├── .github └── workflows │ ├── code-quality.yml │ ├── release.yml │ └── sanity.yml ├── .gitignore ├── ChatDBG.pdf ├── Dockerfile ├── LICENSE ├── README.md ├── check-version-bump.py ├── media ├── gdb.svg ├── lldb.svg └── pdb.svg ├── pyproject.toml ├── rust-support ├── Cargo.toml ├── chatdbg │ ├── Cargo.toml │ └── src │ │ └── lib.rs ├── chatdbg_macros │ ├── Cargo.toml │ └── src │ │ └── lib.rs └── fill-crate-version.py ├── samples ├── cpp │ ├── test-deep-recursion.cpp │ ├── test-definition-likely.cpp │ ├── test-failed-assert.cpp │ ├── test-out-of-bounds-read.cpp │ ├── test-overflow.cpp │ ├── test-pointers-loop.cpp │ ├── test-pointers.cpp │ ├── test-stack-overflow.cpp │ └── test-use-after-free.cpp ├── python │ ├── README.md │ ├── bootstrap.py │ ├── bootstrap2.py │ ├── ds101.py │ ├── marble-sample.csv │ ├── marbles.py │ ├── mean.py │ ├── nb.ipynb │ ├── requirements.txt │ ├── sample.py │ └── testme.py └── rust │ ├── .gitignore │ ├── Cargo.toml │ └── test-failed-assert.rs └── src └── chatdbg ├── __init__.py ├── __main__.py ├── assistant ├── __init__.py ├── assistant.py └── listeners.py ├── chatdbg_gdb.py ├── chatdbg_lldb.py ├── chatdbg_pdb.py ├── custom_pdb ├── prompts.py └── text.py ├── native_util ├── clangd_lsp_integration.py ├── code.py ├── dbg_dialog.py ├── safety.py └── stacks.py ├── pdb_util ├── capture.py ├── locals.py ├── paths.py └── sandbox.py └── util ├── config.py ├── exit_message.py ├── help.py ├── history.py ├── instructions ├── default.txt └── gpt-4o.txt ├── jupyter.py ├── log.py ├── markdown.py ├── module_whitelist.txt ├── plog.py ├── printer.py ├── prompts.py ├── stream.py ├── text.py ├── trim.py └── wrap.py /.github/workflows/code-quality.yml: -------------------------------------------------------------------------------- 1 | name: Code quality 2 | 3 | on: 4 | push: 5 | branches: ["main"] 6 | pull_request: 7 | branches: ["main"] 8 | 9 | jobs: 10 | code-quality: 11 | strategy: 12 | matrix: 13 | python: ["3.9","3.10", "3.11", "3.12", "3.13"] 14 | 15 | runs-on: ubuntu-latest 16 | 17 | steps: 18 | - uses: actions/checkout@v4 19 | 20 | - uses: actions/setup-python@v5 21 | with: 22 | python-version: ${{ matrix.python }} 23 | 24 | - name: Install 25 | run: python3 -m pip install . 26 | 27 | - name: Install dependencies 28 | run: | 29 | python3 -m pip install --upgrade pip 30 | python3 -m pip install black 31 | python3 -m pip install mypy 32 | 33 | - name: Run compileall 34 | run: python3 -m compileall src 35 | 36 | - name: Run black 37 | run: python3 -m black --check src 38 | -------------------------------------------------------------------------------- /.github/workflows/release.yml: -------------------------------------------------------------------------------- 1 | name: Release 2 | 3 | on: 4 | release: 5 | types: [published] 6 | 7 | permissions: 8 | contents: read 9 | 10 | jobs: 11 | release: 12 | runs-on: ubuntu-latest 13 | 14 | steps: 15 | - uses: actions/checkout@v4 16 | 17 | - name: Set up Python 18 | uses: actions/setup-python@v5 19 | with: 20 | python-version: '>=3.11' 21 | 22 | - name: Install dependencies 23 | run: | 24 | python -m pip install --upgrade pip 25 | pip install build packaging requests 26 | 27 | - name: Check version bump 28 | run: python3 check-version-bump.py 29 | 30 | - name: Sync pyproject.toml/Cargo.toml versions 31 | run: python3 rust-support/fill-crate-version.py 32 | 33 | - name: Build package 34 | run: python -m build 35 | 36 | - name: Publish to crates.io 37 | run: | 38 | cargo publish -p chatdbg_macros --allow-dirty 39 | cargo publish -p chatdbg --allow-dirty 40 | env: 41 | CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }} 42 | working-directory: rust-support 43 | 44 | - name: Publish to PyPI 45 | uses: pypa/gh-action-pypi-publish@release/v1 46 | with: 47 | user: __token__ 48 | password: ${{ secrets.PYPI_API_TOKEN }} 49 | -------------------------------------------------------------------------------- /.github/workflows/sanity.yml: -------------------------------------------------------------------------------- 1 | name: Sanity 2 | 3 | on: 4 | push: 5 | branches: ["main"] 6 | pull_request: 7 | branches: ["main"] 8 | 9 | jobs: 10 | sanity: 11 | strategy: 12 | matrix: 13 | platform: ["ubuntu-latest", "macos-latest", "windows-latest"] 14 | python: ["3.9","3.10", "3.11", "3.12", "3.13"] 15 | 16 | runs-on: ${{ matrix.platform }} 17 | 18 | steps: 19 | - uses: actions/checkout@v4 20 | 21 | - uses: actions/setup-python@v5 22 | with: 23 | python-version: ${{ matrix.python }} 24 | 25 | - name: Update pip 26 | run: python3 -m pip install --upgrade pip 27 | 28 | - name: Install 29 | run: python3 -m pip install . 30 | 31 | - name: Check help message 32 | run: python3 -m chatdbg --help 33 | 34 | - name: Check calling executable directly 35 | run: chatdbg --help 36 | -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | develop-eggs/ 12 | dist/ 13 | downloads/ 14 | eggs/ 15 | .eggs/ 16 | lib64/ 17 | parts/ 18 | sdist/ 19 | var/ 20 | wheels/ 21 | share/python-wheels/ 22 | *.egg-info/ 23 | .installed.cfg 24 | *.egg 25 | MANIFEST 26 | 27 | # PyInstaller 28 | # Usually these files are written by a python script from a template 29 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 30 | *.manifest 31 | 32 | # Installer logs 33 | pip-log.txt 34 | pip-delete-this-directory.txt 35 | 36 | # Unit test / coverage reports 37 | htmlcov/ 38 | .tox/ 39 | .nox/ 40 | .coverage 41 | .coverage.* 42 | .cache 43 | nosetests.xml 44 | coverage.xml 45 | *.cover 46 | *.py,cover 47 | .hypothesis/ 48 | .pytest_cache/ 49 | cover/ 50 | 51 | # Translations 52 | *.mo 53 | *.pot 54 | 55 | # Django stuff: 56 | *.log 57 | local_settings.py 58 | db.sqlite3 59 | db.sqlite3-journal 60 | 61 | # Flask stuff: 62 | instance/ 63 | .webassets-cache 64 | 65 | # Scrapy stuff: 66 | .scrapy 67 | 68 | # Sphinx documentation 69 | docs/_build/ 70 | 71 | # PyBuilder 72 | .pybuilder/ 73 | target/ 74 | 75 | # Jupyter Notebook 76 | .ipynb_checkpoints 77 | 78 | # IPython 79 | profile_default/ 80 | ipython_config.py 81 | 82 | # pyenv 83 | # For a library or package, you might want to ignore these files since the code is 84 | # intended to run in multiple environments; otherwise, check them in: 85 | # .python-version 86 | 87 | # pipenv 88 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 89 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 90 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 91 | # install all needed dependencies. 92 | #Pipfile.lock 93 | 94 | # poetry 95 | # Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control. 96 | # This is especially recommended for binary packages to ensure reproducibility, and is more 97 | # commonly ignored for libraries. 98 | # https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control 99 | #poetry.lock 100 | 101 | # pdm 102 | # Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control. 103 | #pdm.lock 104 | # pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it 105 | # in version control. 106 | # https://pdm.fming.dev/#use-with-ide 107 | .pdm.toml 108 | 109 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm 110 | __pypackages__/ 111 | 112 | # Celery stuff 113 | celerybeat-schedule 114 | celerybeat.pid 115 | 116 | # SageMath parsed files 117 | *.sage.py 118 | 119 | # Environments 120 | .env 121 | .venv 122 | env/ 123 | venv/ 124 | ENV/ 125 | env.bak/ 126 | venv.bak/ 127 | 128 | # Spyder project settings 129 | .spyderproject 130 | .spyproject 131 | 132 | # Rope project settings 133 | .ropeproject 134 | 135 | # mkdocs documentation 136 | /site 137 | 138 | # mypy 139 | .mypy_cache/ 140 | .dmypy.json 141 | dmypy.json 142 | 143 | # Pyre type checker 144 | .pyre/ 145 | 146 | # pytype static type analyzer 147 | .pytype/ 148 | 149 | # Cython debug symbols 150 | cython_debug/ 151 | 152 | # PyCharm 153 | # JetBrains specific template is maintained in a separate JetBrains.gitignore that can 154 | # be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore 155 | # and can be added to the global gitignore or merged into this file. For a more nuclear 156 | # option (not recommended) you can uncomment the following to ignore the entire idea folder. 157 | #.idea/ 158 | 159 | a.out 160 | log.yaml 161 | -------------------------------------------------------------------------------- /ChatDBG.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/plasma-umass/ChatDBG/805a3d1900aebaaa26603b334e688d70c3d3a7bf/ChatDBG.pdf -------------------------------------------------------------------------------- /Dockerfile: -------------------------------------------------------------------------------- 1 | # A Dockerfile to make development setup easier. Run with: 2 | # docker run -it --rm -e OPENAI_API_KEY=$OPENAI_API_KEY $(docker build -q .) 3 | 4 | FROM ubuntu 5 | 6 | ARG LLVM_VERSION=20 7 | 8 | RUN apt update \ 9 | && DEBIAN_FRONTEND=noninteractive apt install -y tzdata \ 10 | && apt install -y python3 python3-pip \ 11 | && apt install -y locales \ 12 | && apt install -y autoconf automake bear bison build-essential cmake flex gdb git libgdbm-dev m4 texinfo \ 13 | && apt install -y curl lsb-release wget software-properties-common gnupg \ 14 | && curl -sSf https://apt.llvm.org/llvm.sh | bash -s -- ${LLVM_VERSION} all \ 15 | && apt clean \ 16 | && rm -rf /var/lib/apt/lists/* 17 | 18 | # UTF-8. 19 | RUN locale-gen en_US.UTF-8 20 | RUN update-locale LANG=en_US.UTF-8 21 | ENV LANG=en_US.UTF-8 22 | ENV LANGUAGE=en_US:en 23 | ENV LC_ALL=en_US.UTF-8 24 | 25 | # LLVM/Clang. 26 | RUN update-alternatives --install /usr/bin/clang clang /usr/bin/clang-${LLVM_VERSION} 100 27 | RUN update-alternatives --install /usr/bin/clang++ clang++ /usr/bin/clang++-${LLVM_VERSION} 100 28 | RUN update-alternatives --install /usr/bin/clangd clangd /usr/bin/clangd-${LLVM_VERSION} 100 29 | RUN update-alternatives --install /usr/bin/lldb lldb /usr/bin/lldb-${LLVM_VERSION} 100 30 | 31 | RUN update-alternatives --install /usr/bin/cc cc /usr/bin/clang 100 32 | RUN update-alternatives --install /usr/bin/c++ c++ /usr/bin/clang++ 100 33 | ENV CC=clang 34 | ENV CXX=clang++ 35 | 36 | # ChatDBG. 37 | COPY . /root/ChatDBG 38 | RUN pip install --break-system-packages -e /root/ChatDBG 39 | RUN python3 -c 'import chatdbg; print(f"command script import {chatdbg.__path__[0]}/chatdbg_lldb.py")' >> ~/.lldbinit 40 | RUN echo 'settings set target.disable-aslr false' >> ~/.lldbinit 41 | RUN python3 -c 'import chatdbg; print(f"source {chatdbg.__path__[0]}/chatdbg_gdb.py")' >> ~/.gdbinit 42 | ENV CHATDBG_UNSAFE=1 43 | 44 | RUN pip install --break-system-packages -r /root/ChatDBG/samples/python/requirements.txt 45 | 46 | # Bugbench. 47 | RUN git clone https://github.com/nicovank/bugbench.git /root/ChatDBG/samples/bugbench 48 | RUN cd /root/ChatDBG/samples/bugbench && make all 49 | 50 | ENV TERM=xterm-256color 51 | WORKDIR /root/ChatDBG 52 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "[]" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright [yyyy] [name of copyright owner] 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. 202 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # ChatDBG 2 | 3 | by [Emery Berger](https://emeryberger.com), [Stephen Freund](https://www.cs.williams.edu/~freund/index.html), [Kyla Levin](https://khlevin.github.io/KylaHLevin/index.html), [Nicolas van Kempen](https://nvankempen.com/) (ordered alphabetically) 4 | 5 | [![PyPI Latest Release](https://img.shields.io/pypi/v/chatdbg.svg)](https://pypi.org/project/chatdbg/) 6 | [![Downloads](https://static.pepy.tech/badge/chatdbg)](https://pepy.tech/project/chatdbg) 7 | [![Downloads](https://static.pepy.tech/badge/chatdbg/month)](https://pepy.tech/project/chatdbg) 8 | 9 | [Read the paper!](https://raw.githubusercontent.com/plasma-umass/ChatDBG/main/ChatDBG.pdf) 10 | 11 | ChatDBG is an AI-based debugging assistant for C/C++/Python/Rust code that integrates large language models into a standard debugger (`pdb`, `lldb`, `gdb`) to help debug your code. With ChatDBG, you can engage in a dialog with your debugger, asking open-ended questions about your program, like `why is x null?`. ChatDBG will _take the wheel_ and steer the debugger to answer your queries. ChatDBG can provide error diagnoses and suggest fixes. 12 | 13 | As far as we are aware, ChatDBG is the _first_ debugger to automatically perform root cause analysis and to provide suggested fixes. 14 | 15 | **Watch ChatDBG in action!** 16 | | LLDB on [test-overflow.cpp](https://github.com/plasma-umass/ChatDBG/blob/main/samples/cpp/test-overflow.cpp) | GDB on [test-overflow.cpp](https://github.com/plasma-umass/ChatDBG/blob/main/samples/cpp/test-overflow.cpp) | Pdb on [bootstrap.py](https://github.com/plasma-umass/ChatDBG/blob/main/samples/python/bootstrap.py) | 17 | |:-------------------------:|:-------------------------:|:-------------------------:| 18 | | | || 19 | 20 | For technical details and a complete evaluation, see our FSE'25 paper, [_ChatDBG: An AI-Powered Debugging Assistant_](https://dl.acm.org/doi/10.1145/3729355) ([PDF](https://raw.githubusercontent.com/plasma-umass/ChatDBG/main/ChatDBG.pdf)). 21 | 22 | > [!NOTE] 23 | > 24 | > ChatDBG for `pdb`, `lldb`, and `gdb` are feature-complete; we are currently backporting features for these debuggers into the other debuggers. 25 | 26 | ## Installation 27 | 28 | > [!IMPORTANT] 29 | > 30 | > ChatDBG currently needs to be connected to an [OpenAI account](https://openai.com/api/). _Your account will need to have a positive balance for this to work_ ([check your balance](https://platform.openai.com/account/usage)). If you have never purchased credits, you will need to purchase at least \$1 in credits (if your API account was created before August 13, 2023) or \$0.50 (if you have a newer API account) in order to have access to GPT-4, which ChatDBG uses. [Get a key here.](https://platform.openai.com/account/api-keys) 31 | > 32 | > Once you have an API key, set it as an environment variable called `OPENAI_API_KEY`. 33 | > 34 | > ```bash 35 | > export OPENAI_API_KEY= 36 | > ``` 37 | 38 | Install ChatDBG using `pip` (you need to do this whether you are debugging Python, C, or C++ code): 39 | 40 | ```bash 41 | python3 -m pip install chatdbg 42 | ``` 43 | 44 | If you are using ChatDBG to debug Python programs, you are done. If you want to use ChatDBG to debug native code with `gdb` or `lldb`, follow the installation instructions below. 45 | 46 | ### Installing as an `lldb` extension 47 | 48 |
49 | 50 | lldb installation instructions 51 | 52 | 53 | Install ChatDBG into the `lldb` debugger by running the following command: 54 | 55 | #### Linux 56 | 57 | ```bash 58 | python3 -m pip install ChatDBG 59 | python3 -c 'import chatdbg; print(f"command script import {chatdbg.__path__[0]}/chatdbg_lldb.py")' >> ~/.lldbinit 60 | ``` 61 | 62 | If you encounter an error, you may be using an older version of LLVM. Update to the latest version as follows: 63 | 64 | ``` 65 | sudo apt install -y lsb-release wget software-properties-common gnupg 66 | curl -sSf https://apt.llvm.org/llvm.sh | sudo bash -s -- 18 all 67 | # LLDB now available as `lldb-18`. 68 | ``` 69 | 70 | #### Mac 71 | 72 | ```bash 73 | xcrun python3 -m pip install ChatDBG 74 | xcrun python3 -c 'import chatdbg; print(f"command script import {chatdbg.__path__[0]}/chatdbg_lldb.py")' >> ~/.lldbinit 75 | ``` 76 | 77 | This will install ChatDBG as an LLVM extension. 78 | 79 |
80 | 81 | ### Installing as a `gdb` extension 82 | 83 |
84 | 85 | gdb installation instructions 86 | 87 | 88 | Install ChatDBG into the `gdb` debugger by running the following command: 89 | 90 | ```bash 91 | python3 -m pip install ChatDBG 92 | python3 -c 'import chatdbg; print(f"source {chatdbg.__path__[0]}/chatdbg_gdb.py")' >> ~/.gdbinit 93 | ``` 94 | 95 | This will install ChatDBG as a GDB extension. 96 | 97 |
98 | 99 | ## Usage 100 | 101 | ### Debugging Python 102 | 103 | To use ChatDBG to debug Python programs, simply run your Python script as follows: 104 | 105 | ```bash 106 | chatdbg -c continue yourscript.py 107 | ``` 108 | 109 | ChatDBG is an extension of the standard Python debugger `pdb`. Like 110 | `pdb`, when your script encounters an uncaught exception, ChatDBG will 111 | enter post mortem debugging mode. 112 | 113 | Unlike other debuggers, you can then use the `why` command to ask 114 | ChatDBG why your program failed and get a suggested fix. After the LLM responds, 115 | you may issue additional debugging commands or continue the conversation by entering 116 | any other text. 117 | 118 | #### IPython and Jupyter Support 119 | 120 | To use ChatDBG as the default debugger for IPython or inside Jupyter Notebooks, 121 | create a IPython profile and then add the necessary exensions on startup. (Modify 122 | these lines as necessary if you already have a customized profile file.) 123 | 124 | ```bash 125 | ipython profile create 126 | echo "c.InteractiveShellApp.extensions = ['chatdbg.chatdbg_pdb', 'ipyflow']" >> ~/.ipython/profile_default/ipython_config.py 127 | ``` 128 | 129 | On the command line, you can then run: 130 | 131 | ```bash 132 | ipython --pdb yourscript.py 133 | ``` 134 | 135 | Inside Jupyter, run your notebook with the [ipyflow kernel](https://github.com/ipyflow/ipyflow) and include this line magic at the top of the file. 136 | 137 | ``` 138 | %pdb 139 | ``` 140 | 141 | ### Debugging native code (C, C++, or Rust with lldb / gdb) 142 | 143 | To use ChatDBG with `lldb` or `gdb`, just run native code (compiled with `-g` for debugging symbols) with your choice of debugger; when it crashes, ask `why`. This also works for post mortem debugging (when you load a core with the `-c` option). 144 | 145 | The native debuggers work slightly differently than Pdb. After the debugger responds to your question, you will enter into ChatDBG's command loop, as indicated by the `(ChatDBG chatting)` prompt. You may continue issuing debugging commands and you may send additional messages to the LLM by starting those messages with "chat". When you are done, type `quit` to return to the debugger's main command loop. 146 | 147 |
148 | 149 | Debugging Rust programs 150 | 151 | 152 | To use ChatDBG with Rust, you need to do two steps: modify your 153 | `Cargo.toml` file and add one line to your source program. 154 | 155 | 1. Add this to your `Cargo.toml` file: 156 | 157 | ```toml 158 | [dependencies] 159 | chatdbg = "0.6.2" 160 | 161 | [profile.dev] 162 | panic = "abort" 163 | 164 | [profile.release] 165 | panic = "abort" 166 | ``` 167 | 168 | 2. In your program, apply the `#[chatdbg::main]` attribute to your `main` 169 | function: 170 | 171 | ```rust 172 | #[chatdbg::main] 173 | fn main() { 174 | ``` 175 | 176 | Now you can debug your Rust code with `gdb` or `lldb`. 177 | 178 |
179 | 180 | ### Examples 181 | 182 |
183 | 184 | ChatDBG example in lldb 185 | 186 | 187 | ```gdb 188 | (ChatDBG lldb) run 189 | Process 85494 launched: '/Users/emery/git/ChatDBG/test/a.out' (arm64) 190 | TEST 1 191 | TEST -422761288 192 | TEST 0 193 | TEST 0 194 | TEST 0 195 | TEST 0 196 | TEST 0 197 | TEST 0 198 | Process 85494 stopped 199 | * thread #1, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x100056200) 200 | frame #0: 0x0000000100002f64 a.out`foo(n=8, b=1) at test.cpp:7:22 201 | 4 int x[] = { 1, 2, 3, 4, 5 }; 202 | 5 203 | 6 void foo(int n, float b) { 204 | -> 7 cout << "TEST " << x[n * 10000] << endl; 205 | 8 } 206 | 9 207 | 10 int main() 208 | Target 0: (a.out) stopped. 209 | ``` 210 | 211 | Ask `why` to have ChatDBG provide a helpful explanation why this program failed, and suggest a fix: 212 | 213 | ````gdb 214 | (ChatDBG lldb) why 215 | The root cause of this error is accessing an index of the array `x` 216 | that is out of bounds. In `foo()`, the index is calculated as `n * 217 | 10000`, which can be much larger than the size of the array `x` (which 218 | is only 5 elements). In the given trace, the program is trying to 219 | access the memory address `0x100056200`, which is outside of the range 220 | of allocated memory for the array `x`. 221 | 222 | To fix this error, we need to ensure that the index is within the 223 | bounds of the array. One way to do this is to check the value of `n` 224 | before calculating the index, and ensure that it is less than the size 225 | of the array divided by the size of the element. For example, we can 226 | modify `foo()` as follows: 227 | 228 | ``` 229 | void foo(int n, float b) { 230 | if (n < 0 || n >= sizeof(x)/sizeof(x[0])) { 231 | cout << "ERROR: Invalid index" << endl; 232 | return; 233 | } 234 | cout << "TEST " << x[n] << endl; 235 | } 236 | ``` 237 | 238 | This code checks if `n` is within the valid range, and prints an error 239 | message if it is not. If `n` is within the range, the function prints 240 | the value of the element at index `n` of `x`. With this modification, 241 | the program will avoid accessing memory outside the bounds of the 242 | array, and will print the expected output for valid indices. 243 | ```` 244 | 245 |
