Git3 Compat + gix speedup for GitPython - #2274

PR #2274's Python 3.12 CI jobs spent 57m 25s in Windows `pytest` versus 5m 2s
on Ubuntu, with 85,777 and 82,683 `Git.execute` launches respectively. A local
profile of the file-handle and recursive submodule regressions spent about 65 of
77 seconds executing Git, including 44 seconds in repository construction.
Repeated process startup makes the separate metadata probes expensive.

Query reference format, object format, bare status and the absolute common
directory in one `git rev-parse` call on the CLI construction path. Keep the
path last and split only the three fixed scalar fields so embedded newlines
remain intact. Git still computes every value. Supported Gix discovery keeps
its native metadata and existing CLI format-validation query, with unchanged
native launch ceilings and fallbacks.

The focused selection drops from 2,447 to 2,009 launches: three saved for
each of 146 repository opens. Three paired Windows measurements of 20 opens
each reduce the median from 300.74 ms to 221.18 ms per open, with matching
metadata and 11 versus eight launches. Add `--durations=30` to CI so future
slow tests are visible, and record the profiling evidence and reproduction
requirements in `doc/gix-backend.md`.

Cover both object formats, both reference formats, bare and non-bare repos,
spaces and embedded newlines. The eight applicable Windows cases fail before
the fix and pass afterward; all 16 cases pass on Ubuntu WSL. The optional
Windows Gix selection reproduces the same 17 failures on the original and
patched sources. Document the demonstrated path-formatting and undecodable
`commondir` differences under `GIX-14`, distinguishing them from Windows test
assumptions. Keep these bugs open pending compatible upstream behavior.

Validation: the full Windows CLI suite passes with 1,662 passed, 60 skipped,
nine expected failures, two unexpected passes and 40 passing subtests. It
records 77,732 launches in 3,715.16s. The initial baseline records 86,032 in
3,990.28s; its only two failures pass on unchanged code after restoring CI's
`core.autocrlf=true` and placing `pytest`'s temporary directory outside a Git
checkout. The overlapping full runs are not a controlled elapsed-time
comparison. Repository-wide Ruff lint/format, all pinned pre-commit hooks,
`mypy` (47 files), `basedpyright` (zero errors or warnings), the Sphinx HTML
build and `git diff --check` pass.

PR #2274's Python 3.12 CI jobs spent 57m 25s in Windows `pytest` versus 5m 2s
on Ubuntu, with 85,777 and 82,683 `Git.execute` launches respectively. A local
profile of the file-handle and recursive submodule regressions spent about 65 of
77 seconds executing Git, including 44 seconds in repository construction.
Repeated process startup makes the separate metadata probes expensive.

Query reference format, object format, bare status and the absolute common
directory in one `git rev-parse` call on the CLI construction path. Keep the
path last and split only the three fixed scalar fields so embedded newlines
remain intact. Git still computes every value. Supported Gix discovery keeps
its native metadata and existing CLI format-validation query, with unchanged
native launch ceilings and fallbacks.

The focused selection drops from 2,447 to 2,009 launches: three saved for
each of 146 repository opens. Three paired Windows measurements of 20 opens
each reduce the median from 300.74 ms to 221.18 ms per open, with matching
metadata and 11 versus eight launches. Add `--durations=30` to CI so future
slow tests are visible, and record the profiling evidence and reproduction
requirements in `doc/gix-backend.md`.

Cover both object formats, both reference formats, bare and non-bare repos,
spaces and embedded newlines. The eight applicable Windows cases fail before
the fix and pass afterward; all 16 cases pass on Ubuntu WSL. The optional
Windows Gix selection reproduces the same 17 failures on the original and
patched sources. Document the demonstrated path-formatting and undecodable
`commondir` differences under `GIX-14`, distinguishing them from Windows test
assumptions. Keep these bugs open pending compatible upstream behavior.

Validation: the full Windows CLI suite passes with 1,662 passed, 60 skipped,
nine expected failures, two unexpected passes and 40 passing subtests. It
records 77,732 launches in 3,715.16s. The initial baseline records 86,032 in
3,990.28s; its only two failures pass on unchanged code after restoring CI's
`core.autocrlf=true` and placing `pytest`'s temporary directory outside a Git
checkout. The overlapping full runs are not a controlled elapsed-time
comparison. Repository-wide Ruff lint/format, all pinned pre-commit hooks,
`mypy` (47 files), `basedpyright` (zero errors or warnings), the Sphinx HTML
build and `git diff --check` pass.

