docs(cmd): clarify Git.execute() string vs list command argument#2144
Merged
Byron merged 1 commit intogitpython-developers:mainfrom May 7, 2026
Merged
Conversation
Closes gitpython-developers#2016 Users routinely hit GitCommandNotFound by passing a single string with spaces to repo.git.execute(...). With shell=False (default) subprocess treats the entire string as the executable name and fails. Document the recommended list form, the string-as-single-executable behavior, and the two ways to coerce a string into argv tokens (shlex.split or shell=True).
EliahKagan
added a commit
to EliahKagan/GitPython
that referenced
this pull request
May 8, 2026
Update the rewritten :param command: block from gitpython-developers#2144 in two ways: * Add a Windows bullet. gitpython-developers#2144 said the string is "passed as a single executable name to subprocess.Popen" with shell=False. That is accurate on POSIX, but on Windows subprocess.Popen forwards the string to CreateProcessW and Windows command-line parsing produces the program's argv. So e.g. "git version" actually runs. * Name the tokenization risks specifically. gitpython-developers#2144 hedged with "possible security implications" for shlex.split and pointed at the existing shell-parameter warning for shell=True. Be concrete: under shell=True the shell interprets ;, |, &, $(...), etc. as syntax, so metacharacters in interpolated values can execute arbitrary commands; shlex.split is preferable on POSIX but follows POSIX rules on Windows that may diverge from Windows command-line conventions; and embedded whitespace or quotes can shift tokenization either way. Neither is safe with untrusted input (branch names, URLs, filenames, etc.); the sequence form is, because each interpolated value occupies a single argv slot. Documentation only; behavior is unchanged. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EliahKagan
added a commit
to EliahKagan/GitPython
that referenced
this pull request
May 8, 2026
Rework :param command: as three parts: 1. Brief description: parameter type, sequence recommended, brief platform-dependent string note. Corrects gitpython-developers#2144's claim that with shell=False a string is "passed as a single executable name to subprocess.Popen" -- accurate on POSIX, but on Windows subprocess.Popen forwards the string to CreateProcessW, which tokenizes via Windows command-line parsing. 2. Asymmetric security paragraphs: * shell=True (or Git.USE_SHELL) runs the command through the shell, which interprets metacharacters anywhere in it; with untrusted input that is OS command injection. Cross-references USE_SHELL and the shell parameter for detail. * shlex.split runs no shell, but tokenizes by POSIX shell rules. On Windows those rules differ from both shell=False's OS argv parsing and shell=True's cmd.exe parsing, so untrusted whitespace or quoting can shift token boundaries and inject extra arguments into git's own option parser. 3. Conclusion: neither automatic-splitting approach is safe with untrusted input; build the sequence form directly, one value per argv slot. Replaces gitpython-developers#2144's hedged "possible security implications" wording with named mechanisms and keeps the asymmetry between command injection (shell=True, catastrophic) and argument injection (shlex.split on Windows, milder) visible. No worked examples to keep the docstring compact; the existing USE_SHELL and shell- parameter docstrings give the full picture for shell=True. Documentation only; behavior is unchanged. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EliahKagan
added a commit
to EliahKagan/GitPython
that referenced
this pull request
May 8, 2026
Rework :param command: as four parts: 1. Brief parameter description, with a recommendation to pass a sequence and a platform-dependent note on string handling: on POSIX a string is the program name, on Windows the OS splits it into argv. Corrects gitpython-developers#2144's claim that with shell=False the string is "passed as a single executable name to subprocess.Popen" -- accurate on POSIX, but on Windows subprocess.Popen forwards the string to CreateProcessW, which tokenizes via Windows command-line parsing. 2. shell=True (or Git.USE_SHELL) explanation: it sends the command to the platform shell rather than executing it directly, and the shell interprets ;, |, &, $(...), etc. as syntax. With untrusted text in the command -- paths, branch names, URLs, etc. -- this is arbitrary OS command execution. Cross-references Git.USE_SHELL for the long-form discussion. 3. shlex.split explanation: runs no shell, so the command-injection risk does not apply, but its POSIX shell rules on Windows match neither the shell=False OS argv parsing nor the shell=True cmd.exe parsing. Untrusted whitespace or quoting can therefore shift token boundaries, injecting extra arguments into git's option parser. 4. Asymmetric conclusion: build the sequence form directly; shell=True is the more dangerous route (arbitrary command execution), but no automatic-splitting route is safe with untrusted input. Replaces gitpython-developers#2144's hedged "possible security implications" wording with named mechanisms; preserves the asymmetry between command injection (shell=True) and argument injection (shlex.split on Windows). No worked examples (verbosity); the existing USE_SHELL docstring carries the full attack discussion. Documentation only; behavior is unchanged. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
Rewrites the
:param command:docstring onGit.execute()ingit/cmd.pyto clarify the string-vs-sequence semantics. The previous wording ("the program to execute is the first item in the args sequence or string") was misleading: withshell=False(the default),subprocess.Popentreats the entire string as a single executable name, so"git log -n 1"looks for an executable literally named "git log -n 1" and raisesGitCommandNotFound.The new docstring:
["git", "log", "-n", "1"]) for the defaultshell=Falsecase.shell=False, with the exact failure mode users see in [Bug] GitCommandNotFound when executing repo.git.execute on macOS #2016.shlex.split(...)(sequence) orshell=True(with the existing warning) for the case where they want a string that gets tokenised.Closes #2016
Why
This is the third recurrence of this confusion that surfaces in issues — string commands look like the obvious shape for "run this git invocation", and the failure mode is opaque. The
subprocess.Popensemantics are doing exactly what they always do; the GitPython docstring was the surface that misled users into expecting otherwise. Updating it costs ~10 lines and short-circuits the next round of issues.I deliberately kept this docs-only. Auto-splitting a string when it contains spaces would be a behavior change that could break existing callers who do pass a single executable path with whitespace. The acknowledged label on the issue suggests maintainer interest without committing to a specific behavior fix; clarifying the docs is the smallest useful step.
Verification
python -c "import git; help(git.cmd.Git.execute)"renders the new param block cleanly.python -m py_compile git/cmd.pyis clean. No tests were modified or broken.