246 | 247 |
248 | 249 | ChatDBG example in Python (pdb) 250 | 251 | 252 | ```python 253 | Traceback (most recent call last): 254 | File "yourscript.py", line 9, in 255 | print(tryme(100)) 256 | File "yourscript.py", line 4, in tryme 257 | if x / i > 2: 258 | ZeroDivisionError: division by zero 259 | Uncaught exception. Entering post mortem debugging 260 | Running 'cont' or 'step' will restart the program 261 | > yourscript.py(4)tryme() 262 | -> if x / i > 2: 263 | ``` 264 | 265 | Ask `why` to have ChatDBG provide a helpful explanation why this program failed, and suggest a fix: 266 | 267 | ```python 268 | (ChatDBG Pdb) why 269 | The root cause of the error is that the code is attempting to 270 | divide by zero in the line "if x / i > 2". As i ranges from 0 to 99, 271 | it will eventually reach the value of 0, causing a ZeroDivisionError. 272 | 273 | A possible fix for this would be to add a check for i being equal to 274 | zero before performing the division. This could be done by adding an 275 | additional conditional statement, such as "if i == 0: continue", to 276 | skip over the iteration when i is zero. The updated code would look 277 | like this: 278 | 279 | def tryme(x): 280 | count = 0 281 | for i in range(100): 282 | if i == 0: 283 | continue 284 | if x / i > 2: 285 | count += 1 286 | return count 287 | 288 | if __name__ == '__main__': 289 | print(tryme(100)) 290 | ``` 291 | 292 |
293 | -------------------------------------------------------------------------------- /check-version-bump.py: -------------------------------------------------------------------------------- 1 | import tomllib 2 | import sys 3 | 4 | from packaging.version import Version 5 | import requests 6 | 7 | PYPI_URL = "https://pypi.org/pypi/chatdbg/json" 8 | 9 | latest = requests.get(PYPI_URL).json()["info"]["version"] 10 | with open("pyproject.toml", "rb") as f: 11 | current = tomllib.load(f)["project"]["version"] 12 | 13 | if Version(current) <= Version(latest): 14 | print(f"Latest version is {latest} on PyPI. Please bump versions.") 15 | sys.exit(1) 16 | -------------------------------------------------------------------------------- /media/gdb.svg: -------------------------------------------------------------------------------- 1 | 2 | 3 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | For help, type "help". 58 | 59 | 60 | 61 | Type "apropos word" to search for commands related to "word"... 62 | 63 | 64 | 65 | Reading symbols from 66 | 67 | a.out 68 | 69 | ... 70 | 71 | 72 | 73 | (ChatDBG gdb) run 74 | 75 | 76 | 77 | Starting program: 78 | 79 | /home/steve/ChatDBG/samples/cpp/a.out 80 | 81 | 82 | 83 | [Thread debugging using libthread_db enabled] 84 | 85 | 86 | 87 | Using host libthread_db library " 88 | 89 | /lib/x86_64-linux-gnu/libthread_db.so.1 90 | 91 | ". 92 | 93 | 94 | 95 | TEST 1 96 | 97 | 98 | 99 | TEST 0 100 | 101 | 102 | 103 | TEST 0 104 | 105 | 106 | 107 | TEST 0 108 | 109 | 110 | 111 | Program received signal SIGSEGV, Segmentation fault. 112 | 113 | 114 | 115 | 0x0000555555555213 116 | 117 | in 118 | 119 | foo 120 | 121 | ( 122 | 123 | n 124 | 125 | =4, 126 | 127 | b 128 | 129 | =1) at 130 | 131 | test-overflow.cpp 132 | 133 | :7 134 | 135 | 136 | 137 | 7 138 | 139 | cout 140 | 141 | << 142 | 143 | "TEST " 144 | 145 | << 146 | 147 | x 148 | 149 | [ 150 | 151 | n 152 | 153 | * 154 | 155 | 10000 156 | 157 | ] 158 | 159 | << 160 | 161 | endl 162 | 163 | ; 164 | 165 | 166 | 167 | (ChatDBG gdb) why? 168 | 169 | 170 | 171 | (ChatDBG gdb) code test-overflow.cpp:7 172 | 173 | 174 | 175 | 1 #include <iostream> 176 | 177 | 178 | 179 | 2 using namespace std; 180 | 181 | 182 | 183 | 3 184 | 185 | 186 | 187 | 4 int x[] = { 1, 2, 3, 4, 5 }; 188 | 189 | 190 | 191 | 5 192 | 193 | 194 | 195 | 6 void foo(int n, float b) { 196 | 197 | 198 | 199 | 7 200 | 201 | cout << "TEST " << x[n * 10000] << endl; 202 | 203 | 204 | 205 | 8 } 206 | 207 | 208 | 209 | 9 210 | 211 | 212 | 213 | 10 int main() 214 | 215 | 216 | 217 | The stack trace shows that the program crashed due to a 218 | 219 | SIGSEGV 220 | 221 | (segmentation fault), which 222 | 223 | 224 | 225 | 226 | 227 | 228 | 229 | 230 | 231 | 232 | 233 | 234 | 235 | 236 | 237 | 238 | 239 | 240 | 241 | 242 | 243 | 244 | -------------------------------------------------------------------------------- /media/lldb.svg: -------------------------------------------------------------------------------- 1 | 2 | 3 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | (ChatDBG lldb) 58 | 59 | run 60 | 61 | 62 | 63 | Process 23558 launched: '/home/steve/ChatDBG/samples/cpp/a.out' (x86_64) 64 | 65 | 66 | 67 | TEST 1 68 | 69 | 70 | 71 | TEST 0 72 | 73 | 74 | 75 | TEST 0 76 | 77 | 78 | 79 | TEST 0 80 | 81 | 82 | 83 | Process 23558 stopped 84 | 85 | 86 | 87 | * thread #1, name = 'a.out', stop reason = 88 | 89 | signal SIGSEGV: address not mapped to object (fault address: 0x55555557f110) 90 | 91 | 92 | 93 | frame #0: 94 | 95 | 0x0000555555555213 96 | 97 | a.out`foo(n=4, b=1) at 98 | 99 | test-overflow.cpp 100 | 101 | : 102 | 103 | 7 104 | 105 | : 106 | 107 | 33 108 | 109 | 110 | 111 | 4 112 | 113 | int 114 | 115 | x[] = { 116 | 117 | 1 118 | 119 | , 120 | 121 | 2 122 | 123 | , 124 | 125 | 3 126 | 127 | , 128 | 129 | 4 130 | 131 | , 132 | 133 | 5 134 | 135 | }; 136 | 137 | 138 | 139 | 5 140 | 141 | 142 | 143 | 6 144 | 145 | void 146 | 147 | foo( 148 | 149 | int 150 | 151 | n, 152 | 153 | float 154 | 155 | b) { 156 | 157 | 158 | 159 | -> 160 | 161 | 7 162 | 163 | cout << "TEST " << x[n * 164 | 165 | 10000 166 | 167 | ] 168 | 169 | << endl; 170 | 171 | 172 | 173 | 8 174 | 175 | } 176 | 177 | 178 | 179 | 9 180 | 181 | 182 | 183 | 10 184 | 185 | int 186 | 187 | main() 188 | 189 | 190 | 191 | (ChatDBG lldb) 192 | 193 | why 194 | 195 | 196 | 197 | The SIGSEGV (segmentation fault) signal indicates that the program attempted to access memory that it's not 198 | 199 | 200 | 201 | allowed to. Examining the provided stack trace and the corresponding source code reveals the potential cause of 202 | 203 | 204 | 205 | the segmentation fault. 206 | 207 | 208 | 209 | In frame #0, within the 210 | 211 | foo 212 | 213 | function, there is an attempt to access an element of the array 214 | 215 | x 216 | 217 | using an index 218 | 219 | n * 220 | 221 | 222 | 223 | 10000 224 | 225 | : 226 | 227 | 228 | 229 | 7 230 | 231 | cout 232 | 233 | << 234 | 235 | "TEST " 236 | 237 | << 238 | 239 | x[n 240 | 241 | * 242 | 243 | 10000 244 | 245 | ] 246 | 247 | << 248 | 249 | endl; 250 | 251 | 252 | 253 | The array 254 | 255 | x 256 | 257 | is defined with only 5 elements: 258 | 259 | 260 | 261 | 4 262 | 263 | int 264 | 265 | x[] 266 | 267 | = 268 | 269 | { 270 | 271 | 1 272 | 273 | , 274 | 275 | 276 | 277 | 278 | 279 | 280 | 281 | 282 | 283 | 284 | 285 | 286 | 287 | 288 | 289 | 290 | 291 | 292 | 293 | 294 | 295 | 296 | -------------------------------------------------------------------------------- /media/pdb.svg: -------------------------------------------------------------------------------- 1 | 2 | 3 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | exec(cmd, globals, locals) 48 | 49 | 50 | 51 | File "<string>", line 1, in <module> 52 | 53 | 54 | 55 | File "/home/steve/ChatDBG/samples/python/bootstrap.py", line 19, in <module> 56 | 57 | 58 | 59 | stats = resampled_stats(observed_marbles, 5) 60 | 61 | 62 | 63 | File "/home/steve/ChatDBG/samples/python/bootstrap.py", line 15, in resampled_stats 64 | 65 | 66 | 67 | assert len(stats) == num_trials 68 | 69 | 70 | 71 | AssertionError 72 | 73 | 74 | 75 | Uncaught exception. Entering post mortem debugging 76 | 77 | 78 | 79 | Running 'cont' or 'step' will restart the program 80 | 81 | 82 | 83 | > 84 | 85 | /home/steve/ChatDBG/samples/python/bootstrap.py 86 | 87 | (15) 88 | 89 | resampled_stats 90 | 91 | () 92 | 93 | 94 | 95 | 10 96 | 97 | 98 | 99 | 11 def 100 | 101 | resampled_stats 102 | 103 | ( 104 | 105 | observed_marbles 106 | 107 | , 108 | 109 | num_trials 110 | 111 | ): 112 | 113 | 114 | 115 | 12 116 | 117 | stats = bootstrap_statistic(observed_marbles, 118 | 119 | 120 | 121 | 13 122 | 123 | proportion_blue 124 | 125 | , 126 | 127 | 128 | 129 | 14 130 | 131 | num_trials) 132 | 133 | 134 | 135 | ---> 15 136 | 137 | assert 138 | 139 | len 140 | 141 | ( 142 | 143 | stats 144 | 145 | ) 146 | 147 | == 148 | 149 | num_trials 150 | 151 | 152 | 153 | 16 154 | 155 | return 156 | 157 | stats 158 | 159 | 160 | 161 | 17 162 | 163 | 164 | 165 | 18 166 | 167 | observed_marbles 168 | 169 | = 170 | 171 | make_marble_sample 172 | 173 | () 174 | 175 | 176 | 177 | 19 178 | 179 | stats 180 | 181 | = 182 | 183 | resampled_stats 184 | 185 | ( 186 | 187 | observed_marbles 188 | 189 | , 190 | 191 | 5 192 | 193 | ) 194 | 195 | 196 | 197 | (ChatDBG) 198 | 199 | why 200 | 201 | 202 | 203 | (ChatDBG) p proportion_blue 204 | 205 | 206 | 207 | <function proportion_blue at 0x7f5655935480> 208 | 209 | 210 | 211 | (ChatDBG) p bootstrap_statistic 212 | 213 | 214 | 215 | <function bootstrap_statistic at 0x7f5655e26440> 216 | 217 | 218 | 219 | 220 | 221 | 222 | 223 | 224 | 225 | 226 | 227 | 228 | 229 | 230 | 231 | 232 | 233 | 234 | 235 | 236 | 237 | 238 | -------------------------------------------------------------------------------- /pyproject.toml: -------------------------------------------------------------------------------- 1 | [build-system] 2 | requires = ["hatchling"] 3 | build-backend = "hatchling.build" 4 | 5 | [project] 6 | name = "ChatDBG" 7 | version = "1.0.0" 8 | authors = [ 9 | { name="Emery Berger", email="emery.berger@gmail.com" }, 10 | { name="Stephen Freund", email="sfreund@williams.edu" }, 11 | { name="Kyla Levin", email="khlevin@umass.edu" }, 12 | { name="Nicolas van Kempen", email="nvankemp@gmail.com" }, 13 | ] 14 | dependencies = [ 15 | "llm-utils>=0.2.8", 16 | "openai>=1.29.0", 17 | "rich>=13.7.0", 18 | "ansicolors>=1.1.8", 19 | "traitlets>=5.14.1", 20 | "ipdb>=0.13.13", 21 | "ipython==8.18.1", 22 | "litellm==1.55.9", 23 | "PyYAML>=6.0.1", 24 | "ipyflow>=0.0.130", 25 | "numpy>=1.26.3", 26 | ] 27 | description = "AI-assisted debugging. Uses AI to answer 'why'." 28 | readme = "README.md" 29 | requires-python = ">=3.9" 30 | classifiers = [ 31 | "Programming Language :: Python :: 3", 32 | "License :: OSI Approved :: Apache Software License", 33 | "Operating System :: OS Independent", 34 | ] 35 | 36 | [project.scripts] 37 | chatdbg = "chatdbg.__main__:main" 38 | print_chatdbg_log = "chatdbg.util.plog:main" 39 | 40 | [project.urls] 41 | "Homepage" = "https://github.com/plasma-umass/ChatDBG" 42 | "Bug Tracker" = "https://github.com/plasma-umass/ChatDBG/issues" 43 | -------------------------------------------------------------------------------- /rust-support/Cargo.toml: -------------------------------------------------------------------------------- 1 | [workspace] 2 | resolver = "2" 3 | members = [ 4 | "chatdbg", 5 | "chatdbg_macros", 6 | ] 7 | -------------------------------------------------------------------------------- /rust-support/chatdbg/Cargo.toml: -------------------------------------------------------------------------------- 1 | [package] 2 | name = "chatdbg" 3 | version = "##VERSION##" # Will be auto-synced with pyproject.toml on release. 4 | edition = "2021" 5 | description = "Rust-specific support for ChatDBG." 6 | authors = ["Emery Berger "] 7 | license = "Apache-2.0" 8 | readme = "../../README.md" 9 | homepage = "https://github.com/plasma-umass/ChatDBG/" 10 | repository = "https://github.com/plasma-umass/ChatDBG/" 11 | 12 | [dependencies] 13 | chatdbg_macros = "##VERSION##" # Sync with pyproject.toml on release. 14 | lazy_static = "1.4.0" 15 | -------------------------------------------------------------------------------- /rust-support/chatdbg/src/lib.rs: -------------------------------------------------------------------------------- 1 | pub use chatdbg_macros::main; 2 | 3 | use std::fs::{File, OpenOptions}; 4 | use std::io::Write; 5 | use std::panic; 6 | use std::sync::{ 7 | atomic::{AtomicBool, Ordering}, 8 | Mutex, 9 | }; 10 | use std::thread; 11 | 12 | // Global Mutex to synchronize file writes across threads. 13 | lazy_static::lazy_static! { 14 | static ref FILE_MUTEX: Mutex<()> = Mutex::new(()); 15 | static ref FILE_CREATED: AtomicBool = AtomicBool::new(false); 16 | } 17 | 18 | pub fn chatdbg() { 19 | // Set a custom panic hook. 20 | panic::set_hook(Box::new(|info| { 21 | let _guard = FILE_MUTEX.lock().unwrap(); // Lock Mutex to synchronize access. 22 | 23 | // Format the panic message similarly to the default panic handler. 24 | let payload = if let Some(s) = info.payload().downcast_ref::<&str>() { 25 | *s 26 | } else { 27 | "Box" 28 | }; 29 | 30 | let location = if let Some(location) = info.location() { 31 | format!(" at '{}' line {}", location.file(), location.line()) 32 | } else { 33 | String::from("") 34 | }; 35 | 36 | let message = format!( 37 | "thread '{}' panicked with '{}'{}", 38 | thread::current().name().unwrap_or(""), 39 | payload, 40 | location 41 | ); 42 | 43 | // Print to stderr. 44 | eprintln!("{}", message); 45 | 46 | // Specify the filename without including the process id. 47 | let filename = "panic_log.txt"; 48 | 49 | // Open the file with appropriate options. 50 | let mut file = if FILE_CREATED.swap(true, Ordering::SeqCst) { 51 | // If the file is already created by another thread, open it in append mode. 52 | OpenOptions::new() 53 | .create(true) 54 | .append(true) 55 | .open(filename) 56 | .expect("Unable to open file") 57 | } else { 58 | // If this is the first thread to create the file, overwrite any existing file. 59 | File::create(filename).expect("Unable to create file") 60 | }; 61 | 62 | // Write to the file. 63 | writeln!(file, "{}", message).expect("Unable to write to file"); 64 | })); 65 | } 66 | -------------------------------------------------------------------------------- /rust-support/chatdbg_macros/Cargo.toml: -------------------------------------------------------------------------------- 1 | [package] 2 | name = "chatdbg_macros" 3 | version = "##VERSION##" # Will be auto-synced with pyproject.toml on release. 4 | edition = "2021" 5 | description = "Rust-specific (macro) support for ChatDBG." 6 | authors = ["Noah Lev Bartell-Mangel "] 7 | license = "Apache-2.0" 8 | readme = "../../README.md" 9 | homepage = "https://github.com/plasma-umass/ChatDBG/" 10 | repository = "https://github.com/plasma-umass/ChatDBG/" 11 | 12 | [dependencies] 13 | quote = "1.0.33" 14 | syn = { version = "2.0.37", features = ["full"] } 15 | 16 | [lib] 17 | proc-macro = true 18 | -------------------------------------------------------------------------------- /rust-support/chatdbg_macros/src/lib.rs: -------------------------------------------------------------------------------- 1 | use proc_macro::TokenStream; 2 | use quote::quote; 3 | use syn::{parse, ItemFn}; 4 | 5 | #[proc_macro_attribute] 6 | pub fn main(_attr: TokenStream, input: TokenStream) -> TokenStream { 7 | let item = match parse(input) { 8 | Ok(i) => i, 9 | Err(_) => return quote! {}.into(), 10 | }; 11 | let ItemFn { 12 | attrs, 13 | vis, 14 | sig, 15 | block, 16 | } = item; 17 | let stmts = &block.stmts; 18 | quote! { 19 | #(#attrs)* #vis #sig { 20 | ::chatdbg::chatdbg(); 21 | #(#stmts)* 22 | } 23 | } 24 | .into() 25 | } 26 | -------------------------------------------------------------------------------- /rust-support/fill-crate-version.py: -------------------------------------------------------------------------------- 1 | import os 2 | import tomllib 3 | import re 4 | 5 | DIRNAME = os.path.dirname(__file__) 6 | PYPROJECT_FILE = os.path.join(DIRNAME, "../pyproject.toml") 7 | CARGO_TOML_FILES = [ 8 | os.path.join(DIRNAME, "chatdbg/Cargo.toml"), 9 | os.path.join(DIRNAME, "chatdbg_macros/Cargo.toml"), 10 | ] 11 | 12 | 13 | if __name__ == "__main__": 14 | with open(PYPROJECT_FILE, "rb") as f: 15 | version = tomllib.load(f)["project"]["version"] 16 | for file in CARGO_TOML_FILES: 17 | with open(file, "r") as f: 18 | content = f.read() 19 | content = re.sub(r"##VERSION##", version, content) 20 | with open(file, "w") as f: 21 | f.write(content) 22 | -------------------------------------------------------------------------------- /samples/cpp/test-deep-recursion.cpp: -------------------------------------------------------------------------------- 1 | struct David 2 | { 3 | int *data; 4 | }; 5 | struct Charlie 6 | { 7 | David *d1; 8 | }; 9 | struct Bob 10 | { 11 | 12 | Charlie *c1; 13 | Charlie *c2; 14 | David *d2; 15 | }; 16 | struct Adam 17 | { 18 | Bob *b1; 19 | Bob *b2; 20 | }; 21 | 22 | int main() 23 | { 24 | int arrayofints[] = {21, 24, 85, 56, 37, 10, 34, 48, 92, 13}; 25 | int *p = &arrayofints[0]; 26 | David david1 = {&arrayofints[4]}; 27 | David david2 = {&arrayofints[2]}; 28 | David david3 = {&arrayofints[8]}; 29 | Charlie charlie1 = {&david1}; 30 | Charlie charlie2 = {&david2}; 31 | Bob bob1 = {&charlie1, &charlie2, &david3}; 32 | David david4 = {&arrayofints[7]}; 33 | David david5 = {nullptr}; 34 | David david6 = {&arrayofints[0]}; 35 | Charlie charlie3 = {&david4}; 36 | Charlie charlie4 = {&david5}; 37 | Bob bob2 = {&charlie3, &charlie4, &david6}; 38 | Adam adam1 = {&bob1, &bob2}; 39 | 40 | int n = *adam1.b2->c2->d1->data; 41 | return 0; 42 | } 43 | -------------------------------------------------------------------------------- /samples/cpp/test-definition-likely.cpp: -------------------------------------------------------------------------------- 1 | static int* p = nullptr; 2 | 3 | struct Bob 4 | { 5 | int ****data; 6 | }; 7 | struct Adam 8 | { 9 | Bob *b1; 10 | Bob *b2; 11 | }; 12 | 13 | int main() 14 | { 15 | int **p2 = &p; 16 | int ***p3 = &p2; 17 | int ****p4 = &p3; 18 | 19 | Bob bob1 = {p4}; 20 | Bob bob2 = {p4}; 21 | Adam adam1 = {&bob1, &bob2}; 22 | 23 | int n = ****adam1.b1->data; 24 | return 0; 25 | } 26 | -------------------------------------------------------------------------------- /samples/cpp/test-failed-assert.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | 4 | float fact(float n) { 5 | auto x = 1.0; 6 | for (auto i = 0.0; i < n; i++) { 7 | x *= i; 8 | } 9 | assert(x != 0.0); 10 | return x; 11 | } 12 | 13 | 14 | int main() 15 | { 16 | std::cout << fact(100) << std::endl; 17 | return 0; 18 | } 19 | -------------------------------------------------------------------------------- /samples/cpp/test-out-of-bounds-read.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | 4 | void f(int value) { 5 | // The value in this function is expected to be in [1, 5]. 6 | std::cout << value << std::endl; 7 | } 8 | 9 | int main() { 10 | int a[5] = {1, 2, 3, 4, 5}; 11 | int b[5] = {6, 7, 8, 9, 10}; 12 | 13 | f(*(a + 5)); 14 | } 15 | -------------------------------------------------------------------------------- /samples/cpp/test-overflow.