PR #2274's Python 3.12 CI jobs spent 57m 25s in Windows `pytest` versus 5m 2s
on Ubuntu, with 85,777 and 82,683 `Git.execute` launches respectively. A local
profile of the file-handle and recursive submodule regressions spent about 65 of
77 seconds executing Git, including 44 seconds in repository construction.
Repeated process startup makes the separate metadata probes expensive.

Query reference format, object format, bare status and the absolute common
directory in one `git rev-parse` call on the CLI construction path. Keep the
path last and split only the three fixed scalar fields so embedded newlines
remain intact. Git still computes every value. Supported Gix discovery keeps
its native metadata and existing CLI format-validation query, with unchanged
native launch ceilings and fallbacks.

The focused selection drops from 2,447 to 2,009 launches: three saved for
each of 146 repository opens. Three paired Windows measurements of 20 opens
each reduce the median from 300.74 ms to 221.18 ms per open, with matching
metadata and 11 versus eight launches. Add `--durations=30` to CI so future
slow tests are visible, and record the profiling evidence and reproduction
requirements in `doc/gix-backend.md`.

Cover both object formats, both reference formats, bare and non-bare repos,
spaces and embedded newlines. The eight applicable Windows cases fail before
the fix and pass afterward; all 16 cases pass on Ubuntu WSL. The optional
Windows Gix selection reproduces the same 17 failures on the original and
patched sources. Document the demonstrated path-formatting and undecodable
`commondir` differences under `GIX-14`, distinguishing them from Windows test
assumptions. Keep these bugs open pending compatible upstream behavior.

Validation: the full Windows CLI suite passes with 1,662 passed, 60 skipped,
nine expected failures, two unexpected passes and 40 passing subtests. It
records 77,732 launches in 3,715.16s. The initial baseline records 86,032 in
3,990.28s; its only two failures pass on unchanged code after restoring CI's
`core.autocrlf=true` and placing `pytest`'s temporary directory outside a Git
checkout. The overlapping full runs are not a controlled elapsed-time
comparison. Repository-wide Ruff lint/format, all pinned pre-commit hooks,
`mypy` (47 files), `basedpyright` (zero errors or warnings), the Sphinx HTML
build and `git diff --check` pass.

PR #2274's Python 3.12 CI jobs spent 57m 25s in Windows `pytest` versus 5m 2s
on Ubuntu, with 85,777 and 82,683 `Git.execute` launches respectively. A local
profile of the file-handle and recursive submodule regressions spent about 65 of
77 seconds executing Git, including 44 seconds in repository construction.
Repeated process startup makes the separate metadata probes expensive.

Query reference format, object format, bare status and the absolute common
directory in one `git rev-parse` call on the CLI construction path. Keep the
path last and split only the three fixed scalar fields so embedded newlines
remain intact. Git still computes every value. Supported Gix discovery keeps
its native metadata and existing CLI format-validation query, with unchanged
native launch ceilings and fallbacks.

The focused selection drops from 2,447 to 2,009 launches: three saved for
each of 146 repository opens. Three paired Windows measurements of 20 opens
each reduce the median from 300.74 ms to 221.18 ms per open, with matching
metadata and 11 versus eight launches. Add `--durations=30` to CI so future
slow tests are visible, and record the profiling evidence and reproduction
requirements in `doc/gix-backend.md`.

Cover both object formats, both reference formats, bare and non-bare repos,
spaces and embedded newlines. The eight applicable Windows cases fail before
the fix and pass afterward; all 16 cases pass on Ubuntu WSL. The optional
Windows Gix selection reproduces the same 17 failures on the original and
patched sources. Document the demonstrated path-formatting and undecodable
`commondir` differences under `GIX-14`, distinguishing them from Windows test
assumptions. Keep these bugs open pending compatible upstream behavior.