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | using namespace std; 3 | 4 | int x[] = { 1, 2, 3, 4, 5 }; 5 | 6 | void foo(int n, float b) { 7 | cout << "TEST " << x[n * 10000] << endl; 8 | } 9 | 10 | int main() 11 | { 12 | for (auto i = 0; i < 10; i++) { 13 | foo(i, 1.0); 14 | } 15 | return 0; 16 | } 17 | -------------------------------------------------------------------------------- /samples/cpp/test-pointers-loop.cpp: -------------------------------------------------------------------------------- 1 | class Node 2 | { 3 | public: 4 | int data; 5 | Node *next; 6 | 7 | Node(int value) : data(value), next(nullptr) {} 8 | }; 9 | 10 | int main() 11 | { 12 | Node *node1 = new Node(10); 13 | Node *node2 = new Node(20); 14 | Node *node3 = new Node(30); 15 | 16 | node1->next = node2; 17 | node2->next = node1; 18 | 19 | Node n = *node3->next; 20 | 21 | delete node1; 22 | delete node2; 23 | delete node3; 24 | 25 | return 0; 26 | } -------------------------------------------------------------------------------- /samples/cpp/test-pointers.cpp: -------------------------------------------------------------------------------- 1 | struct Bob 2 | { 3 | int ****data; 4 | }; 5 | struct Adam 6 | { 7 | Bob *b1; 8 | Bob *b2; 9 | }; 10 | 11 | int main() 12 | { 13 | int num = 1; 14 | int *p1 = # 15 | int **p2 = &p1; 16 | int ***p3 = &p2; 17 | int ****p4 = &p3; 18 | 19 | Bob bob1 = {p4}; 20 | Bob bob2 = {nullptr}; 21 | Adam adam1 = {&bob1, &bob2}; 22 | 23 | int n = ****adam1.b2->data; 24 | return 0; 25 | } 26 | -------------------------------------------------------------------------------- /samples/cpp/test-stack-overflow.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | int fib(int n) { 4 | return fib(n-1) + fib(n-2); 5 | } 6 | 7 | int main() 8 | { 9 | auto const n = 100; 10 | std::cout << "fib(" << n << ") = " << fib(n); 11 | return 0; 12 | } 13 | -------------------------------------------------------------------------------- /samples/cpp/test-use-after-free.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | void doSomething(int * ptr) { 4 | *ptr = 0; 5 | } 6 | 7 | int main() 8 | { 9 | int * n = new int(100); 10 | n--; 11 | delete n; 12 | char * ch = new char[16]; 13 | delete [] ch; 14 | doSomething(n); 15 | std::cout << "n = " << *n << std::endl; 16 | return 0; 17 | } 18 | -------------------------------------------------------------------------------- /samples/python/README.md: -------------------------------------------------------------------------------- 1 | # Python Tests 2 | 3 | * Install the packages in `requirements.txt`. Some of the tests use them. 4 | 5 | * Run, for example: 6 | 7 | ``` 8 | chatdbg -c continue marbles.py 9 | ``` 10 | 11 | * *NOTE*: one python package used by the programs has been replaced 12 | with a small python stub to hide the author's identity. That may 13 | change how ChatDBG works, since it includes that stub as part of 14 | the user-defined code. 15 | -------------------------------------------------------------------------------- /samples/python/bootstrap.py: -------------------------------------------------------------------------------- 1 | from datascience import * 2 | from ds101 import * 3 | 4 | def make_marble_sample(): 5 | table = Table().read_table('marble-sample.csv') 6 | return table.column('color') 7 | 8 | def proportion_blue(sample): 9 | return sample 10 | 11 | def resampled_stats(observed_marbles, num_trials): 12 | stats = bootstrap_statistic(observed_marbles, 13 | proportion_blue, 14 | num_trials) 15 | assert len(stats) == num_trials 16 | return stats 17 | 18 | observed_marbles = make_marble_sample() 19 | stats = resampled_stats(observed_marbles, 5) 20 | 21 | assert np.isclose(np.mean(stats), 0.7) -------------------------------------------------------------------------------- /samples/python/bootstrap2.py: -------------------------------------------------------------------------------- 1 | from datascience import * 2 | from ds101 import * 3 | 4 | def make_marble_sample(): 5 | table = Table().read_table('marble-sample.csv') 6 | return table.column('color') 7 | 8 | def proportion_blue(sample): 9 | return np.count_nonzero(sample == 'B') / len(sample) 10 | 11 | def resampled_stats(observed_marbles, num_trials): 12 | stats = bootstrap_statistic(observed_marbles, 13 | proportion_blue, 14 | num_trials) 15 | assert len(stats) == num_trials 16 | return stats 17 | 18 | observed_marbles = make_marble_sample() 19 | stats = resampled_stats(observed_marbles, 5) 20 | 21 | assert np.isclose(np.mean(stats), 0.7) -------------------------------------------------------------------------------- /samples/python/ds101.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | from datascience import * 3 | 4 | # fake library to hide identities... 5 | 6 | def bootstrap_statistic(observed_sample, compute_statistic, num_trials): 7 | """ 8 | Creates num_trials resamples of the initial sample. 9 | Returns an array of the provided statistic for those samples. 10 | 11 | * observed_sample: the initial sample, as an array. 12 | 13 | * compute_statistic: a function that takes a sample as 14 | an array and returns the statistic for that 15 | sample. 16 | 17 | * num_trials: the number of bootstrap samples to create. 18 | 19 | """ 20 | 21 | # Check that observed_sample is an array! 22 | if not isinstance(observed_sample, np.ndarray): 23 | raise ValueError('The first parameter to bootstrap_statistic must be a sample represented as an array, not a value of type ' + str(type(observed_sample).__name__)) 24 | 25 | statistics = make_array() 26 | 27 | for i in np.arange(0, num_trials): 28 | #Key: in bootstrapping we must always sample with replacement 29 | simulated_resample = np.random.choice(observed_sample, len(observed_sample)) 30 | 31 | resample_statistic = compute_statistic(simulated_resample) 32 | statistics = np.append(statistics, resample_statistic) 33 | 34 | return statistics 35 | -------------------------------------------------------------------------------- /samples/python/marble-sample.csv: -------------------------------------------------------------------------------- 1 | color 2 | R 3 | R 4 | R 5 | R 6 | R 7 | R 8 | R 9 | R 10 | R 11 | B 12 | B 13 | B 14 | B 15 | B 16 | B 17 | B 18 | B 19 | B 20 | B 21 | B 22 | B 23 | B 24 | B 25 | B 26 | B 27 | B 28 | B 29 | B 30 | B 31 | B 32 | -------------------------------------------------------------------------------- /samples/python/marbles.py: -------------------------------------------------------------------------------- 1 | from datascience import * 2 | from ds101 import * 3 | 4 | def make_marble_bag(): 5 | table = Table().read_table('marble-sample.csv') 6 | return table.column('color') 7 | 8 | def ratio(x,y): 9 | return x / y 10 | 11 | def ratio_blue_to_red(sample): 12 | blues = np.count_nonzero(sample == 'B') 13 | reds = np.count_nonzero(sample == 'r') 14 | return ratio(blues, reds) 15 | 16 | marbles = make_marble_bag() 17 | if 'R' in marbles: 18 | print(ratio_blue_to_red(marbles)) 19 | -------------------------------------------------------------------------------- /samples/python/mean.py: -------------------------------------------------------------------------------- 1 | 2 | import numpy as np 3 | from datascience import * 4 | from ds101 import bootstrap_statistic 5 | 6 | def make_marble_bag(): 7 | table = Table().read_table('marble-sample.csv') 8 | return table.column('color') 9 | 10 | observed_marbles = make_marble_bag() 11 | 12 | def percent_blue(sample): 13 | return np.count_nonzero(sample == 'B') / len(sample) 14 | 15 | def main(): 16 | 17 | num_trials = 5 18 | 19 | stats = bootstrap_statistic(observed_marbles, 20 | percent_blue, 21 | num_trials) 22 | 23 | assert np.isclose(np.mean(stats), 0.7) 24 | 25 | main() -------------------------------------------------------------------------------- /samples/python/nb.ipynb: -------------------------------------------------------------------------------- 1 | { 2 | "cells": [ 3 | { 4 | "cell_type": "markdown", 5 | "id": "9706dd9f-da01-4afc-8148-d5e51696e281", 6 | "metadata": {}, 7 | "source": [ 8 | "**You must set the kernel to be the ipyflow kernel for ChatDBG to work.**" 9 | ] 10 | }, 11 | { 12 | "cell_type": "code", 13 | "execution_count": null, 14 | "id": "637bdd5a-8e45-4f1d-8ad6-6e4d75e60d66", 15 | "metadata": {}, 16 | "outputs": [], 17 | "source": [ 18 | "from datascience import *\n", 19 | "%pdb" 20 | ] 21 | }, 22 | { 23 | "cell_type": "code", 24 | "execution_count": null, 25 | "id": "66b2ebbf-584a-4c5f-885d-f2ad667cadd1", 26 | "metadata": {}, 27 | "outputs": [], 28 | "source": [ 29 | "red_blue_proportions = make_array(0.3, 0.7)" 30 | ] 31 | }, 32 | { 33 | "cell_type": "code", 34 | "execution_count": null, 35 | "id": "802a3ab2-599d-49dd-a5b4-3139d5e34269", 36 | "metadata": {}, 37 | "outputs": [], 38 | "source": [ 39 | "def make_marble_bag(size):\n", 40 | " return size * sample_proportions(red_blue_proportions, size)" 41 | ] 42 | }, 43 | { 44 | "cell_type": "code", 45 | "execution_count": null, 46 | "id": "b46dc5a5-c376-40b3-a98d-da0ea012c80a", 47 | "metadata": {}, 48 | "outputs": [], 49 | "source": [ 50 | "make_marble_bag(10)" 51 | ] 52 | } 53 | ], 54 | "metadata": { 55 | "ipyflow": { 56 | "cell_children": { 57 | "11a5ccf7-17d6-4600-ad3b-4369d00eac1a": [], 58 | "44d121b5-07d7-4b90-8fa5-bd0ad7ee2623": [], 59 | "45d9fcde-3d09-4bd0-a1ac-f8470fc81250": [], 60 | "637bdd5a-8e45-4f1d-8ad6-6e4d75e60d66": [], 61 | "66b2ebbf-584a-4c5f-885d-f2ad667cadd1": [ 62 | "b46dc5a5-c376-40b3-a98d-da0ea012c80a" 63 | ], 64 | "679500c8-471d-4dd8-aad4-09f1942805fb": [], 65 | "802a3ab2-599d-49dd-a5b4-3139d5e34269": [ 66 | "b46dc5a5-c376-40b3-a98d-da0ea012c80a" 67 | ], 68 | "a0ef25d9-d30c-4577-8285-c21fdd1fac2d": [], 69 | "b46dc5a5-c376-40b3-a98d-da0ea012c80a": [], 70 | "c13a73c3-68cb-4172-8f9d-0b0387b0c927": [ 71 | "f7c07ae1-735d-4093-bd3e-7efa9291f2d2" 72 | ], 73 | "cf1aa8d4-c2fb-4c68-9861-3431df761d1a": [], 74 | "f7c07ae1-735d-4093-bd3e-7efa9291f2d2": [] 75 | }, 76 | "cell_parents": { 77 | "11a5ccf7-17d6-4600-ad3b-4369d00eac1a": [], 78 | "44d121b5-07d7-4b90-8fa5-bd0ad7ee2623": [], 79 | "45d9fcde-3d09-4bd0-a1ac-f8470fc81250": [], 80 | "637bdd5a-8e45-4f1d-8ad6-6e4d75e60d66": [], 81 | "66b2ebbf-584a-4c5f-885d-f2ad667cadd1": [], 82 | "679500c8-471d-4dd8-aad4-09f1942805fb": [], 83 | "802a3ab2-599d-49dd-a5b4-3139d5e34269": [], 84 | "a0ef25d9-d30c-4577-8285-c21fdd1fac2d": [], 85 | "b46dc5a5-c376-40b3-a98d-da0ea012c80a": [ 86 | "66b2ebbf-584a-4c5f-885d-f2ad667cadd1", 87 | "802a3ab2-599d-49dd-a5b4-3139d5e34269" 88 | ], 89 | "c13a73c3-68cb-4172-8f9d-0b0387b0c927": [], 90 | "cf1aa8d4-c2fb-4c68-9861-3431df761d1a": [], 91 | "f7c07ae1-735d-4093-bd3e-7efa9291f2d2": [ 92 | "c13a73c3-68cb-4172-8f9d-0b0387b0c927" 93 | ] 94 | } 95 | }, 96 | "kernelspec": { 97 | "display_name": "Python 3 (ipyflow)", 98 | "language": "python", 99 | "name": "ipyflow" 100 | }, 101 | "language_info": { 102 | "codemirror_mode": { 103 | "name": "ipython", 104 | "version": 3 105 | }, 106 | "file_extension": ".py", 107 | "mimetype": "text/x-python", 108 | "name": "python", 109 | "nbconvert_exporter": "python", 110 | "pygments_lexer": "ipython3", 111 | "version": "3.11.7" 112 | } 113 | }, 114 | "nbformat": 4, 115 | "nbformat_minor": 5 116 | } 117 | -------------------------------------------------------------------------------- /samples/python/requirements.txt: -------------------------------------------------------------------------------- 1 | # packages necessary for the test files 2 | datascience 3 | -------------------------------------------------------------------------------- /samples/python/sample.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | 3 | red_blue_proportions = np.array([0.3, 0.7]) 4 | 5 | def make_sample(sample_size, probabilities): 6 | return np.random.multinomial(sample_size, probabilities) 7 | 8 | def make_marble_bag(size): 9 | return make_sample(red_blue_proportions, size) 10 | 11 | make_marble_bag(10) 12 | -------------------------------------------------------------------------------- /samples/python/testme.py: -------------------------------------------------------------------------------- 1 | def tryme(x): 2 | count = 0 3 | for i in range(100): 4 | if x / i > 2: 5 | count += 1 6 | return count 7 | 8 | if __name__ == '__main__': 9 | print(tryme(100)) 10 | 11 | 12 | 13 | -------------------------------------------------------------------------------- /samples/rust/.gitignore: -------------------------------------------------------------------------------- 1 | target 2 | Cargo.lock 3 | panic_log.txt 4 | -------------------------------------------------------------------------------- /samples/rust/Cargo.toml: -------------------------------------------------------------------------------- 1 | [package] 2 | name = "chatdbg_rust_tests" 3 | version = "1.0.0" 4 | edition = "2021" 5 | 6 | [[bin]] 7 | name = "test-failed-assert" 8 | path = "test-failed-assert.rs" 9 | 10 | [dependencies] 11 | chatdbg = { path = "../../rust-support/chatdbg" } 12 | 13 | [profile.dev] 14 | panic = "abort" 15 | 16 | [profile.release] 17 | panic = "abort" 18 | 19 | -------------------------------------------------------------------------------- /samples/rust/test-failed-assert.rs: -------------------------------------------------------------------------------- 1 | fn fact(n: f32) -> f32 { 2 | let mut x = 1.0; 3 | for i in 0..n as i32 { 4 | x *= i as f32; 5 | } 6 | assert!(x != 0.0); 7 | x 8 | } 9 | 10 | #[chatdbg::main] 11 | fn main() { 12 | println!("{}", fact(100.0)); 13 | } 14 | -------------------------------------------------------------------------------- /src/chatdbg/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/plasma-umass/ChatDBG/805a3d1900aebaaa26603b334e688d70c3d3a7bf/src/chatdbg/__init__.py -------------------------------------------------------------------------------- /src/chatdbg/__main__.py: -------------------------------------------------------------------------------- 1 | import sys 2 | from getopt import GetoptError 3 | 4 | import ipdb 5 | 6 | from chatdbg.chatdbg_pdb import ChatDBG 7 | from chatdbg.util.config import chatdbg_config 8 | from chatdbg.util.help import print_help 9 | 10 | 11 | def main() -> None: 12 | ipdb.__main__._get_debugger_cls = lambda: ChatDBG 13 | 14 | args = chatdbg_config.parse_user_flags(sys.argv[1:]) 15 | 16 | if "-h" in args or "--help" in args: 17 | print_help() 18 | 19 | sys.argv = [sys.argv[0]] + args 20 | 21 | try: 22 | ipdb.__main__.main() 23 | except GetoptError as e: 24 | print(f"Unrecognized option: {e.opt}\n") 25 | print_help() 26 | sys.exit(1) 27 | 28 | 29 | if __name__ == "__main__": 30 | main() 31 | -------------------------------------------------------------------------------- /src/chatdbg/assistant/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/plasma-umass/ChatDBG/805a3d1900aebaaa26603b334e688d70c3d3a7bf/src/chatdbg/assistant/__init__.py -------------------------------------------------------------------------------- /src/chatdbg/assistant/assistant.py: -------------------------------------------------------------------------------- 1 | import json 2 | import string 3 | import textwrap 4 | import time 5 | import pprint 6 | 7 | import warnings 8 | 9 | with warnings.catch_warnings(): 10 | warnings.simplefilter("ignore") 11 | import litellm 12 | 13 | import openai 14 | 15 | from ..util.trim import sandwich_tokens, trim_messages 16 | from ..util.text import strip_ansi 17 | from .listeners import Printer 18 | 19 | 20 | class AssistantError(Exception): 21 | def __init__(self, *args: object) -> None: 22 | super().__init__(*args) 23 | 24 | 25 | def remove_non_printable_chars(s: str) -> str: 26 | printable_chars = set(string.printable) 27 | filtered_string = "".join(filter(lambda x: x in printable_chars, s)) 28 | return filtered_string 29 | 30 | 31 | class Assistant: 32 | def __init__( 33 | self, 34 | instructions, 35 | model="gpt-4o", 36 | timeout=30, 37 | listeners=[Printer()], 38 | functions=[], 39 | max_call_response_tokens=2048, 40 | ): 41 | 42 | # Hide their debugging info -- it messes with our error handling 43 | litellm.suppress_debug_info = True 44 | 45 | self._clients = listeners 46 | 47 | self._functions = {} 48 | for f in functions: 49 | self._add_function(f) 50 | 51 | self._model = model 52 | self._timeout = timeout 53 | self._conversation = [{"role": "system", "content": instructions}] 54 | self._max_call_response_tokens = max_call_response_tokens 55 | 56 | self._check_model() 57 | self._broadcast("on_begin_dialog", instructions) 58 | 59 | def close(self): 60 | self._broadcast("on_end_dialog") 61 | 62 | def _warn_about_exception(self, e, message="Unexpected Exception"): 63 | import traceback 64 | 65 | tb_lines = traceback.format_exception(type(e), e, e.__traceback__) 66 | tb_string = "".join(tb_lines) 67 | self._broadcast("on_error", f"{message}\n\n{e}\n{tb_string}") 68 | 69 | def query(self, prompt: str, user_text): 70 | """ 71 | Send a query to the LLM. 72 | - prompt is the prompt to send. 73 | - user_text is what the user typed (which may or not be the same as prompt) 74 | 75 | Returns a dictionary containing: 76 | - "completed": True of the query ran to completion. 77 | - "cost": Cost of query, or 0 if not completed. 78 | Other fields only if completed is True 79 | - "time": completion time in seconds 80 | - "model": the model used. 81 | - "tokens": total tokens 82 | - "prompt_tokens": our prompts 83 | - "completion_tokens": the LLM completions part 84 | """ 85 | stats = {"completed": False, "cost": 0} 86 | start = time.time() 87 | 88 | self._broadcast("on_begin_query", prompt, user_text) 89 | try: 90 | stats = self._streamed_query(prompt, user_text) 91 | elapsed = time.time() - start 92 | 93 | stats["time"] = elapsed 94 | stats["model"] = self._model 95 | stats["completed"] = True 96 | stats["message"] = f"\n[Cost: ~${stats['cost']:.2f} USD]" 97 | except openai.OpenAIError as e: 98 | self._warn_about_exception(e, f"Unexpected OpenAI Error. Retry the query.") 99 | stats["message"] = f"[Exception: {e}]" 100 | except KeyboardInterrupt: 101 | # user action -- just ignore 102 | stats["message"] = "[Chat Interrupted]" 103 | except Exception as e: 104 | self._warn_about_exception(e, f"Unexpected Exception.") 105 | stats["message"] = f"[Exception: {e}]" 106 | 107 | self._broadcast("on_end_query", stats) 108 | return stats 109 | 110 | def _report(self, stats): 111 | if stats["completed"]: 112 | print() 113 | else: 114 | print("[Chat Interrupted]") 115 | 116 | def _broadcast(self, method_name, *args): 117 | for client in self._clients: 118 | method = getattr(client, method_name, None) 119 | if callable(method): 120 | method(*args) 121 | 122 | def _check_model(self): 123 | result = litellm.validate_environment(self._model) 124 | missing_keys = result["missing_keys"] 125 | if missing_keys != []: 126 | _, provider, _, _ = litellm.get_llm_provider(self._model) 127 | if provider == "openai": 128 | raise AssistantError( 129 | textwrap.dedent( 130 | f"""\ 131 | You need an OpenAI key to use the {self._model} model. 132 | You can get a key here: https://platform.openai.com/api-keys. 133 | Set the environment variable OPENAI_API_KEY to your key value.""" 134 | ) 135 | ) 136 | else: 137 | raise AssistantError( 138 | textwrap.dedent( 139 | f"""\ 140 | You need to set the following environment variables 141 | to use the {self._model} model: {', '.join(missing_keys)}.""" 142 | ) 143 | ) 144 | 145 | try: 146 | if not litellm.supports_function_calling(self._model): 147 | raise AssistantError( 148 | textwrap.dedent( 149 | f"""\ 150 | The {self._model} model does not support function calls. 151 | You must use a model that does, eg. gpt-4.""" 152 | ) 153 | ) 154 | except: 155 | raise AssistantError( 156 | textwrap.dedent( 157 | f"""\ 158 | {self._model} does not appear to be a supported model. 159 | See https://docs.litellm.ai/docs/providers.""" 160 | ) 161 | ) 162 | 163 | def _add_function(self, function): 164 | """ 165 | Add a new function to the list of function tools. 166 | The function should have the necessary json spec as its docstring 167 | """ 168 | schema = json.loads(function.__doc__) 169 | assert "name" in schema, "Bad JSON in docstring for function tool." 170 | self._functions[schema["name"]] = {"function": function, "schema": schema} 171 | 172 | def _make_call(self, tool_call) -> str: 173 | name = tool_call.function.name 174 | try: 175 | args = json.loads(tool_call.function.arguments) 176 | function = self._functions[name] 177 | call, result = function["function"](**args) 178 | result = remove_non_printable_chars(strip_ansi(result).expandtabs()) 179 | self._broadcast("on_function_call", call, result) 180 | except KeyboardInterrupt as e: 181 | raise e 182 | except Exception as e: 183 | # likely to be an exception from the code we ran, not a bug... 184 | result = f"Exception in function call: {e}" 185 | self._broadcast("on_warn", result) 186 | return result 187 | 188 | def _streamed_query(self, prompt: str, user_text): 189 | cost = 0 190 | 191 | self._conversation.append({"role": "user", "content": prompt}) 192 | 193 | while True: 194 | stream = self._stream_completion() 195 | 196 | # litellm.stream_chunk_builder is broken for new GPT models 197 | # that have content before calls, so... 198 | 199 | # stream the response, collecting the tool_call parts separately 200 | # from the content 201 | try: 202 | self._broadcast("on_begin_stream") 203 | chunks = [] 204 | tool_chunks = [] 205 | for chunk in stream: 206 | chunks.append(chunk) 207 | if chunk.choices[0].delta.content: 208 | self._broadcast( 209 | "on_stream_delta", chunk.choices[0].delta.content 210 | ) 211 | else: 212 | tool_chunks.append(chunk) 213 | finally: 214 | self._broadcast("on_end_stream") 215 | 216 | # then compute for the part that litellm gives back. 217 | completion = litellm.stream_chunk_builder( 218 | chunks, messages=self._conversation 219 | ) 220 | cost += litellm.completion_cost(completion) 221 | 222 | # add content to conversation, but if there is no content, then the message 223 | # has only tool calls, and skip this step 224 | response_message = completion.choices[0].message 225 | if response_message.content != None: 226 | # fix: remove tool calls. They get added below. 227 | response_message = response_message.copy() 228 | response_message["tool_calls"] = None 229 | self._conversation.append(response_message.json()) 230 | 231 | if response_message.content != None: 232 | self._broadcast("on_response", response_message.content) 233 | 234 | if completion.choices[0].finish_reason == "tool_calls": 235 | # create a message with just the tool calls, append that to the conversation, and generate the responses. 236 | tool_completion = litellm.stream_chunk_builder( 237 | tool_chunks, self._conversation 238 | ) 239 | 240 | # this part wasn't counted above... 241 | cost += litellm.completion_cost(tool_completion) 242 | 243 | tool_message = tool_completion.choices[0].message 244 | 245 | tool_json = tool_message.json() 246 | 247 | # patch for litellm sometimes putting index fields in the tool calls it constructs 248 | # in stream_chunk_builder. gpt-4-turbo-2024-04-09 can't handle those index fields, so 249 | # just remove them for the moment. 250 | for tool_call in tool_json.get("tool_calls", []): 251 | _ = tool_call.pop("index", None) 252 | 253 | tool_json["role"] = "assistant" 254 | self._conversation.append(tool_json) 255 | self._add_function_results_to_conversation(tool_message) 256 | else: 257 | break 258 | 259 | stats = { 260 | "cost": cost, 261 | "tokens": completion.usage.total_tokens, 262 | "prompt_tokens": completion.usage.prompt_tokens, 263 | "completion_tokens": completion.usage.completion_tokens, 264 | } 265 | return stats 266 | 267 | def _stream_completion(self): 268 | 269 | self._trim_conversation() 270 | 271 | return litellm.completion( 272 | model=self._model, 273 | messages=self._conversation, 274 | tools=[ 275 | {"type": "function", "function": f["schema"]} 276 | for f in self._functions.values() 277 | ], 278 | timeout=self._timeout, 279 | stream=True, 280 | ) 281 | 282 | def _trim_conversation(self): 283 | old_len = litellm.token_counter(self._model, messages=self._conversation) 284 | 285 | self._conversation = trim_messages(self._conversation, self._model) 286 | 287 | new_len = litellm.token_counter(self._model, messages=self._conversation) 288 | if old_len != new_len: 289 | self._broadcast( 290 | "on_warn", f"Trimming conversation from {old_len} to {new_len} tokens." 291 | ) 292 | 293 | def _add_function_results_to_conversation(self, response_message): 294 | response_message["role"] = "assistant" 295 | tool_calls = response_message.tool_calls 296 | try: 297 | for tool_call in tool_calls: 298 | function_response = self._make_call(tool_call) 299 | function_response = sandwich_tokens( 300 | function_response, self._model, self._max_call_response_tokens, 0.5 301 | ) 302 | response = { 303 | "tool_call_id": tool_call.id, 304 | "role": "tool", 305 | "name": tool_call.function.name, 306 | "content": function_response, 307 | } 308 | self._conversation.append(response) 309 | except Exception as e: 310 | # Warning: potential infinite loop if the LLM keeps sending 311 | # the same bad call. 312 | self._broadcast( 313 | "on_error", f"An exception occured while processing tool calls: {e}" 314 | ) 315 | -------------------------------------------------------------------------------- /src/chatdbg/assistant/listeners.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import textwrap 3 | 4 | 5 | class BaseAssistantListener: 6 | """ 7 | Events that the Assistant generates. Override these for the client. 8 | """ 9 | 10 | # Dialogs capture 1 or more queries. 11 | 12 | def on_begin_dialog(self, instructions): 13 | pass 14 | 15 | def on_end_dialog(self): 16 | pass 17 | 18 | # Events for a single query 19 | 20 | def on_begin_query(self, prompt, user_text): 21 | pass 22 | 23 | def on_response(self, text): 24 | pass 25 | 26 | def on_function_call(self, call, result): 27 | pass 28 | 29 | def on_end_query(self, stats): 30 | pass 31 | 32 | # For clients wishing to stream responses 33 | 34 | def on_begin_stream(self): 35 | pass 36 | 37 | def on_stream_delta(self, text): 38 | pass 39 | 40 | def on_end_stream(self): 41 | pass 42 | 43 | # Notifications of non-fatal problems 44 | 45 | def on_warn(self, text): 46 | pass 47 | 48 | def on_error(self, text): 49 | pass 50 | 51 | 52 | class Printer(BaseAssistantListener): 53 | def __init__(self, out=sys.stdout): 54 | self.out = out 55 | 56 | def on_warn(self, text): 57 | print(textwrap.indent(text, "*** "), file=self.out) 58 | 59 | def on_error(self, text): 60 | print(textwrap.indent(text, "*** "), file=self.out) 61 | 62 | def on_begin_stream(self): 63 | pass 64 | 65 | def on_stream_delta(self, text): 66 | print(text, end="", file=self.out, flush=True) 67 | 68 | def on_end_stream(self): 69 | pass 70 | 71 | def on_begin_query(self, prompt, user_text): 72 | pass 73 | 74 | def on_end_query(self, stats): 75 | pass 76 | 77 | def on_response(self, text): 78 | if text != None: 79 | print(text, file=self.out) 80 | 81 | def on_function_call(self, call, result): 82 | if result and len(result) > 0: 83 | entry = f"{call}\n{result}" 84 | else: 85 | entry = f"{call}" 86 | print(entry, file=self.out) 87 | 88 | 89 | class StreamingPrinter(Printer): 90 | def __init__(self, out=sys.stdout): 91 | super().__init__(out) 92 | 93 | def on_begin_stream(self): 94 | print("", flush=True) 95 | 96 | def on_stream_delta(self, text): 97 | print(text, end="", file=self.out, flush=True) 98 | 99 | def on_end_stream(self): 100 | print("", flush=True) 101 | 102 | def on_response(self, text): 103 | pass 104 | -------------------------------------------------------------------------------- /src/chatdbg/chatdbg_gdb.py: -------------------------------------------------------------------------------- 1 | import os 2 | import atexit 3 | from typing import List, Optional, Union 4 | 5 | import gdb 6 | 7 | from chatdbg.native_util import clangd_lsp_integration 8 | from chatdbg.native_util.code import code 9 | from chatdbg.native_util.dbg_dialog import DBGDialog 10 | from chatdbg.native_util.stacks import ( 11 | _ArgumentEntry, 12 | _FrameSummaryEntry, 13 | _SkippedFramesEntry, 14 | ) 15 | from chatdbg.util.config import chatdbg_config 16 | from chatdbg.native_util.safety import command_is_safe 17 | from chatdbg.util.exit_message import chatdbg_was_called, print_exit_message 18 | 19 | # The file produced by the panic handler if the Rust program is using the chatdbg crate. 20 | RUST_PANIC_LOG_FILENAME = "panic_log.txt" 21 | PROMPT = "(ChatDBG gdb) " 22 | 23 | # Set the prompt to ChatDBG gdb 24 | gdb.prompt_hook = lambda current_prompt: PROMPT 25 | 26 | 27 | last_error_type = "" 28 | 29 | atexit.register(print_exit_message) 30 | 31 | 32 | def stop_handler(event): 33 | """Sets last error type so we can report it later.""" 34 | # Check if the event is a stop event 35 | global last_error_type 36 | if not hasattr(event, "stop_signal"): 37 | last_error_type = "" # Not a real error (e.g., a breakpoint) 38 | return 39 | if event.stop_signal is not None: 40 | last_error_type = event.stop_signal 41 | 42 | 43 | gdb.events.stop.connect(stop_handler) 44 | 45 | 46 | class Code(gdb.Command): 47 | 48 | def __init__(self): 49 | gdb.Command.__init__(self, "code", gdb.COMMAND_USER) 50 | 51 | def invoke(self, command, from_tty): 52 | print(code(command)) 53 | return 54 | 55 | 56 | Code() 57 | 58 | 59 | class Definition(gdb.Command): 60 | 61 | def __init__(self): 62 | gdb.Command.__init__(self, "definition", gdb.COMMAND_USER) 63 | 64 | def invoke(self, command, from_tty): 65 | print(clangd_lsp_integration.native_definition(command)) 66 | return 67 | 68 | 69 | Definition() 70 | 71 | 72 | class Config(gdb.Command): 73 | 74 | def __init__(self): 75 | gdb.Command.__init__(self, "config", gdb.COMMAND_USER) 76 | 77 | def invoke(self, command, from_tty): 78 | args = command.split() 79 | message = chatdbg_config.parse_only_user_flags(args) 80 | print(message) 81 | return 82 | 83 | 84 | Config() 85 | 86 | 87 | # Implement the command `why` 88 | class Why(gdb.Command): 89 | """Provides root cause analysis for a failure.""" 90 | 91 | def __init__(self): 92 | gdb.Command.__init__(self, "why", gdb.COMMAND_USER) 93 | 94 | def invoke(self, command, from_tty): 95 | try: 96 | dialog = GDBDialog(PROMPT) 97 | dialog.dialog(command) 98 | except Exception as e: 99 | print(str(e)) 100 | return 101 | 102 | 103 | Why() 104 | 105 | gdb.execute("alias chat = why") 106 | 107 | 108 | class GDBDialog(DBGDialog): 109 | 110 | def __init__(self, prompt) -> None: 111 | chatdbg_was_called() 112 | super().__init__(prompt) 113 | 114 | def _message_is_a_bad_command_error(self, message): 115 | return message.strip().startswith("Undefined command:") 116 | 117 | def _run_one_command(self, command): 118 | try: 119 | return gdb.execute(command, to_string=True) 120 | except Exception as e: 121 | return str(e) 122 | 123 | def check_debugger_state(self): 124 | global last_error_type 125 | if not last_error_type: 126 | # Assume we are running from a core dump, 127 | # which _probably_ means a SEGV. 128 | last_error_type = "SIGSEGV" 129 | try: 130 | frame = gdb.selected_frame() 131 | block = frame.block() 132 | except gdb.error: 133 | self.fail( 134 | "Must be attached to a program that fails to use `why` or `chat`." 135 | ) 136 | except RuntimeError: 137 | self.fail( 138 | "Your program must be compiled with debug information (`-g`) to use `why` or `chat`." 139 | ) 140 | 141 | def _get_frame_summaries( 142 | self, max_entries: int = 20 143 | ) -> Optional[List[Union[_FrameSummaryEntry, _SkippedFramesEntry]]]: 144 | thread = gdb.selected_thread() 145 | if not thread: 146 | return None 147 | 148 | skipped = 0 149 | summaries: List[Union[_FrameSummaryEntry, _SkippedFramesEntry]] = [] 150 | 151 | frame = gdb.selected_frame() 152 | 153 | index = -1 154 | # Walk the stack and build up the frames list. 155 | while frame is not None: 156 | index += 1 157 | 158 | name = frame.name() 159 | if not name: 160 | skipped += 1 161 | frame = frame.older() 162 | continue 163 | symtab_and_line = frame.find_sal() 164 | 165 | # Get frame file path 166 | if symtab_and_line.symtab is not None: 167 | file_path = symtab_and_line.symtab.fullname() 168 | if file_path == None: 169 | file_path = "[unknown]" 170 | else: 171 | # If we are in a subdirectory, use a relative path instead. 172 | if file_path.startswith(os.getcwd()): 173 | file_path = os.path.relpath(file_path) 174 | # Skip frames for which we have no source -- likely system frames. 175 | if not os.path.exists(file_path): 176 | skipped += 1 177 | frame = frame.older() 178 | continue 179 | else: 180 | file_path = None 181 | 182 | # Get frame lineno 183 | if symtab_and_line.line is not None: 184 | lineno = symtab_and_line.line 185 | else: 186 | lineno = None 187 | 188 | # Get arguments 189 | arguments: List[_ArgumentEntry] = [] 190 | block = gdb.Block 191 | try: 192 | block = frame.block() 193 | except Exception: 194 | skipped += 1 195 | frame = frame.older() 196 | continue 197 | for symbol in block: 198 | if symbol.is_argument: 199 | typename = symbol.type 200 | name = symbol.name 201 | value = str(frame.read_var(name)) 202 | arguments.append(_ArgumentEntry(typename, name, value)) 203 | 204 | if skipped > 0: 205 | summaries.append(_SkippedFramesEntry(skipped)) 206 | skipped = 0 207 | 208 | summaries.append( 209 | _FrameSummaryEntry(index, name, arguments, file_path, lineno) 210 | ) 211 | if len(summaries) >= max_entries: 212 | break 213 | frame = frame.older() 214 | 215 | if skipped > 0: 216 | summaries.append(_SkippedFramesEntry(skipped)) 217 | 218 | return summaries 219 | 220 | def _initial_prompt_error_message(self): 221 | # If the Rust panic log exists, append it to the error reason. 222 | global last_error_type 223 | try: 224 | with open(RUST_PANIC_LOG_FILENAME, "r") as log: 225 | panic_log = log.read() 226 | last_error_type = panic_log + "\n" + last_error_type 227 | except: 228 | pass 229 | return last_error_type 230 | 231 | def _initial_prompt_error_details(self): 232 | """Anything more beyond the initial error message to include.""" 233 | return None 234 | 235 | def _initial_prompt_command_line(self): 236 | executable_path = gdb.selected_inferior().progspace.filename 237 | 238 | if executable_path.startswith(os.getcwd()): 239 | executable_path = os.path.join(".", os.path.relpath(executable_path)) 240 | 241 | prefix = "Argument list to give program being debugged when it is started is " 242 | args = gdb.execute("show args", to_string=True).strip() 243 | if args.startswith(prefix): 244 | args = args[len(prefix) :].strip('."') 245 | 246 | return executable_path + " " + args 247 | 248 | def _initial_prompt_input(self): 249 | prefix = "Argument list to give program being debugged when it is started is " 250 | args = gdb.execute("show args", to_string=True).strip() 251 | if args.startswith(prefix): 252 | args = args[len(prefix) :].strip('."') 253 | 254 | input_pipe = args.find("<") 255 | if input_pipe != -1: 256 | input_file = args[input_pipe + 1 :].strip() 257 | try: 258 | content = open(input_file, "r").read() 259 | return content 260 | except Exception: 261 | self.fail(f"The detected input file {input_file} could not be read.") 262 | 263 | def _prompt_stack(self): 264 | """ 265 | Return a simple backtrace to show the LLM where we are on the stack 266 | in followup prompts. 267 | """ 268 | return None 269 | 270 | def llm_debug(self, command: str): 271 | """ 272 | { 273 | "name": "debug", 274 | "description": "The `debug` function runs a GDB command on the stopped program and gets the response.", 275 | "parameters": { 276 | "type": "object", 277 | "properties": { 278 | "command": { 279 | "type": "string", 280 | "description": "The GDB command to run, possibly with arguments." 281 | } 282 | }, 283 | "required": [ "command" ] 284 | } 285 | } 286 | """ 287 | if not chatdbg_config.unsafe and not command_is_safe(command): 288 | self._unsafe_cmd = True 289 | return command, f"Command `{command}` is not allowed." 290 | return command, self._run_one_command(command) 291 | -------------------------------------------------------------------------------- /src/chatdbg/chatdbg_lldb.py: -------------------------------------------------------------------------------- 1 | import json 2 | import os 3 | from typing import Optional, Union 4 | 5 | import lldb 6 | 7 | from chatdbg.native_util import clangd_lsp_integration 8 | from chatdbg.native_util.code import code 9 | from chatdbg.native_util.dbg_dialog import DBGDialog 10 | from chatdbg.native_util.stacks import ( 11 | _ArgumentEntry, 12 | _FrameSummaryEntry, 13 | _SkippedFramesEntry, 14 | ) 15 | from chatdbg.util.config import chatdbg_config 16 | from chatdbg.util.exit_message import chatdbg_was_called, print_exit_message 17 | from chatdbg.native_util.safety import command_is_safe 18 | 19 | 20 | # The file produced by the panic handler if the Rust program is using the chatdbg crate. 21 | RUST_PANIC_LOG_FILENAME = "panic_log.txt" 22 | PROMPT = "(ChatDBG lldb) " 23 | 24 | 25 | def __lldb_init_module(debugger: lldb.SBDebugger, internal_dict: dict) -> None: 26 | debugger.HandleCommand(f"settings set prompt '{PROMPT}'") 27 | debugger.SetDestroyCallback(print_exit_message) 28 | chatdbg_config.format = "md" 29 | 30 | 31 | @lldb.command("code") 32 | def _function_code( 33 | debugger: lldb.SBDebugger, 34 | command: str, 35 | result: lldb.SBCommandReturnObject, 36 | internal_dict: dict, 37 | ) -> None: 38 | result.AppendMessage(code(command)) 39 | 40 | 41 | @lldb.command("definition") 42 | def _function_definition( 43 | debugger: lldb.SBDebugger, 44 | command: str, 45 | result: lldb.SBCommandReturnObject, 46 | internal_dict: dict, 47 | ) -> None: 48 | result.AppendMessage(clangd_lsp_integration.native_definition(command)) 49 | 50 | 51 | @lldb.command("chat") 52 | @lldb.command("why") 53 | def chat( 54 | debugger: lldb.SBDebugger, 55 | command: str, 56 | result: lldb.SBCommandReturnObject, 57 | internal_dict: dict, 58 | ): 59 | try: 60 | dialog = LLDBDialog(PROMPT, debugger) 61 | dialog.dialog(command) 62 | except Exception as e: 63 | result.SetError(str(e)) 64 | 65 | 66 | @lldb.command("config") 67 | def config( 68 | debugger: lldb.SBDebugger, 69 | command: str, 70 | result: lldb.SBCommandReturnObject, 71 | internal_dict: dict, 72 | ): 73 | args = command.split() 74 | message = chatdbg_config.parse_only_user_flags(args) 75 | result.AppendMessage(message) 76 | 77 | 78 | class LLDBDialog(DBGDialog): 79 | 80 | def __init__(self, prompt, debugger) -> None: 81 | super().__init__(prompt) 82 | chatdbg_was_called() 83 | self._debugger = debugger 84 | 85 | def _message_is_a_bad_command_error(self, message): 86 | return message.strip().endswith("is not a valid command.") 87 | 88 | def _run_one_command(self, command): 89 | interpreter = self._debugger.GetCommandInterpreter() 90 | result = lldb.SBCommandReturnObject() 91 | interpreter.HandleCommand(command, result) 92 | 93 | if result.Succeeded(): 94 | return result.GetOutput() 95 | else: 96 | return result.GetError() 97 | 98 | def _is_debug_build(self) -> bool: 99 | """Returns False if not compiled with debug information.""" 100 | target = self._debugger.GetSelectedTarget() 101 | if not target: 102 | return False 103 | for module in target.module_iter(): 104 | for cu in module.compile_unit_iter(): 105 | for line_entry in cu: 106 | if line_entry.GetLine() > 0: 107 | return True 108 | return False 109 | 110 | def get_thread(self) -> Optional[lldb.SBThread]: 111 | """ 112 | Returns a currently stopped thread in the debugged process. 113 | :return: A currently stopped thread or None if no thread is stopped. 114 | """ 115 | process = self._get_process() 116 | if not process: 117 | return None 118 | for thread in process: 119 | reason = thread.GetStopReason() 120 | if reason not in [lldb.eStopReasonNone, lldb.eStopReasonInvalid]: 121 | return thread 122 | return thread 123 | 124 | def check_debugger_state(self): 125 | if not self._debugger.GetSelectedTarget(): 126 | self.fail("Must be attached to a program to use `why` or `chat`.") 127 | 128 | elif not self._is_debug_build(): 129 | self.fail( 130 | "Your program must be compiled with debug information (`-g`) to use `why` or `chat`." 131 | ) 132 | 133 | thread = self.get_thread() 134 | if not thread: 135 | self.fail("must run the code first to use `chat`.") 136 | 137 | if not clangd_lsp_integration.is_available(): 138 | self.warn( 139 | "`clangd` was not found. The `find_definition` function will not be made available." 140 | ) 141 | 142 | def _get_frame_summaries( 143 | self, max_entries: int = 20 144 | ) -> Optional[list[Union[_FrameSummaryEntry, _SkippedFramesEntry]]]: 145 | thread = self.get_thread() 146 | if not thread: 147 | return None 148 | 149 | skipped = 0 150 | summaries: list[Union[_FrameSummaryEntry, _SkippedFramesEntry]] = [] 151 | 152 | index = -1 153 | # For each frame in thread 154 | for frame in thread: 155 | index += 1 156 | if not frame.GetDisplayFunctionName(): 157 | skipped += 1 158 | continue 159 | name = frame.GetDisplayFunctionName().split("(")[0] 160 | # Get function arguments, store as _ArgumentEntries 161 | arguments: list[_ArgumentEntry] = [] 162 | for j in range( 163 | frame.GetFunction().GetType().GetFunctionArgumentTypes().GetSize() 164 | ): 165 | arg = frame.FindVariable(frame.GetFunction().GetArgumentName(j)) 166 | if not arg: 167 | arguments.append( 168 | _ArgumentEntry("[unknown]", "[unknown]", "[unknown]") 169 | ) 170 | continue 171 | # TODO: Check if we should simplify / truncate types, e.g. std::unordered_map. 172 | arguments.append( 173 | _ArgumentEntry(arg.GetTypeName(), arg.GetName(), arg.GetValue()) 174 | ) 175 | 176 | # Look for paths to the function file. If there's no source, skip frame. 177 | line_entry = frame.GetLineEntry() 178 | file_path = line_entry.GetFileSpec().fullpath 179 | if file_path == None: 180 | file_path = "[unknown]" 181 | lineno = line_entry.GetLine() 182 | 183 | # If we are in a subdirectory, use a relative path instead. 184 | if file_path.startswith(os.getcwd()): 185 | file_path = os.path.relpath(file_path) 186 | 187 | # Skip frames for which we have no source -- likely system frames. 188 | if not os.path.exists(file_path): 189 | skipped += 1 190 | continue 191 | 192 | # Add _SkippedFramesEntry onto summaries list 193 | if skipped > 0: 194 | summaries.append(_SkippedFramesEntry(skipped)) 195 | skipped = 0 196 | 197 | # Otherwise, add _FrameSummaryEntries until max_entries, then break 198 | summaries.append( 199 | _FrameSummaryEntry(index, name, arguments, file_path, lineno) 200 | ) 201 | if len(summaries) >= max_entries: 202 | break 203 | 204 | if skipped > 0: 205 | summaries.append(_SkippedFramesEntry(skipped)) 206 | if len(summaries) > max_entries: 207 | summaries.pop(-2) 208 | 209 | total_summary_count = sum( 210 | [s.count() if isinstance(s, _SkippedFramesEntry) else 1 for s in summaries] 211 | ) 212 | 213 | if total_summary_count < len(thread): 214 | if isinstance(summaries[-1], _SkippedFramesEntry): 215 | summaries[-1] = _SkippedFramesEntry( 216 | len(thread) - total_summary_count + summaries[-1].count() 217 | ) 218 | else: 219 | summaries.append( 220 | _SkippedFramesEntry(len(thread) - total_summary_count + 1) 221 | ) 222 | if len(summaries) > max_entries: 223 | summaries.pop(-2) 224 | 225 | assert sum( 226 | [s.count() if isinstance(s, _SkippedFramesEntry) else 1 for s in summaries] 227 | ) == len(thread) 228 | 229 | return summaries 230 | 231 | def _get_process(self) -> Optional[lldb.SBProcess]: 232 | """ 233 | Get the process that the current target owns. 234 | :return: An lldb object representing the process (lldb.SBProcess) that this target owns. 235 | """ 236 | target = self._debugger.GetSelectedTarget() 237 | return target.process if target else None 238 | 239 | def _initial_prompt_error_message(self): 240 | thread = self.get_thread() 241 | 242 | error_message = thread.GetStopDescription(1024) if thread else None 243 | if error_message: 244 | return error_message 245 | else: 246 | self.warn("could not generate an error message.") 247 | return None 248 | 249 | def _initial_prompt_command_line(self): 250 | executable = self._debugger.GetSelectedTarget().GetExecutable() 251 | 252 | executable_path = os.path.join( 253 | executable.GetDirectory(), executable.GetFilename() 254 | ) 255 | if executable_path.startswith(os.getcwd()): 256 | executable_path = os.path.join(".", os.path.relpath(executable_path)) 257 | 258 | command_line_arguments = [ 259 | self._debugger.GetSelectedTarget().GetLaunchInfo().GetArgumentAtIndex(i) 260 | for i in range( 261 | self._debugger.GetSelectedTarget().GetLaunchInfo().GetNumArguments() 262 | ) 263 | ] 264 | 265 | command_line_invocation = " ".join([executable_path, *command_line_arguments]) 266 | if command_line_invocation: 267 | return command_line_invocation 268 | else: 269 | self.warn("could not retrieve the command line invocation.") 270 | return None 271 | 272 | def _initial_prompt_input(self): 273 | stream = lldb.SBStream() 274 | self._debugger.GetSetting("target.input-path").GetAsJSON(stream) 275 | entry = json.loads(stream.GetData()) 276 | 277 | input_path = entry if entry else None 278 | if input_path: 279 | try: 280 | with open(input_path, "r", errors="ignore") as file: 281 | input_contents = file.read() 282 | return input_contents 283 | except FileNotFoundError: 284 | self.warn("could not retrieve the input data.") 285 | return None 286 | 287 | def _initial_prompt_error_details(self): 288 | """Anything more beyond the initial error message to include.""" 289 | return None 290 | 291 | def _prompt_stack(self): 292 | """ 293 | Return a simple backtrace to show the LLM where we are on the stack 294 | in followup prompts. 295 | """ 296 | return None 297 | 298 | def llm_debug(self, command: str): 299 | """ 300 | { 301 | "name": "debug", 302 | "description": "The `debug` function runs an LLDB command on the stopped program and gets the response.", 303 | "parameters": { 304 | "type": "object", 305 | "properties": { 306 | "command": { 307 | "type": "string", 308 | "description": "The LLDB command to run, possibly with arguments." 309 | } 310 | }, 311 | "required": [ "command" ] 312 | } 313 | } 314 | """ 315 | if not chatdbg_config.unsafe and not command_is_safe(command): 316 | self._unsafe_cmd = True 317 | return command, f"Command `{command}` is not allowed." 318 | return command, self._run_one_command(command) 319 | -------------------------------------------------------------------------------- /src/chatdbg/custom_pdb/prompts.py: -------------------------------------------------------------------------------- 1 | _intro = f"""\ 2 | You are a debugging assistant. You will be given a Python stack trace for an 3 | error and answer questions related to the root cause of the error. 4 | """ 5 | 6 | _pdb_function = f"""\ 7 | Call the `pdb` function to run Pdb debugger commands on the stopped program. You 8 | may call the `pdb` function to run the following commands: `bt`, `up`, `down`, 9 | `p expression`, `list`. 10 | 11 | Call `pdb` to print any variable value or expression that you believe may 12 | contribute to the error. 13 | """ 14 | 15 | 16 | _info_function = """\ 17 | Call the `info` function to get the documentation and source code for any 18 | variable, function, package, class, method reference, field reference, or 19 | dotted reference visible in the current frame. Examples include: n, e.n 20 | where e is an expression, and t.n where t is a type. 21 | 22 | Unless it is from a common, widely-used library, you MUST call `info` exactly once on any 23 | symbol that is referenced in code leading up to the error. 24 | """ 25 | 26 | 27 | _slice_function = """\ 28 | Call the `slice` function to get the code used to produce 29 | the value currently stored a variable. You MUST call `slice` exactly once on any 30 | variable used but not defined in the current frame's code. 31 | """ 32 | 33 | _take_the_wheel_instructions = """\ 34 | Call the provided functions as many times as you would like. 35 | """ 36 | 37 | _general_instructions = f"""\ 38 | The root cause of any error is likely due to a problem in the source code from the user. 39 | 40 | Explain why each variable contributing to the error has been set 41 | to the value that it has. 42 | 43 | Continue with your explanations until you reach the root cause of the error. Your answer may be as long as necessary. 44 | 45 | End your answer with a section titled "##### Recommendation\\n" that contains one of: 46 | * a fix if you have identified the root cause 47 | * a numbered list of 1-3 suggestions for how to continue debugging if you have not 48 | """ 49 | 50 | 51 | _wheel_and_slice = f"""\ 52 | {_intro} 53 | {_pdb_function} 54 | {_info_function} 55 | {_slice_function} 56 | {_take_the_wheel_instructions} 57 | {_general_instructions} 58 | """ 59 | 60 | _wheel_no_slice = f"""\ 61 | {_intro} 62 | {_pdb_function} 63 | {_info_function} 64 | {_take_the_wheel_instructions} 65 | {_general_instructions} 66 | """ 67 | 68 | _no_wheel = f"""\ 69 | {_intro} 70 | {_general_instructions} 71 | """ 72 | 73 | 74 | def pdb_instructions(supports_flow: bool, take_the_wheel: bool) -> str: 75 | if take_the_wheel: 76 | if supports_flow: 77 | return _wheel_and_slice 78 | else: 79 | return _wheel_no_slice 80 | else: 81 | return _no_wheel 82 | -------------------------------------------------------------------------------- /src/chatdbg/custom_pdb/text.py: -------------------------------------------------------------------------------- 1 | import re 2 | from typing import Union 3 | 4 | 5 | def make_arrow(pad): 6 | """generate the leading arrow in front of traceback or debugger""" 7 | if pad >= 2: 8 | return "-" * (pad - 2) + "> " 9 | elif pad == 1: 10 | return ">" 11 | return "" 12 | 13 | 14 | def strip_color(s: str) -> str: 15 | ansi_escape = re.compile(r"\x1B(?:[@-Z\\-_]|\[[0-?]*[ -/]*[@-~])") 16 | return ansi_escape.sub("", s) 17 | 18 | 19 | def truncate_proportionally( 20 | text: str, maxlen: int = 32000, top_proportion: Union[float, int] = 0.5 21 | ) -> str: 22 | """Omit part of a string if needed to make it fit in a maximum length.""" 23 | if len(text) > maxlen: 24 | pre = max(0, int((maxlen - 3) * top_proportion)) 25 | post = max(0, maxlen - 3 - pre) 26 | return text[:pre] + "..." + text[len(text) - post :] 27 | return text 28 | -------------------------------------------------------------------------------- /src/chatdbg/native_util/clangd_lsp_integration.py: -------------------------------------------------------------------------------- 1 | import json 2 | import os 3 | import subprocess 4 | import urllib.parse 5 | 6 | import llm_utils 7 | 8 | 9 | def _to_lsp_request(id, method, params): 10 | request = {"jsonrpc": "2.0", "id": id, "method": method} 11 | if params: 12 | request["params"] = params 13 | 14 | content = json.dumps(request) 15 | header = f"Content-Length: {len(content)}\r\n\r\n" 16 | return header + content 17 | 18 | 19 | # Same as a request, but without an id. 20 | def _to_lsp_notification(method, params): 21 | request = {"jsonrpc": "2.0", "method": method} 22 | if params: 23 | request["params"] = params 24 | 25 | content = json.dumps(request) 26 | header = f"Content-Length: {len(content)}\r\n\r\n" 27 | return header + content 28 | 29 | 30 | def _parse_lsp_response(id, file): 31 | # Ignore all messages until the response with the correct id is found. 32 | while True: 33 | header = {} 34 | while True: 35 | line = file.readline().strip() 36 | if not line: 37 | break 38 | key, value = line.split(":", 1) 39 | header[key.strip()] = value.strip() 40 | 41 | content = file.read(int(header["Content-Length"])) 42 | response = json.loads(content) 43 | if "id" in response and response["id"] == id: 44 | return response 45 | 46 | 47 | def _path_to_uri(path): 48 | return "file://" + os.path.abspath(path) 49 | 50 | 51 | def uri_to_path(uri): 52 | data = urllib.parse.urlparse(uri) 53 | 54 | assert data.scheme == "file" 55 | assert not data.netloc 56 | assert not data.params 57 | assert not data.query 58 | assert not data.fragment 59 | 60 | path = data.path 61 | if path.startswith(os.getcwd()): 62 | path = os.path.relpath(path, os.getcwd()) 63 | return urllib.parse.unquote(path) # clangd seems to escape paths. 64 | 65 | 66 | def is_available(executable="clangd"): 67 | try: 68 | clangd = subprocess.run( 69 | [executable, "--version"], 70 | stdout=subprocess.DEVNULL, 71 | stderr=subprocess.DEVNULL, 72 | ) 73 | return clangd.returncode == 0 74 | except FileNotFoundError: 75 | return False 76 | 77 | 78 | class clangd: 79 | def __init__( 80 | self, 81 | executable="clangd", 82 | working_directory=os.getcwd(), 83 | stderr=subprocess.DEVNULL, 84 | ): 85 | self.id = 0 86 | self.process = subprocess.Popen( 87 | [executable], 88 | text=True, 89 | stdin=subprocess.PIPE, 90 | stdout=subprocess.PIPE, 91 | stderr=stderr, 92 | cwd=working_directory, 93 | ) 94 | self.initialize() 95 | 96 | def __del__(self): 97 | self.process.terminate() 98 | 99 | def initialize(self): 100 | self.id += 1 101 | request = _to_lsp_request(self.id, "initialize", {"processId": os.getpid()}) 102 | self.process.stdin.write(request) 103 | self.process.stdin.flush() 104 | return _parse_lsp_response(self.id, self.process.stdout) 105 | # TODO: Assert there is no error. 106 | 107 | def didOpen(self, filename, languageId): 108 | with open(filename, "r") as file: 109 | text = file.read() 110 | 111 | notification = _to_lsp_notification( 112 | "textDocument/didOpen", 113 | { 114 | "textDocument": { 115 | "uri": _path_to_uri(filename), 116 | "languageId": languageId, 117 | "version": 1, 118 | "text": text, 119 | } 120 | }, 121 | ) 122 | self.process.stdin.write(notification) 123 | self.process.stdin.flush() 124 | 125 | def didClose(self, filename): 126 | notification = _to_lsp_notification( 127 | "textDocument/didClose", {"textDocument": {"uri": _path_to_uri(filename)}} 128 | ) 129 | self.process.stdin.write(notification) 130 | self.process.stdin.flush() 131 | 132 | def definition(self, filename, line, character): 133 | self.id += 1 134 | request = _to_lsp_request( 135 | self.id, 136 | "textDocument/definition", 137 | { 138 | "textDocument": {"uri": _path_to_uri(filename)}, 139 | "position": { 140 | # Things are 0-indexed in LSP. 141 | "line": line - 1, 142 | "character": character - 1, 143 | }, 144 | }, 145 | ) 146 | self.process.stdin.write(request) 147 | self.process.stdin.flush() 148 | return _parse_lsp_response(self.id, self.process.stdout) 149 | 150 | 151 | def native_definition(command): 152 | if not is_available(): 153 | return "`clangd` was not found. The `definition` function will not be made available." 154 | last_space_index = command.rfind(" ") 155 | if last_space_index == -1: 156 | return "usage: definition : " 157 | filename_lineno = command[:last_space_index] 158 | symbol = command[last_space_index + 1 :] 159 | parts = filename_lineno.split(":") 160 | if len(parts) != 2: 161 | return "usage: definition : " 162 | filename, lineno = parts[0], int(parts[1]) 163 | 164 | try: 165 | with open(filename, "r") as file: 166 | lines = file.readlines() 167 | except FileNotFoundError: 168 | return f"file '{filename}' not found." 169 | 170 | if lineno - 1 >= len(lines): 171 | return "symbol not found at that location." 172 | 173 | # We just return the first match here. Maybe we should find all definitions. 174 | character = lines[lineno - 1].find(symbol) 175 | 176 | # Now, some heuristics to make up for GPT's terrible math skills. 177 | if character == -1: 178 | symbol = symbol.lstrip("*") 179 | character = lines[lineno - 1].find(symbol) 180 | 181 | if character == -1: 182 | symbol = symbol.split("::")[-1] 183 | character = lines[lineno - 1].find(symbol) 184 | 185 | # Check five lines above and below. 186 | if character == -1: 187 | for i in range(-5, 6, 1): 188 | if lineno - 1 + i < 0 or lineno - 1 + i >= len(lines): 189 | continue 190 | character = lines[lineno - 1 + i].find(symbol) 191 | if character != -1: 192 | lineno += i 193 | break 194 | 195 | if character == -1: 196 | return "symbol not found at that location." 197 | 198 | _clangd = None 199 | if is_available(): 200 | _clangd = clangd() 201 | 202 | _clangd.didOpen(filename, "c" if filename.endswith(".c") else "cpp") 203 | definition = _clangd.definition(filename, lineno, character + 1) 204 | _clangd.didClose(filename) 205 | 206 | if "result" not in definition or not definition["result"]: 207 | return "No definition found." 208 | 209 | path = uri_to_path(definition["result"][0]["uri"]) 210 | start_lineno = definition["result"][0]["range"]["start"]["line"] + 1 211 | end_lineno = definition["result"][0]["range"]["end"]["line"] + 1 212 | lines, first = llm_utils.read_lines(path, start_lineno - 5, end_lineno + 5) 213 | content = llm_utils.number_group_of_lines(lines, first) 214 | line_string = ( 215 | f"line {start_lineno}" 216 | if start_lineno == end_lineno 217 | else f"lines {start_lineno}-{end_lineno}" 218 | ) 219 | return f"""File '{path}' at {line_string}:\n```\n{content}\n```""" 220 | -------------------------------------------------------------------------------- /src/chatdbg/native_util/code.py: -------------------------------------------------------------------------------- 1 | import llm_utils 2 | 3 | 4 | def code(command): 5 | parts = command.split(":") 6 | if len(parts) != 2: 7 | return "usage: code :" 8 | filename, lineno = parts[0], int(parts[1]) 9 | try: 10 | lines, first = llm_utils.read_lines(filename, lineno - 7, lineno + 3) 11 | except FileNotFoundError: 12 | return f"file '{filename}' not found." 13 | return llm_utils.number_group_of_lines(lines, first) 14 | -------------------------------------------------------------------------------- /src/chatdbg/native_util/dbg_dialog.py: -------------------------------------------------------------------------------- 1 | import sys 2 | 3 | from . import clangd_lsp_integration 4 | from ..util.prompts import ( 5 | build_followup_prompt, 6 | build_initial_prompt, 7 | initial_instructions, 8 | ) 9 | 10 | from ..assistant.assistant import Assistant 11 | from ..util.config import chatdbg_config 12 | from ..util.history import CommandHistory 13 | from ..util.log import ChatDBGLog 14 | from .stacks import build_enriched_stacktrace 15 | 16 | 17 | class DBGError(Exception): 18 | 19 | def __init__(self, message): 20 | self.message = message 21 | super().__init__(self.message) 22 | 23 | 24 | class DBGDialog: 25 | # The log file used by the listener on the Assistant 26 | _log = ChatDBGLog( 27 | log_filename=chatdbg_config.log, 28 | config=chatdbg_config.to_json(), 29 | capture_streams=False, # don't have access to target's stdout/stderr here. 30 | ) 31 | 32 | def __init__(self, prompt) -> None: 33 | self._prompt = prompt 34 | self._history = CommandHistory(self._prompt) 35 | self._unsafe_cmd = False 36 | 37 | def query_and_print(self, assistant, user_text, is_followup): 38 | prompt = self.build_prompt(user_text, is_followup) 39 | 40 | self._history.clear() 41 | print(assistant.query(prompt, user_text)["message"]) 42 | if self._unsafe_cmd: 43 | self.warn( 44 | f"Warning: One or more debugger commands were blocked as potentially unsafe.\nWarning: You can disable sanitizing with `config --unsafe` and try again at your own risk." 45 | ) 46 | self._unsafe_cmd = False 47 | 48 | def dialog(self, user_text): 49 | assistant = self._make_assistant() 50 | self.check_debugger_state() 51 | 52 | self.query_and_print(assistant, user_text, False) 53 | while True: 54 | try: 55 | command = input("(ChatDBG chatting) ").strip() 56 | if command in ["exit", "quit"]: 57 | break 58 | if command in ["chat", "why"]: 59 | self.query_and_print(assistant, command, True) 60 | elif command == "history": 61 | print(self._history) 62 | else: 63 | # Send the next input as an LLDB command 64 | result = self._run_one_command(command) 65 | if self._message_is_a_bad_command_error(result): 66 | # If result is not a recognized command, pass it as a query 67 | self.query_and_print(assistant, command, True) 68 | else: 69 | if command != "test_prompt": 70 | self._history.append(command, result) 71 | print(result) 72 | except EOFError: 73 | # If it causes an error, break 74 | break 75 | 76 | assistant.close() 77 | 78 | # Return string for valid command. None if the command is not valid. 79 | def _run_one_command(self, command): 80 | pass 81 | 82 | def _message_is_a_bad_command_error(self, message): 83 | pass 84 | 85 | def check_debugger_state(self): 86 | pass 87 | 88 | def _get_frame_summaries(self, max_entries: int = 20): 89 | pass 90 | 91 | def initial_prompt_instructions(self): 92 | functions = self._supported_functions() 93 | return initial_instructions(functions) 94 | 95 | def _initial_prompt_enchriched_stack_trace(self): 96 | return build_enriched_stacktrace(self._get_frame_summaries()) 97 | 98 | def _initial_prompt_error_message(self): 99 | return None 100 | 101 | def _initial_prompt_error_details(self): 102 | """Anything more beyond the initial error message to include.""" 103 | return None 104 | 105 | def _initial_prompt_command_line(self): 106 | return None 107 | 108 | def _initial_prompt_input(self): 109 | return None 110 | 111 | def _prompt_stack(self): 112 | """ 113 | Return a simple backtrace to show the LLM where we are on the stack 114 | in followup prompts. 115 | """ 116 | return None 117 | 118 | def _prompt_history(self): 119 | return str(self._history) 120 | 121 | def build_prompt(self, arg, conversing): 122 | if not conversing: 123 | return build_initial_prompt( 124 | self._initial_prompt_enchriched_stack_trace(), 125 | self._initial_prompt_error_message(), 126 | self._initial_prompt_error_details(), 127 | self._initial_prompt_command_line(), 128 | self._initial_prompt_input(), 129 | self._prompt_history(), 130 | user_text=arg, 131 | ) 132 | else: 133 | return build_followup_prompt( 134 | self._prompt_history(), self._prompt_stack(), arg 135 | ) 136 | 137 | def llm_debug(self, command: str) -> str: 138 | pass 139 | 140 | def llm_get_code_surrounding(self, filename: str, line_number: int) -> str: 141 | """ 142 | { 143 | "name": "get_code_surrounding", 144 | "description": "The `get_code_surrounding` function returns the source code in the given file surrounding and including the provided line number.", 145 | "parameters": { 146 | "type": "object", 147 | "properties": { 148 | "filename": { 149 | "type": "string", 150 | "description": "The filename to read from." 151 | }, 152 | "line_number": { 153 | "type": "integer", 154 | "description": "The line number to focus on. Some context before and after that line will be provided." 155 | } 156 | }, 157 | "required": [ "filename", "line_number" ] 158 | } 159 | } 160 | """ 161 | return f"code {filename}:{line_number}", self._run_one_command( 162 | f"code {filename}:{line_number}" 163 | ) 164 | 165 | def llm_find_definition(self, filename: str, line_number: int, symbol: str) -> str: 166 | """ 167 | { 168 | "name": "find_definition", 169 | "description": "The `find_definition` function returns the source code for the definition for the given symbol at the given source line number. Call `find_definition` on every symbol that could be linked to the issue.", 170 | "parameters": { 171 | "type": "object", 172 | "properties": { 173 | "filename": { 174 | "type": "string", 175 | "description": "The filename the symbol is from." 176 | }, 177 | "line_number": { 178 | "type": "integer", 179 | "description": "The line number where the symbol is present." 180 | }, 181 | "symbol": { 182 | "type": "string", 183 | "description": "The symbol to lookup." 184 | } 185 | }, 186 | "required": [ "filename", "line_number", "symbol" ] 187 | } 188 | } 189 | """ 190 | return f"definition {filename}:{line_number} {symbol}", self._run_one_command( 191 | f"definition {filename}:{line_number} {symbol}" 192 | ) 193 | 194 | def _supported_functions(self): 195 | functions = [self.llm_debug, self.llm_get_code_surrounding] 196 | if clangd_lsp_integration.is_available(): 197 | functions += [self.llm_find_definition] 198 | return functions 199 | 200 | def _make_assistant(self) -> Assistant: 201 | 202 | functions = self._supported_functions() 203 | instruction_prompt = self.initial_prompt_instructions() 204 | 205 | # gdb overwrites sys.stdin to be a file object that doesn't seem 206 | # to support colors or streaming. So, just use the original stdout 207 | # here for all subclasses. 208 | printer = chatdbg_config.make_printer(sys.__stdout__, self._prompt, " ", 80) 209 | 210 | assistant = Assistant( 211 | instruction_prompt, 212 | model=chatdbg_config.model, 213 | functions=functions, 214 | listeners=[ 215 | printer, 216 | self._log, 217 | ], 218 | ) 219 | 220 | return assistant 221 | 222 | def warn(self, message): 223 | print(message) 224 | 225 | def fail(self, message): 226 | raise DBGError(message) 227 | -------------------------------------------------------------------------------- /src/chatdbg/native_util/safety.py: -------------------------------------------------------------------------------- 1 | import re 2 | 3 | 4 | # A very simple whitelist-based approach. 5 | # If ChatDBG wants to call other commands not listed here, they should be 6 | # evaluated and added if not possibly harmful. 7 | def command_is_safe(cmd: str) -> bool: 8 | cmd = cmd.strip() 9 | command_name = cmd.split()[0] 10 | 11 | # Allowed unconditionally. 12 | if command_name in [ 13 | "apropos", 14 | "bt", 15 | "down", 16 | "frame", 17 | "h", 18 | "help", 19 | "info", 20 | "language", 21 | "l", 22 | "list", 23 | "source", 24 | "up", 25 | "version", 26 | ]: 27 | return True 28 | 29 | # Allowed conditionally. 30 | if command_name in ["p", "print"]: 31 | return re.fullmatch(r"([a-zA-Z0-9_ *.]|->)*", cmd) is not None 32 | 33 | return False 34 | -------------------------------------------------------------------------------- /src/chatdbg/native_util/stacks.py: -------------------------------------------------------------------------------- 1 | import textwrap 2 | 3 | import llm_utils 4 | 5 | 6 | class _ArgumentEntry: 7 | def __init__(self, type: str, name: str, value: str): 8 | self._type = type 9 | self._name = name 10 | self._value = value 11 | 12 | def __str__(self): 13 | return f"({self._type}) {self._name} = {self._value if self._value is not None else '[unknown]'}" 14 | 15 | def __repr__(self): 16 | return f"_ArgumentEntry({repr(self._type)}, {repr(self._name)}, {repr(self._value) if self._value is not None else '[unknown]'})" 17 | 18 | 19 | class _FrameSummaryEntry: 20 | def __init__( 21 | self, 22 | index: int, 23 | name: str, 24 | arguments: list[_ArgumentEntry], 25 | file_path: str, 26 | lineno: int, 27 | ): 28 | self._index = index 29 | self._name = name 30 | self._arguments = arguments 31 | self._file_path = file_path 32 | self._lineno = lineno 33 | 34 | def index(self): 35 | return self._index 36 | 37 | def file_path(self): 38 | return self._file_path 39 | 40 | def lineno(self): 41 | return self._lineno 42 | 43 | def __str__(self): 44 | a = ", ".join([str(a) for a in self._arguments]) 45 | return f"{self._index}: {self._name}({a}) at {self._file_path}:{self._lineno}" 46 | 47 | def __repr__(self): 48 | return f"_FrameSummaryEntry({self._index}, {repr(self._name)}, {repr(self._arguments)}, {repr(self._file_path)}, {self._lineno})" 49 | 50 | 51 | class _SkippedFramesEntry: 52 | def __init__(self, count: int): 53 | self._count = count 54 | 55 | def count(self): 56 | return self._count 57 | 58 | def __str__(self): 59 | return f"[{self._count} skipped frame{'s' if self._count > 1 else ''}...]" 60 | 61 | def __repr__(self): 62 | return f"_SkippedFramesEntry({self._count})" 63 | 64 | 65 | def build_enriched_stacktrace(summaries): 66 | parts = [] 67 | if not summaries: 68 | print("could not generate any frame summary.") 69 | return 70 | else: 71 | frame_summary = "\n".join([str(s) for s in summaries]) 72 | parts.append(frame_summary) 73 | 74 | total_frames = sum( 75 | [s.count() if isinstance(s, _SkippedFramesEntry) else 1 for s in summaries] 76 | ) 77 | 78 | if total_frames > 1000: 79 | parts.append( 80 | "Note that there are over 1000 frames in the stack trace, hinting at a possible stack overflow error." 81 | ) 82 | 83 | max_initial_locations_to_send = 3 84 | source_code_entries = [] 85 | for summary in summaries: 86 | if isinstance(summary, _FrameSummaryEntry): 87 | file_path, lineno = summary.file_path(), summary.lineno() 88 | lines, first = llm_utils.read_lines(file_path, lineno - 10, lineno + 9) 89 | block = llm_utils.number_group_of_lines(lines, first) 90 | block = textwrap.indent(block, " ") 91 | source_code_entries.append( 92 | f"Frame #{summary.index()} at {file_path}:{lineno}:\n{block}\n" 93 | ) 94 | 95 | if len(source_code_entries) == max_initial_locations_to_send: 96 | break 97 | 98 | if source_code_entries: 99 | parts.append( 100 | f"Here is the source code for the first {len(source_code_entries)} frames:\n\n" 101 | + "\n\n".join(source_code_entries) 102 | ) 103 | else: 104 | print("could not retrieve source code for any frames.") 105 | return "\n\n".join(parts) 106 | -------------------------------------------------------------------------------- /src/chatdbg/pdb_util/capture.py: -------------------------------------------------------------------------------- 1 | from io import StringIO, TextIOWrapper 2 | 3 | 4 | class CaptureInput: 5 | def __init__(self, input_stream): 6 | input_stream = TextIOWrapper(input_stream.buffer, encoding="utf-8", newline="") 7 | 8 | self.original_input = input_stream 9 | self.capture_buffer = StringIO() 10 | self.original_readline = input_stream.buffer.raw.readline 11 | 12 | def custom_readline(*args, **kwargs): 13 | input_data = self.original_readline(*args, **kwargs) 14 | self.capture_buffer.write(input_data.decode()) 15 | return input_data 16 | 17 | input_stream.buffer.raw.readline = custom_readline 18 | 19 | def readline(self, *args, **kwargs): 20 | input_data = self.original_input.readline(*args, **kwargs) 21 | self.capture_buffer.write(input_data) 22 | self.capture_buffer.flush() 23 | return input_data 24 | 25 | def read(self, *args, **kwargs): 26 | input_data = self.original_input.read(*args, **kwargs) 27 | self.capture_buffer.write(input_data) 28 | self.capture_buffer.flush() 29 | return input_data 30 | 31 | def get_captured_input(self): 32 | return self.capture_buffer.getvalue() 33 | 34 | 35 | class CaptureOutput: 36 | """ 37 | File wrapper that will stash a copy of everything written. 38 | """ 39 | 40 | def __init__(self, file): 41 | self.file = file 42 | self.buffer = StringIO() 43 | 44 | def write(self, data): 45 | self.buffer.write(data) 46 | return self.file.write(data) 47 | 48 | def getvalue(self): 49 | return self.buffer.getvalue() 50 | 51 | def getfile(self): 52 | return self.file 53 | 54 | def __getattr__(self, attr): 55 | # Delegate attribute access to the file object 56 | return getattr(self.file, attr) 57 | -------------------------------------------------------------------------------- /src/chatdbg/pdb_util/locals.py: -------------------------------------------------------------------------------- 1 | import ast 2 | import inspect 3 | import itertools 4 | import numbers 5 | import textwrap 6 | 7 | import numpy as np 8 | 9 | from io import StringIO 10 | from types import FrameType 11 | from typing import Any, Union 12 | 13 | 14 | class SymbolFinder(ast.NodeVisitor): 15 | def __init__(self): 16 | self.defined_symbols = set() 17 | 18 | def visit_Assign(self, node: ast.Assign) -> None: 19 | for target in node.targets: 20 | if isinstance(target, ast.Name): 21 | self.defined_symbols.add(target.id) 22 | self.generic_visit(node) 23 | 24 | def visit_For(self, node: ast.For) -> None: 25 | if isinstance(node.target, ast.Name): 26 | self.defined_symbols.add(node.target.id) 27 | self.generic_visit(node) 28 | 29 | def visit_comprehension(self, node: ast.Name) -> None: 30 | if isinstance(node.target, ast.Name): 31 | self.defined_symbols.add(node.target.id) 32 | self.generic_visit(node) 33 | 34 | 35 | def _extract_locals(frame: FrameType) -> set[str]: 36 | try: 37 | source = textwrap.dedent(inspect.getsource(frame)) 38 | tree = ast.parse(source) 39 | 40 | finder = SymbolFinder() 41 | finder.visit(tree) 42 | 43 | args, varargs, keywords, locals = inspect.getargvalues(frame) 44 | parameter_symbols = set(args + [varargs, keywords]) 45 | parameter_symbols.discard(None) 46 | 47 | return (finder.defined_symbols | parameter_symbols) & locals.keys() 48 | except: 49 | # ipes 50 | return set() 51 | 52 | 53 | def _is_iterable(obj: Any) -> bool: 54 | try: 55 | iter(obj) 56 | return True 57 | except TypeError: 58 | return False 59 | 60 | 61 | def _repr_if_defined(obj: Any) -> bool: 62 | if obj.__class__ in [np.ndarray, dict, list, tuple]: 63 | # handle these at iterables to truncate reasonably 64 | return False 65 | result = ( 66 | "__repr__" in dir(obj.__class__) 67 | and obj.__class__.__repr__ is not object.__repr__ 68 | ) 69 | return result 70 | 71 | 72 | def _format_limited( 73 | value: Union[int, np.ndarray], limit: int = 10, depth: int = 3 74 | ) -> str: 75 | def format_tuple(t, depth): 76 | return tuple([helper(x, depth) for x in t]) 77 | 78 | def format_list(list, depth): 79 | return [helper(x, depth) for x in list] 80 | 81 | def format_dict(items, depth): 82 | return {k: helper(v, depth) for k, v in items} 83 | 84 | def format_object(obj, depth): 85 | attributes = dir(obj) 86 | fields = { 87 | attr: getattr(obj, attr, None) 88 | for attr in attributes 89 | if not callable(getattr(obj, attr, None)) and not attr.startswith("__") 90 | } 91 | return format( 92 | f"{type(obj).__name__} object with fields {format_dict(fields.items(), depth)}" 93 | ) 94 | 95 | def helper(value, depth): 96 | if depth == 0: 97 | return ... 98 | if value is Ellipsis: 99 | return ... 100 | if isinstance(value, dict): 101 | if len(value) > limit: 102 | return format_dict( 103 | list(value.items())[: limit - 1] + [(..., ...)], depth - 1 104 | ) 105 | else: 106 | return format_dict(value.items(), depth - 1) 107 | elif isinstance(value, (str, bytes)): 108 | if len(value) > 254: 109 | value = str(value)[0:253] + "..." 110 | return value 111 | elif isinstance(value, tuple): 112 | if len(value) > limit: 113 | return format_tuple(value[0 : limit - 1] + (...,), depth - 1) 114 | else: 115 | return format_tuple(value, depth - 1) 116 | elif value is None or isinstance( 117 | value, (int, float, bool, type, numbers.Number) 118 | ): 119 | return value 120 | elif isinstance(value, np.ndarray): 121 | with np.printoptions(threshold=limit): 122 | return np.array_repr(value) 123 | elif inspect.isclass(type(value)) and _repr_if_defined(value): 124 | return repr(value) 125 | elif _is_iterable(value): 126 | value = list(itertools.islice(value, 0, limit + 1)) 127 | if len(value) > limit: 128 | return format_list(value[: limit - 1] + [...], depth - 1) 129 | else: 130 | return format_list(value, depth - 1) 131 | elif inspect.isclass(type(value)): 132 | return format_object(value, depth - 1) 133 | else: 134 | return value 135 | 136 | result = str(helper(value, depth=3)).replace("Ellipsis", "...") 137 | if len(result) > 1024 * 2: 138 | result = result[: 1024 * 2 - 3] + "..." 139 | if type(value) == str: 140 | return "'" + result + "'" 141 | else: 142 | return result 143 | 144 | 145 | def print_locals(file: StringIO, frame: FrameType) -> None: 146 | locals = frame.f_locals 147 | in_global_scope = locals is frame.f_globals 148 | defined_locals = _extract_locals(frame) 149 | # Unclear benefit: possibly some benefit w/ stack only runs, but large context... 150 | # if in_global_scope and "In" in locals: # in notebook 151 | # defined_locals = defined_locals | extract_nb_globals(locals) 152 | if len(defined_locals) > 0: 153 | if in_global_scope: 154 | print(f" Global variables:", file=file) 155 | else: 156 | print(f" Variables in this frame:", file=file) 157 | for name in sorted(defined_locals): 158 | value = locals[name] 159 | t = type(value).__name__ 160 | prefix = f" {name}: {t} = " 161 | rep_list = _format_limited(value, limit=20).split("\n") 162 | if len(rep_list) > 1: 163 | rep = ( 164 | prefix 165 | + rep_list[0] 166 | + "\n" 167 | + textwrap.indent("\n".join(rep_list[1:]), prefix=" " * len(prefix)) 168 | ) 169 | else: 170 | rep = prefix + rep_list[0] 171 | print(rep, file=file) 172 | print(file=file) 173 | -------------------------------------------------------------------------------- /src/chatdbg/pdb_util/paths.py: -------------------------------------------------------------------------------- 1 | import os 2 | import sys 3 | 4 | 5 | def is_library_file(file_path): 6 | # Check if file is in standard library directory 7 | std_lib_path = os.path.dirname(os.__file__) 8 | if os.path.commonpath([file_path, std_lib_path]) == std_lib_path: 9 | return True 10 | # Check if file is in site-packages or dist-packages (common locations for installed libraries) 11 | for path in sys.path: 12 | if "site-packages" in path or "dist-packages" in path: 13 | if os.path.commonpath([file_path, path]) == path: 14 | return True 15 | # Assume the file is a user-written file if it doesn't match the above criteria 16 | return False 17 | 18 | 19 | def main(): 20 | user_path = os.path.abspath(sys.path[0]) 21 | print(f"*** user path: {user_path} ***") 22 | 23 | library_paths = [os.path.dirname(os.__file__)] + [ 24 | path for path in sys.path if "site-packages" in path or "dist-packages" in path 25 | ] 26 | 27 | print(library_paths) 28 | 29 | file_path = "/path/to/your/file.py" 30 | if is_library_file(file_path): 31 | print(f"{file_path} is likely a library file.") 32 | else: 33 | print(f"{file_path} is likely a user-written file.") 34 | 35 | 36 | if __name__ == "__main__": 37 | print(f"std lib: {os.path.dirname(os.__file__)}") 38 | for path in sys.path: 39 | if "site-packages" in path or "dist-packages" in path: 40 | print(f"packages: {path}") 41 | 42 | main() 43 | -------------------------------------------------------------------------------- /src/chatdbg/pdb_util/sandbox.py: -------------------------------------------------------------------------------- 1 | import ast 2 | import re 3 | from chatdbg.util.config import chatdbg_config 4 | 5 | 6 | def _sandboxed_call(func, *args, **kwargs): 7 | """ 8 | Check if the function is in the module whitelist before calling it. 9 | """ 10 | allowed_modules = chatdbg_config.get_module_whitelist() 11 | 12 | # Get the module name of the function. 13 | # If the module name is None, use the __name__ attribute of the globals dictionary. 14 | module_name = func.__module__ 15 | if module_name is None: 16 | module_name = func.__globals__.get("__name__", None) 17 | 18 | # Check if the function is in the module whitelist. If it is, call the function. 19 | # Otherwise, raise an ImportError. 20 | if any( 21 | re.fullmatch(allowed, f"{module_name}.{func.__name__}") 22 | for allowed in allowed_modules 23 | ): 24 | return func(*args, **kwargs) 25 | else: 26 | raise ImportError( 27 | f"Calling function {func.__name__} from module {module_name} is not allowed." 28 | ) 29 | 30 | 31 | class SandboxTransformer(ast.NodeTransformer): 32 | """ 33 | Wrap all function calls in the expression with a call to _sandboxed_call. 34 | """ 35 | 36 | def visit_Call(self, node): 37 | new_node = ast.Call( 38 | func=ast.Name(id="_sandboxed_call", ctx=ast.Load()), 39 | args=[node.func] + node.args, 40 | keywords=node.keywords, 41 | ) 42 | return ast.copy_location(new_node, node) 43 | 44 | 45 | def sandbox_eval(expression, globals, locals): 46 | """ 47 | Wrap all function calls in the expression with a call to _sandboxed_call. 48 | This function will raise an ImportError if the function is not in the module whitelist. 49 | """ 50 | tree = ast.parse(expression, mode="eval") 51 | tree = SandboxTransformer().visit(tree) 52 | ast.fix_missing_locations(tree) 53 | code = compile(tree, filename="", mode="eval") 54 | globals = globals.copy() 55 | globals["_sandboxed_call"] = _sandboxed_call 56 | return eval(code, globals, locals) 57 | -------------------------------------------------------------------------------- /src/chatdbg/util/config.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | import os 3 | 4 | from traitlets import Bool, Int, Unicode 5 | from traitlets.config import Configurable 6 | 7 | from chatdbg.assistant.listeners import BaseAssistantListener 8 | from chatdbg.util.markdown import ChatDBGMarkdownPrinter 9 | from chatdbg.util.printer import ChatDBGPrinter 10 | 11 | from io import TextIOWrapper 12 | from typing import Union 13 | 14 | from chatdbg.util.jupyter import ChatDBGJupyterPrinter 15 | 16 | 17 | def _chatdbg_get_env( 18 | option_name: str, default_value: Union[bool, int, str] 19 | ) -> Union[bool, int, str]: 20 | env_name = "CHATDBG_" + option_name.upper() 21 | v = os.getenv(env_name, str(default_value)) 22 | if type(default_value) == int: 23 | return int(v) 24 | elif type(default_value) == bool: 25 | return v.lower() == "true" or v.lower() == "1" 26 | else: 27 | return v 28 | 29 | 30 | class DBGParser(argparse.ArgumentParser): 31 | def __init__(self, **kwargs) -> None: 32 | super().__init__(**kwargs) 33 | 34 | def error(self, message): 35 | raise Exception(f"Error: {message}\n") 36 | 37 | 38 | class ChatDBGConfig(Configurable): 39 | model = Unicode(_chatdbg_get_env("model", "gpt-4o"), help="The LLM model").tag( 40 | config=True 41 | ) 42 | 43 | log = Unicode(_chatdbg_get_env("log", "log.yaml"), help="The log file").tag( 44 | config=True 45 | ) 46 | 47 | tag = Unicode(_chatdbg_get_env("tag", ""), help="Any extra info for log file").tag( 48 | config=True 49 | ) 50 | 51 | rc_lines = Unicode( 52 | _chatdbg_get_env("rc_lines", "[]"), help="lines to run at startup" 53 | ).tag(config=True) 54 | 55 | context = Int( 56 | _chatdbg_get_env("context", 10), 57 | help="lines of source code to show when displaying stacktrace information", 58 | ).tag(config=True) 59 | 60 | show_locals = Bool( 61 | _chatdbg_get_env("show_locals", True), 62 | help="show local var values in stacktrace", 63 | ).tag(config=True) 64 | 65 | show_libs = Bool( 66 | _chatdbg_get_env("show_libs", False), help="show library frames in stacktrace" 67 | ).tag(config=True) 68 | 69 | show_slices = Bool( 70 | _chatdbg_get_env("show_slices", True), help="support the `slice` command" 71 | ).tag(config=True) 72 | 73 | take_the_wheel = Bool( 74 | _chatdbg_get_env("take_the_wheel", True), help="Let LLM take the wheel" 75 | ).tag(config=True) 76 | 77 | format = Unicode( 78 | _chatdbg_get_env("format", "md"), 79 | help="The output format (text or md or md:simple or jupyter)", 80 | ).tag(config=True) 81 | 82 | instructions = Unicode( 83 | _chatdbg_get_env("instructions", ""), 84 | help="The file for the initial instructions to the LLM, or '' for the default (possibly-model specific) version", 85 | ).tag(config=True) 86 | 87 | module_whitelist = Unicode( 88 | _chatdbg_get_env("module_whitelist", ""), help="The module whitelist file" 89 | ).tag(config=True) 90 | 91 | unsafe = Bool( 92 | _chatdbg_get_env("unsafe", False), 93 | help="Disable any protections against GPT running harmful code or commands", 94 | ).tag(config=True) 95 | 96 | _user_configurable = [ 97 | log, 98 | model, 99 | instructions, 100 | format, 101 | module_whitelist, 102 | unsafe, 103 | ] 104 | 105 | def _parser(self): 106 | parser = DBGParser(add_help=False) 107 | 108 | for trait in self._user_configurable: 109 | name = f"--{trait.name}" 110 | value = self._trait_values[trait.name] 111 | t = type(value) 112 | if t == bool: 113 | parser.add_argument(name, default=value, action="store_true") 114 | else: 115 | parser.add_argument(name, default=value, type=t) 116 | 117 | return parser 118 | 119 | def to_json(self) -> dict[str, Union[int, str, bool]]: 120 | """Serialize the object to a JSON string.""" 121 | return { 122 | "model": self.model, 123 | "log": self.log, 124 | "tag": self.tag, 125 | "rc_lines": self.rc_lines, 126 | "context": self.context, 127 | "show_locals": self.show_locals, 128 | "show_libs": self.show_libs, 129 | "show_slices": self.show_slices, 130 | "take_the_wheel": self.take_the_wheel, 131 | "format": self.format, 132 | "instructions": self.instructions, 133 | "module_whitelist": self.module_whitelist, 134 | } 135 | 136 | def parse_user_flags(self, argv: list[str]) -> None: 137 | 138 | args, unknown_args = self._parser().parse_known_args(argv) 139 | 140 | for x in self._user_configurable: 141 | self.set_trait(x.name, getattr(args, x.name)) 142 | 143 | return unknown_args 144 | 145 | def user_flags_help(self) -> str: 146 | return "\n".join( 147 | [ 148 | self.class_get_trait_help(x, self).replace("ChatDBGConfig.", "") 149 | for x in self._user_configurable 150 | ] 151 | ) 152 | 153 | def user_flags(self) -> str: 154 | return "\n".join( 155 | [ 156 | f" --{x.name:10}{self._trait_values[x.name]}" 157 | for x in self._user_configurable 158 | ] 159 | ) 160 | 161 | def parse_only_user_flags(self, args: list[str]) -> str: 162 | try: 163 | unknown = chatdbg_config.parse_user_flags(args) 164 | if unknown: 165 | return ( 166 | f"Unrecognized arguments: {' '.join(unknown)}\n\n" 167 | + f"ChatDBG arguments:\n\n{self.user_flags_help()}" 168 | ) 169 | return chatdbg_config.user_flags() 170 | except Exception as e: 171 | return str(e) + f"\nChatDBG arguments:\n\n{self.user_flags_help()}" 172 | 173 | def make_printer( 174 | self, stdout: TextIOWrapper, prompt: str, prefix: str, width: int 175 | ) -> BaseAssistantListener: 176 | format = chatdbg_config.format 177 | split = format.split(":") 178 | if split[0] == "md": 179 | theme = split[1] if len(split) == 2 else None 180 | return ChatDBGMarkdownPrinter(stdout, prompt, prefix, width, theme=theme) 181 | elif format == "text": 182 | return ChatDBGPrinter(stdout, prompt, prefix, width) 183 | elif format == "jupyter": 184 | return ChatDBGJupyterPrinter(prompt, prefix, width) 185 | else: 186 | print("*** Unknown format '{format}'. Defaulting to 'text'", file=stdout) 187 | return ChatDBGPrinter(stdout, prompt, prefix, width) 188 | 189 | def get_module_whitelist(self) -> str: 190 | if self.module_whitelist == "": 191 | file_path = os.path.join(os.path.dirname(__file__), f"module_whitelist.txt") 192 | else: 193 | file_path = self.module_whitelist 194 | 195 | with open(file_path, "r") as file: 196 | return [module.rstrip() for module in file if module.rstrip() != ""] 197 | 198 | 199 | chatdbg_config: ChatDBGConfig = ChatDBGConfig() 200 | -------------------------------------------------------------------------------- /src/chatdbg/util/exit_message.py: -------------------------------------------------------------------------------- 1 | from typing import Any 2 | 3 | _chatdbg_was_called = False 4 | 5 | 6 | def chatdbg_was_called() -> None: 7 | global _chatdbg_was_called 8 | _chatdbg_was_called = True 9 | 10 | 11 | def print_exit_message(*args: Any, **kwargs: Any) -> None: 12 | global _chatdbg_was_called 13 | if _chatdbg_was_called: 14 | print("Thank you for using ChatDBG!") 15 | print( 16 | "Share your success stories here: https://github.com/plasma-umass/ChatDBG/issues/53" 17 | ) 18 | -------------------------------------------------------------------------------- /src/chatdbg/util/help.py: -------------------------------------------------------------------------------- 1 | import sys 2 | 3 | from chatdbg.util.config import chatdbg_config 4 | 5 | _usage = """\ 6 | usage: python -m ipdb [-m] [-c command] ... pyfile [arg] ... 7 | 8 | Debug the Python program given by pyfile. 9 | 10 | Initial commands are read from .pdbrc files in your home directory 11 | and in the current directory, if they exist. Commands supplied with 12 | -c are executed after commands from .pdbrc files. 13 | 14 | To let the script run until an exception occurs, use "-c continue". 15 | To let the script run up to a given line X in the debugged file, use 16 | "-c 'until X'" 17 | 18 | Option -m is available only in Python 3.7 and later. 19 | 20 | ChatDBG-specific options may appear anywhere before pyfile: 21 | """ 22 | 23 | 24 | def print_help(): 25 | print(_usage) 26 | print(chatdbg_config.user_flags_help()) 27 | sys.exit() 28 | -------------------------------------------------------------------------------- /src/chatdbg/util/history.py: -------------------------------------------------------------------------------- 1 | class CommandHistory: 2 | def __init__(self, prompt: str): 3 | self._history: list[tuple[str, str]] = [] 4 | self._prompt = prompt 5 | 6 | def append(self, command: str, result: str) -> None: 7 | self._history += [(command, result)] 8 | 9 | def clear(self) -> None: 10 | self._history = [] 11 | 12 | def _format_history_entry(self, entry: tuple[str, str]) -> str: 13 | line, output = entry 14 | if output: 15 | return f"{self._prompt}{line}\n{output}" 16 | else: 17 | return f"{self._prompt}{line}" 18 | 19 | def __str__(self) -> str: 20 | entry_strs = [self._format_history_entry(x) for x in self._history] 21 | return "\n".join(entry_strs) 22 | -------------------------------------------------------------------------------- /src/chatdbg/util/instructions/default.txt: -------------------------------------------------------------------------------- 1 | You are a debugging assistant. You will be given a stack trace for 2 | an error and answer questions related to the root cause of the 3 | error. 4 | 5 | {functions} 6 | 7 | Call any provided functions as many times as you would like. 8 | 9 | The root cause of any error is likely due to a problem in the source 10 | code from the user. 11 | 12 | Explain why each variable contributing to the error has been set 13 | to the value that it has. 14 | 15 | Continue with your explanations until you reach the root cause of 16 | the error. Your answer may be as long as necessary. 17 | 18 | End your answer with a section titled "##### Recommendation\\n" that 19 | contains one of: 20 | * a fix if you have identified the root cause 21 | * a numbered list of 1-3 suggestions for how to continue debugging if 22 | you have not 23 | -------------------------------------------------------------------------------- /src/chatdbg/util/instructions/gpt-4o.txt: -------------------------------------------------------------------------------- 1 | You are a debugging assistant. You will be given a stack trace for 2 | an error and answer questions related to the root cause of the 3 | error. 4 | 5 | {functions} 6 | 7 | Call any provided functions as many times as you would like. 8 | 9 | The root cause of any error is likely due to a problem in the source 10 | code from the user. 11 | 12 | Continue with your explanations until you reach the root cause of 13 | the error. Your answer may be as long as necessary. 14 | 15 | End your answer with a section titled "##### Recommendation\\n" that 16 | contains one of: 17 | * a fix if you have identified the root cause 18 | * a numbered list of 1-3 suggestions for how to continue debugging if 19 | you have not 20 | -------------------------------------------------------------------------------- /src/chatdbg/util/jupyter.py: -------------------------------------------------------------------------------- 1 | from io import StringIO 2 | 3 | from IPython.display import HTML, display, update_display 4 | from rich.console import Console 5 | from rich.markdown import Markdown 6 | from chatdbg.util.markdown import ChatDBGMarkdownPrinter 7 | 8 | 9 | class ChatDBGJupyterPrinter(ChatDBGMarkdownPrinter): 10 | 11 | def __init__(self, debugger_prompt, chat_prefix, width): 12 | super().__init__(StringIO(), debugger_prompt, chat_prefix, width) 13 | 14 | def _make_console(self, out): 15 | return Console( 16 | soft_wrap=False, file=out, record=True, theme=self._theme, width=self._width 17 | ) 18 | 19 | # Call backs 20 | 21 | # override to flush to the display 22 | def _print(self, text, end=""): 23 | super()._print(text, end=end) 24 | display(HTML(self._export_html())) 25 | 26 | def _export_html(self): 27 | exported_html = self._console.export_html(clear=True, inline_styles=True) 28 | custom_css = """ 29 | 40 | """ 41 | exported_html = f'
{exported_html}
' 42 | modified_html = custom_css + exported_html 43 | return modified_html 44 | 45 | def _stream_append(self, text): 46 | self._streamed += text 47 | m = self._wrap_in_panel(Markdown(self._streamed, code_theme=self._code_theme)) 48 | self._console.print(m) 49 | exported_html = self._export_html() 50 | update_display(HTML(exported_html), display_id=self._display_handle.display_id) 51 | 52 | def on_begin_stream(self): 53 | self._streamed = "" 54 | 55 | def on_stream_delta(self, text): 56 | if self._streamed == "": 57 | self._display_handle = display(HTML(""), display_id=True) 58 | self._stream_append(text) 59 | 60 | def on_end_stream(self): 61 | pass 62 | -------------------------------------------------------------------------------- /src/chatdbg/util/log.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import uuid 3 | from datetime import datetime 4 | 5 | import yaml 6 | 7 | from ..assistant.listeners import BaseAssistantListener 8 | from ..pdb_util.capture import CaptureOutput 9 | from .wrap import word_wrap_except_code_blocks 10 | 11 | 12 | class ChatDBGLog(BaseAssistantListener): 13 | 14 | def __init__(self, log_filename, config, capture_streams=True): 15 | self._log_filename = log_filename 16 | self.config = config 17 | if capture_streams: 18 | self._stdout_wrapper = CaptureOutput(sys.stdout) 19 | self._stderr_wrapper = CaptureOutput(sys.stderr) 20 | sys.stdout = self._stdout_wrapper 21 | sys.stderr = self._stdout_wrapper 22 | else: 23 | self._stderr_wrapper = None 24 | self._stderr_wrapper = None 25 | 26 | self._log = self._make_log() 27 | self._current_chat = None 28 | 29 | def _make_log(self): 30 | meta = { 31 | "time": datetime.now(), 32 | "command_line": " ".join(sys.argv), 33 | "uid": str(uuid.uuid4()), 34 | "config": self.config, 35 | } 36 | return { 37 | "steps": [], 38 | "meta": meta, 39 | "instructions": None, 40 | "stdout": ( 41 | None 42 | if self._stderr_wrapper == None 43 | else self._stdout_wrapper.getvalue() 44 | ), 45 | "stderr": ( 46 | None 47 | if self._stderr_wrapper == None 48 | else self._stderr_wrapper.getvalue() 49 | ), 50 | } 51 | 52 | def _dump(self): 53 | log = self._log 54 | 55 | def total(key): 56 | return sum( 57 | x["stats"][key] 58 | for x in log["steps"] 59 | if x["output"]["type"] == "chat" and "stats" in x["output"] 60 | ) 61 | 62 | log["meta"]["total_tokens"] = total("tokens") 63 | log["meta"]["total_time"] = total("time") 64 | log["meta"]["total_cost"] = total("cost") 65 | 66 | print(f"*** Writing ChatDBG dialog log to {self._log_filename}") 67 | 68 | with open(self._log_filename, "a") as file: 69 | 70 | def literal_presenter(dumper, data): 71 | if "\n" in data: 72 | return dumper.represent_scalar( 73 | "tag:yaml.org,2002:str", data, style="|" 74 | ) 75 | else: 76 | return dumper.represent_scalar("tag:yaml.org,2002:str", data) 77 | 78 | yaml.add_representer(str, literal_presenter) 79 | yaml.dump([log], file, default_flow_style=False, indent=2) 80 | 81 | def on_begin_dialog(self, instructions): 82 | log = self._log 83 | assert log != None 84 | log["instructions"] = instructions 85 | 86 | def on_end_dialog(self): 87 | if self._log != None: 88 | self._dump() 89 | self._log = self._make_log() 90 | 91 | def on_begin_query(self, prompt, extra): 92 | log = self._log 93 | assert log != None 94 | assert self._current_chat == None 95 | self._current_chat = { 96 | "input": extra, 97 | "prompt": prompt, 98 | "output": {"type": "chat", "outputs": []}, 99 | } 100 | 101 | def on_end_query(self, stats): 102 | log = self._log 103 | assert log != None 104 | assert self._current_chat != None 105 | log["steps"] += [self._current_chat] 106 | log["stats"] = stats 107 | self._current_chat = None 108 | 109 | def _post(self, text, kind): 110 | log = self._log 111 | assert log != None 112 | if self._current_chat != None: 113 | self._current_chat["output"]["outputs"].append( 114 | {"type": "text", "output": f"*** {kind}: {text}"} 115 | ) 116 | else: 117 | log["steps"].append( 118 | { 119 | "type": "call", 120 | "input": f"*** {kind}", 121 | "output": {"type": "text", "output": text}, 122 | } 123 | ) 124 | 125 | def on_warn(self, text): 126 | self._post(text, "Warning") 127 | 128 | def on_response(self, text): 129 | log = self._log 130 | assert log != None 131 | assert self._current_chat != None 132 | text = word_wrap_except_code_blocks(text) 133 | self._current_chat["output"]["outputs"].append({"type": "text", "output": text}) 134 | 135 | def on_function_call(self, call, result): 136 | log = self._log 137 | assert log != None 138 | if self._current_chat != None: 139 | self._current_chat["output"]["outputs"].append( 140 | { 141 | "type": "call", 142 | "input": call, 143 | "output": {"type": "text", "output": result}, 144 | } 145 | ) 146 | else: 147 | log["steps"].append( 148 | { 149 | "type": "call", 150 | "input": call, 151 | "output": {"type": "text", "output": result}, 152 | } 153 | ) 154 | -------------------------------------------------------------------------------- /src/chatdbg/util/markdown.py: -------------------------------------------------------------------------------- 1 | import shutil 2 | import textwrap 3 | 4 | from rich import box 5 | import rich 6 | from rich.console import Console 7 | from rich.live import Live 8 | from rich.markdown import Markdown 9 | from rich.panel import Panel 10 | from rich.theme import Theme 11 | from rich.table import Table 12 | from rich.markup import escape 13 | 14 | from chatdbg.util.text import fill_to_width, wrap_long_lines 15 | 16 | from ..assistant.listeners import BaseAssistantListener 17 | 18 | 19 | _themes = { 20 | "default": Theme( 21 | { 22 | "markdown.block": "black on light_steel_blue1", 23 | "markdown.paragraph": "black on light_steel_blue1", 24 | "markdown.text": "black on light_steel_blue1", 25 | "markdown.code": "blue", 26 | "markdown.code_block": "blue", 27 | "markdown.item.bullet": "bold blue", 28 | "markdown.item.number": "bold blue", 29 | "markdown.h1": "bold black", 30 | "markdown.h2": "bold black", 31 | "markdown.h3": "bold black", 32 | "markdown.h4": "bold black", 33 | "markdown.h5": "bold black", 34 | "command": "bold gray11 on wheat1", 35 | "result": "grey35 on wheat1", 36 | "warning": "bright_white on green", 37 | "error": "bright_white on red", 38 | } 39 | ), 40 | "basic": Theme( 41 | { 42 | "markdown.block": "bright_blue on bright_white", 43 | "markdown.paragraph": "bright_blue on bright_white", 44 | "markdown.text": "bright_blue on bright_white", 45 | "markdown.code": "cyan", 46 | "markdown.code_block": "bright_blue", 47 | "markdown.item.bullet": "bold cyan", 48 | "markdown.item.number": "bold cyan", 49 | "markdown.h1": "bold black", 50 | "markdown.h2": "bold black", 51 | "markdown.h3": "bold black", 52 | "markdown.h4": "bold black", 53 | "markdown.h5": "bold bright_blue on bright_white", 54 | "command": "bold bright_yellow on white", 55 | "result": "yellow on white", 56 | "warning": "bright_white on green", 57 | "error": "bright_white on red", 58 | } 59 | ), 60 | } 61 | 62 | _simple_box = Box = box.Box( 63 | " \n" " \n" " \n" " \n" " \n" " \n" " \n" " \n", ascii=True 64 | ) 65 | 66 | from rich.markdown import ConsoleOptions, loop_first, RenderResult, Segment 67 | 68 | 69 | class MyListItem(rich.markdown.ListItem): 70 | """An item in a list.""" 71 | 72 | style_name = "markdown.item" 73 | 74 | def __init__(self) -> None: 75 | super().__init__() 76 | 77 | def render_bullet(self, console: Console, options: ConsoleOptions) -> RenderResult: 78 | render_options = options.update(width=options.max_width - 3) 79 | lines = console.render_lines(self.elements, render_options, style=self.style) 80 | bullet_style = console.get_style("markdown.item.bullet", default="none") 81 | 82 | bullet = Segment(" * ", bullet_style) 83 | padding = Segment(" " * 3, bullet_style) 84 | new_line = Segment("\n") 85 | for first, line in loop_first(lines): 86 | yield bullet if first else padding 87 | yield from line 88 | yield new_line 89 | 90 | 91 | class ChatDBGMarkdownPrinter(BaseAssistantListener): 92 | 93 | def __init__(self, out, debugger_prompt, chat_prefix, width, theme=None): 94 | self._out = out 95 | self._debugger_prompt = debugger_prompt 96 | self._chat_prefix = chat_prefix 97 | self._left_indent = 4 98 | self._width = shutil.get_terminal_size(fallback=(width, 24)).columns 99 | self._theme = _themes["default"] if theme == None else _themes[theme] 100 | self._code_theme = "default" 101 | # used to keep track of streaming 102 | self._streamed = "" 103 | 104 | self._console = self._make_console(out) 105 | 106 | if theme == "basic": 107 | Markdown.elements["list_item_open"] = MyListItem 108 | self._code_theme = "monokai" 109 | 110 | def _make_console(self, out): 111 | return Console(soft_wrap=False, file=out, theme=self._theme, width=self._width) 112 | 113 | # Call backs 114 | 115 | def on_begin_query(self, prompt, user_text): 116 | pass 117 | 118 | def on_end_query(self, stats): 119 | pass 120 | 121 | def _print(self, renderable, end=""): 122 | self._console.print(renderable, end=end) 123 | 124 | def _wrap_in_panel(self, rich_element): 125 | 126 | left_panel = Panel("", box=_simple_box, style="on default") 127 | right_panel = Panel( 128 | rich_element, 129 | box=_simple_box, 130 | style=self._console.get_style("markdown.block"), 131 | ) 132 | 133 | # Create a table to hold the panels side by side 134 | table = Table.grid(padding=0) 135 | table.add_column(justify="left", width=self._left_indent - 2) 136 | table.add_column(justify="left") 137 | table.add_row(left_panel, right_panel) 138 | return table 139 | 140 | def _message(self, text, style): 141 | self._print( 142 | self._wrap_in_panel(self._wrap_and_fill_and_indent(text, " *** ", style)) 143 | ) 144 | 145 | def on_warn(self, text): 146 | self._message(text, "warning") 147 | 148 | def on_error(self, text): 149 | self._message(text, "error") 150 | 151 | def _stream_append(self, text): 152 | self._streamed += text 153 | m = self._wrap_in_panel(Markdown(self._streamed, code_theme=self._code_theme)) 154 | self._live.update(m) 155 | 156 | def on_begin_stream(self): 157 | self._streamed = "" 158 | 159 | def on_stream_delta(self, text): 160 | if self._streamed == "": 161 | self._live = Live(vertical_overflow="visible", console=self._console) 162 | self._live.start(True) 163 | self._stream_append(text) 164 | 165 | def on_end_stream(self): 166 | if self._streamed != "": 167 | self._live.stop() 168 | 169 | def on_response(self, text): 170 | if self._streamed == "" and text != None: 171 | m = self._wrap_in_panel(Markdown(text, code_theme=self._code_theme)) 172 | self._print(m, end="\n") 173 | self._streamed = "" 174 | 175 | def on_function_call(self, call, result): 176 | prefix = self._chat_prefix 177 | line = fill_to_width(f"\n{prefix}{self._debugger_prompt}{call}", self._width) 178 | entry = f"[command]{escape(line)}[/]\n" 179 | 180 | entry += self._wrap_and_fill_and_indent( 181 | result.rstrip() + "\n", prefix, "result" 182 | ) 183 | m = self._wrap_in_panel(entry) 184 | self._print(m, end="") 185 | 186 | def _wrap_and_fill_and_indent(self, text, prefix, style_name): 187 | line_width = self._width - len(prefix) - self._left_indent - 2 188 | text = wrap_long_lines(text.expandtabs(), line_width, subsequent_indent=" ") 189 | text = fill_to_width(text, line_width) 190 | text = textwrap.indent(text, prefix, lambda _: True) 191 | text = escape(text) 192 | return f"[{style_name}]{text}[/]" 193 | -------------------------------------------------------------------------------- /src/chatdbg/util/module_whitelist.txt: -------------------------------------------------------------------------------- 1 | __main__ 2 | 3 | pandas.* 4 | numpy.* 5 | scipy.* 6 | datascience.* 7 | cs104.* 8 | 9 | builtins.abs 10 | builtins.aiter 11 | builtins.all 12 | builtins.anext 13 | builtins.any 14 | builtins.ascii 15 | builtins.bin 16 | builtins.bool 17 | builtins.bytearray 18 | builtins.bytes 19 | builtins.callable 20 | builtins.chr 21 | builtins.classmethod 22 | builtins.complex 23 | builtins.delattr 24 | builtins.dict 25 | builtins.dir 26 | builtins.divmod 27 | builtins.enumerate 28 | builtins.filter 29 | builtins.float 30 | builtins.format 31 | builtins.frozenset 32 | builtins.getattr 33 | builtins.globals 34 | builtins.hasattr 35 | builtins.hash 36 | builtins.help 37 | builtins.hex 38 | builtins.id 39 | builtins.input 40 | builtins.int 41 | builtins.isinstance 42 | builtins.issubclass 43 | builtins.iter 44 | builtins.len 45 | builtins.list 46 | builtins.locals 47 | builtins.map 48 | builtins.max 49 | builtins.memoryview 50 | builtins.min 51 | builtins.next 52 | builtins.object 53 | builtins.oct 54 | builtins.open 55 | builtins.ord 56 | builtins.pow 57 | builtins.print 58 | builtins.property 59 | builtins.range 60 | builtins.repr 61 | builtins.reversed 62 | builtins.round 63 | builtins.set 64 | builtins.setattr 65 | builtins.slice 66 | builtins.sorted 67 | builtins.staticmethod 68 | builtins.str 69 | builtins.sum 70 | builtins.super 71 | builtins.tuple 72 | builtins.type 73 | builtins.vars 74 | builtins.zip 75 | -------------------------------------------------------------------------------- /src/chatdbg/util/plog.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | import sys 3 | import textwrap 4 | 5 | import yaml 6 | 7 | 8 | class LogPrinter: 9 | def __init__(self, file): 10 | self.file = file 11 | 12 | def print(self, x=""): 13 | print(x, file=self.file) 14 | 15 | def _do_message(self, x): 16 | text = x["output"].strip() 17 | if len(text) > 0: 18 | self.print(textwrap.indent(text, prefix=" ")) 19 | 20 | def _do_function(self, x): 21 | prompt = x["input"].strip() 22 | assert x["output"]["type"] == "text" 23 | body = x["output"]["output"].rstrip() 24 | if len(body) > 0: 25 | self.print(f" (ChatDBG) {prompt}") 26 | self.print(textwrap.indent(body, prefix=" ")) 27 | 28 | def _do_assistant(self, x): 29 | for output in x["outputs"]: 30 | if output["type"] == "call": 31 | self._do_function(output) 32 | else: 33 | self._do_message(output) 34 | self.print() 35 | 36 | def _do_step(self, x): 37 | prompt = x["input"].strip() 38 | 39 | self.print(f"(ChatDBG) {prompt}") 40 | 41 | output = x["output"] 42 | if output["type"] == "text": 43 | body = output["output"].rstrip() 44 | if len(body) > 0: 45 | self.print(textwrap.indent(body, prefix=" ")) 46 | else: 47 | self._do_assistant(output) 48 | 49 | def do_one(self, x): 50 | for step in x["steps"]: 51 | self.print() 52 | self._do_step(step) 53 | self.print() 54 | 55 | 56 | def main(): 57 | parser = argparse.ArgumentParser(description="ChatDBG log printer") 58 | parser.add_argument("filenames", nargs="*", help="log files to print") 59 | 60 | args = parser.parse_args() 61 | 62 | for file in args.filenames: 63 | with open(file, "r") as log: 64 | full = yaml.safe_load(log) 65 | 66 | for i, x in enumerate(full): 67 | print() 68 | print(f"{i} " + ("-" * 78)) 69 | print(x["instructions"], file=sys.stdout) 70 | print("-" * 80) 71 | LogPrinter(sys.stdout).do_one(x) 72 | print() 73 | print() 74 | 75 | 76 | if __name__ == "__main__": 77 | main() 78 | -------------------------------------------------------------------------------- /src/chatdbg/util/printer.py: -------------------------------------------------------------------------------- 1 | import textwrap 2 | from ..assistant.listeners import BaseAssistantListener 3 | from .stream import StreamingTextWrapper 4 | from .wrap import word_wrap_except_code_blocks 5 | import os 6 | 7 | 8 | class ChatDBGPrinter(BaseAssistantListener): 9 | def __init__(self, out, debugger_prompt, chat_prefix, width): 10 | self._out = out 11 | self._debugger_prompt = debugger_prompt 12 | self._chat_prefix = chat_prefix 13 | try: 14 | self._width = min(width, os.get_terminal_size().columns - len(chat_prefix)) 15 | except: 16 | # get_terminal_size() may fail in notebooks 17 | self._width = width 18 | 19 | # used to keep track of streaming 20 | self._at_start = True 21 | 22 | # Call backs 23 | 24 | def on_begin_query(self, prompt, user_text): 25 | pass 26 | 27 | def on_end_query(self, stats): 28 | pass 29 | 30 | def _print(self, text, **kwargs): 31 | print( 32 | textwrap.indent(text, self._chat_prefix, lambda _: True), 33 | file=self._out, 34 | **kwargs, 35 | ) 36 | 37 | def on_warn(self, text): 38 | print(textwrap.indent(text, "*** "), file=self._out) 39 | 40 | def on_error(self, text): 41 | print(textwrap.indent(text, "*** "), file=self._out) 42 | 43 | def on_begin_stream(self): 44 | self._stream_wrapper = StreamingTextWrapper( 45 | self._chat_prefix, width=self._width 46 | ) 47 | self._at_start = True 48 | 49 | def on_stream_delta(self, text): 50 | if self._at_start: 51 | self._at_start = False 52 | print( 53 | self._stream_wrapper.append("\n(Message) ", False), 54 | end="", 55 | flush=True, 56 | file=self._out, 57 | ) 58 | print( 59 | self._stream_wrapper.append(text, False), end="", flush=True, file=self._out 60 | ) 61 | 62 | def on_end_stream(self): 63 | print(self._stream_wrapper.flush(), end="", flush=True, file=self._out) 64 | self._at_start = True 65 | 66 | def on_response(self, text): 67 | if self._at_start and text != None: 68 | text = "(Message) " + text 69 | text = word_wrap_except_code_blocks( 70 | text, self._width - len(self._chat_prefix) 71 | ) 72 | self._print(text) 73 | 74 | def on_function_call(self, call, result): 75 | if result and len(result) > 0: 76 | entry = f"{self._debugger_prompt}{call}\n{result}" 77 | else: 78 | entry = f"{self._debugger_prompt}{call}" 79 | self._print(entry) 80 | -------------------------------------------------------------------------------- /src/chatdbg/util/prompts.py: -------------------------------------------------------------------------------- 1 | import json 2 | import os 3 | from chatdbg.util.config import chatdbg_config 4 | from .text import truncate_proportionally 5 | from typing import Any, Callable, List 6 | 7 | 8 | def _wrap_it(before: str, text: str, after: str = "", maxlen: int = 2048) -> str: 9 | if text: 10 | text = truncate_proportionally(text, maxlen, 0.5) 11 | before = before + ":\n" if before else "" 12 | after = after + "\n" if after else "" 13 | return f"{before}```\n{text}\n```\n{after}" 14 | else: 15 | return "" 16 | 17 | 18 | def _concat_prompt(*args) -> str: 19 | args = [a for a in args if len(a) > 0] 20 | return "\n".join(args) 21 | 22 | 23 | def _user_text_it(user_text: str) -> str: 24 | return user_text if len(user_text) > 0 else "What's the bug? Give me a fix." 25 | 26 | 27 | def build_initial_prompt( 28 | stack: str, 29 | error: str, 30 | details: str, 31 | command_line: str, 32 | inputs: str, 33 | history: str, 34 | extra: str = "", 35 | user_text: str = "", 36 | ) -> str: 37 | return _concat_prompt( 38 | _wrap_it("The program has this stack trace", stack), 39 | _wrap_it("The program encountered the following error", error, details), 40 | _wrap_it("This was the command line", command_line), 41 | _wrap_it("This was the program's input", inputs), 42 | _wrap_it("This is the history of some debugger commands I ran", history), 43 | _wrap_it("", extra), 44 | _user_text_it(user_text), 45 | ) 46 | 47 | 48 | def build_followup_prompt(history: str, extra: str, user_text: str) -> str: 49 | return _concat_prompt( 50 | _wrap_it("This is the history of some debugger commands I ran", history), 51 | _wrap_it("", extra), 52 | _user_text_it(user_text), 53 | ) 54 | 55 | 56 | def initial_instructions(functions: List[Callable[[Any], Any]]) -> str: 57 | if chatdbg_config.instructions == "": 58 | file_path = os.path.join( 59 | os.path.dirname(__file__), f"instructions/{chatdbg_config.model}.txt" 60 | ) 61 | if not os.path.exists(file_path): 62 | file_path = os.path.join( 63 | os.path.dirname(__file__), f"instructions/default.txt" 64 | ) 65 | else: 66 | file_path = chatdbg_config.instructions 67 | 68 | function_instructions = [json.loads(f.__doc__)["description"] for f in functions] 69 | with open(file_path, "r") as file: 70 | template = file.read() 71 | return template.format_map({"functions": "\n\n".join(function_instructions)}) 72 | -------------------------------------------------------------------------------- /src/chatdbg/util/stream.py: -------------------------------------------------------------------------------- 1 | import textwrap 2 | import re 3 | import sys 4 | from .wrap import word_wrap_except_code_blocks 5 | 6 | 7 | class StreamingTextWrapper: 8 | 9 | def __init__(self, indent=" ", width=80): 10 | self._buffer = "" # the raw text so far 11 | self._wrapped = "" # the successfully wrapped text do far 12 | self._pending = ( 13 | "" # the part after the last space in buffer -- has not been wrapped yet 14 | ) 15 | self._indent = indent 16 | self._width = width - len(indent) 17 | 18 | def append(self, text, flush=False): 19 | if flush: 20 | self._buffer += self._pending + text 21 | self._pending = "" 22 | else: 23 | text_bits = re.split(r"(\s+)", self._pending + text) 24 | self._pending = text_bits[-1] 25 | self._buffer += "".join(text_bits[0:-1]) 26 | 27 | wrapped = word_wrap_except_code_blocks(self._buffer, width=self._width) 28 | wrapped = textwrap.indent(wrapped, self._indent, lambda _: True) 29 | wrapped_delta = wrapped[len(self._wrapped) :] 30 | self._wrapped = wrapped 31 | return wrapped_delta 32 | 33 | def flush(self): 34 | if len(self._buffer) > 0: 35 | result = self.append("\n", flush=True) 36 | else: 37 | result = self.append("", flush=True) 38 | self._buffer = "" 39 | self._wrapped = "" 40 | return result 41 | 42 | 43 | if __name__ == "__main__": 44 | s = StreamingTextWrapper(3, 20) 45 | for x in sys.argv[1:]: 46 | y = s.append(" " + x) 47 | print(y, end="", flush=True) 48 | print(s.flush()) 49 | -------------------------------------------------------------------------------- /src/chatdbg/util/text.py: -------------------------------------------------------------------------------- 1 | import re 2 | import textwrap 3 | from typing import Union 4 | 5 | 6 | def make_arrow(pad): 7 | """generate the leading arrow in front of traceback or debugger""" 8 | if pad >= 2: 9 | return "-" * (pad - 2) + "> " 10 | elif pad == 1: 11 | return ">" 12 | return "" 13 | 14 | 15 | def strip_ansi(s: str) -> str: 16 | ansi_escape = re.compile(r"\x1B(?:[@-Z\\-_]|\[[0-?]*[ -/]*[@-~])") 17 | return ansi_escape.sub("", s) 18 | 19 | 20 | def truncate_proportionally( 21 | text: str, maxlen: int = 32000, top_proportion: Union[float, int] = 0.5 22 | ) -> str: 23 | """Omit part of a string if needed to make it fit in a maximum length.""" 24 | if len(text) > maxlen: 25 | pre = max(0, int((maxlen - 5) * top_proportion)) 26 | post = max(0, maxlen - 5 - pre) 27 | return text[:pre] + "[...]" + text[len(text) - post :] 28 | return text 29 | 30 | 31 | def wrap_long_lines(text: str, width: int = 80, subsequent_indent: str = " ") -> str: 32 | wrapped_lines = [] 33 | for line in text.split("\n"): 34 | if len(line) > width: 35 | wrapped_lines.extend( 36 | textwrap.wrap(line, width, subsequent_indent=subsequent_indent) 37 | ) 38 | else: 39 | wrapped_lines.append(line) 40 | return "\n".join(wrapped_lines) 41 | 42 | 43 | def fill_to_width(text: str, width: int = 80) -> str: 44 | return "\n".join([line.ljust(width) for line in text.split("\n")]) 45 | -------------------------------------------------------------------------------- /src/chatdbg/util/trim.py: -------------------------------------------------------------------------------- 1 | import copy 2 | import warnings 3 | 4 | with warnings.catch_warnings(): 5 | warnings.simplefilter("ignore") 6 | import litellm 7 | 8 | 9 | def sandwich_tokens( 10 | text: str, model: str, max_tokens: int = 1024, top_proportion: float = 0.5 11 | ) -> str: 12 | if max_tokens == None: 13 | return text 14 | tokens = litellm.encode(model, text) 15 | if len(tokens) <= max_tokens: 16 | return text 17 | else: 18 | total_len = max_tokens - 5 # some slop for the ... 19 | top_len = int(top_proportion * total_len) 20 | bot_start = len(tokens) - (total_len - top_len) 21 | return ( 22 | litellm.decode(model, tokens[0:top_len]) 23 | + " [...] " 24 | + litellm.decode(model, tokens[bot_start:]) 25 | ) 26 | 27 | 28 | def _sum_messages(messages, model): 29 | return litellm.token_counter(model, messages=messages) 30 | 31 | 32 | def _sum_kept_chunks(chunks, model): 33 | return sum(_sum_messages(messages, model) for (messages, kept) in chunks if kept) 34 | 35 | 36 | def _extract(messages, model, tool_call_ids): 37 | tools = [] 38 | other = [] 39 | for m in messages: 40 | if m.get("tool_call_id", -1) in tool_call_ids: 41 | content = sandwich_tokens(m["content"], model, 512, 1.0) 42 | # print(len(litellm.encode(model, m['content'])), '->', len(litellm.encode(model, content))) 43 | m["content"] = content 44 | tools += [m] 45 | else: 46 | other += [m] 47 | return tools, other 48 | 49 | 50 | def _chunkify(messages, model): 51 | if not messages: 52 | return [] 53 | m = messages[0] 54 | if "tool_calls" not in m: 55 | m["content"] = sandwich_tokens(m["content"], model, 1024, 0) 56 | return [([m], False)] + _chunkify(messages[1:], model) 57 | else: 58 | ids = [tool_call["id"] for tool_call in m["tool_calls"]] 59 | tools, other = _extract(messages[1:], model, ids) 60 | return [([m] + tools, False)] + _chunkify(other, model) 61 | 62 | 63 | def trim_messages( 64 | messages: list[dict[str, str]], # list of JSON objects encoded as dicts 65 | model: str, 66 | trim_ratio: float = 0.75, 67 | ) -> list: 68 | """ 69 | Strategy: 70 | - chunk messages: 71 | - single message, or 72 | - tool request and all the tool responses 73 | - keep the system messages 74 | - keep the first user message 75 | - go most recent to oldest, keeping chunks until we are at the limit 76 | 77 | Also, shorten tool call results along the way. 78 | -""" 79 | 80 | messages = copy.deepcopy(messages) 81 | 82 | max_tokens_for_model = litellm.model_cost[model]["max_input_tokens"] 83 | max_tokens = int(max_tokens_for_model * trim_ratio) 84 | 85 | if litellm.token_counter(model, messages=messages) < max_tokens: 86 | return messages 87 | 88 | chunks = _chunkify(messages=messages, model=model) 89 | # print("0", sum_all_chunks(chunks, model), max_tokens) 90 | 91 | # 1. System messages 92 | chunks = [(m, b or m[0]["role"] == "system") for (m, b) in chunks] 93 | # print("1", sum_kept_chunks(chunks, model)) 94 | 95 | # 2. First User Message 96 | for i in range(len(chunks)): 97 | messages, kept = chunks[i] 98 | if messages[0]["role"] == "user": 99 | chunks[i] = (messages, True) 100 | # print("2", sum_kept_chunks(chunks, model)) 101 | 102 | # 3. Fill it up 103 | for i in range(len(chunks))[::-1]: 104 | messages, kept = chunks[i] 105 | if kept: 106 | # print('+') 107 | continue 108 | elif ( 109 | _sum_kept_chunks(chunks, model) + _sum_messages(messages, model) 110 | < max_tokens 111 | ): 112 | # print('-', len(messages)) 113 | chunks[i] = (messages, True) 114 | else: 115 | # print("N", sum_kept_chunks(chunks, model), sum_messages(messages, model)) 116 | break 117 | 118 | # print("3", sum_kept_chunks(chunks, model)) 119 | 120 | assert ( 121 | _sum_kept_chunks(chunks, model) < max_tokens 122 | ), f"New conversation too big {_sum_kept_chunks(chunks, model)} vs {max_tokens}!" 123 | 124 | return [m for (messages, kept) in chunks if kept for m in messages] 125 | -------------------------------------------------------------------------------- /src/chatdbg/util/wrap.py: -------------------------------------------------------------------------------- 1 | import textwrap 2 | 3 | 4 | def word_wrap_except_code_blocks(text: str, width: int = 80) -> str: 5 | """ 6 | Wraps text except for code blocks for nice terminal formatting. 7 | 8 | Splits the text into paragraphs and wraps each paragraph, 9 | except for paragraphs that are inside of code blocks denoted 10 | by ` ``` `. Returns the updated text. 11 | 12 | Args: 13 | text (str): The text to wrap. 14 | width (int): The width of the lines to wrap at, passed to `textwrap.fill`. 15 | 16 | Returns: 17 | The wrapped text. 18 | """ 19 | blocks = text.split("```") 20 | for i in range(0, len(blocks), 2): 21 | paras = blocks[i].split("\n") 22 | wrapped = [textwrap.fill(para, width=width) for para in paras] 23 | blocks[i] = "\n".join(wrapped) 24 | 25 | return "```".join(blocks) 26 | --------------------------------------------------------------------------------