Validation: the full Windows CLI suite passes with 1,662 passed, 60 skipped,
nine expected failures, two unexpected passes and 40 passing subtests. It
records 77,732 launches in 3,715.16s. The initial baseline records 86,032 in
3,990.28s; its only two failures pass on unchanged code after restoring CI's
`core.autocrlf=true` and placing `pytest`'s temporary directory outside a Git
checkout. The overlapping full runs are not a controlled elapsed-time
comparison. Repository-wide Ruff lint/format, all pinned pre-commit hooks,
`mypy` (47 files), `basedpyright` (zero errors or warnings), the Sphinx HTML
build and `git diff --check` pass.

The Windows Python 3.14 job for PR #2274 failed in
`TestSubmodule.test_base_rw` when forced module-only removal called
`rmtree`: Windows reported `WinError 32` for `git/ext/gitdb`.
The legacy test retains that checkout's `Repo` through `csm.repo` and
starts its persistent `cat-file` process while reading parent commits.
Closing only the fresh handle created by `Submodule.remove()` cannot
release this caller-owned process.

Close the retained test-owned parent before deleting its checkout. Keep
strict removal and all existing happy-path assertions, and remove the
obsolete commented-out skip that claimed removal ignored this error.
The original test predates the PR's base, so record this as a follow-up
rather than rewriting shared history.

The focused `test_base_rw` passed locally with both CLI and Gix.
Relevant pre-commit hooks passed. Hosted Windows validation remains
part of the PR CI loop.

The Windows Python 3.14 job for PR #2274 failed in
`TestSubmodule.test_base_rw` when forced module-only removal called
`rmtree`: Windows reported `WinError 32` for `git/ext/gitdb`.
The legacy test retains that checkout's `Repo` through `csm.repo` and
starts its persistent `cat-file` process while reading parent commits.
Closing only the fresh handle created by `Submodule.remove()` cannot
release this caller-owned process.

Close the retained test-owned parent before deleting its checkout. Keep
strict removal and all existing happy-path assertions, and remove the
obsolete commented-out skip that claimed removal ignored this error.
The original test predates the PR's base, so record this as a follow-up
rather than rewriting shared history.

The focused `test_base_rw` passed locally with both CLI and Gix.
Relevant pre-commit hooks passed. Hosted Windows validation remains
part of the PR CI loop.

PR #2274's Python 3.12 CI jobs spent 57m 25s in Windows `pytest` versus 5m 2s
on Ubuntu, with 85,777 and 82,683 `Git.execute` launches respectively. A local
profile of the file-handle and recursive submodule regressions spent about 65 of
77 seconds executing Git, including 44 seconds in repository construction.
Repeated process startup makes the separate metadata probes expensive.

Query reference format, object format, bare status and the absolute common
directory in one `git rev-parse` call on the CLI construction path. Keep the
path last and split only the three fixed scalar fields so embedded newlines
remain intact. Git still computes every value. Supported Gix discovery keeps
its native metadata and existing CLI format-validation query, with unchanged
native launch ceilings and fallbacks.

The focused selection drops from 2,447 to 2,009 launches: three saved for
each of 146 repository opens. Three paired Windows measurements of 20 opens
each reduce the median from 300.74 ms to 221.18 ms per open, with matching
metadata and 11 versus eight launches. Add `--durations=30` to CI so future
slow tests are visible, and record the profiling evidence and reproduction
requirements in `doc/gix-backend.md`.

Cover both object formats, both reference formats, bare and non-bare repos,
spaces and embedded newlines. The eight applicable Windows cases fail before
the fix and pass afterward; all 16 cases pass on Ubuntu WSL. The optional
Windows Gix selection reproduces the same 17 failures on the original and
patched sources. Document the demonstrated path-formatting and undecodable
`commondir` differences under `GIX-14`, distinguishing them from Windows test
assumptions. Keep these bugs open pending compatible upstream behavior.

Validation: the full Windows CLI suite passes with 1,662 passed, 60 skipped,
nine expected failures, two unexpected passes and 40 passing subtests. It
records 77,732 launches in 3,715.16s. The initial baseline records 86,032 in
3,990.28s; its only two failures pass on unchanged code after restoring CI's
`core.autocrlf=true` and placing `pytest`'s temporary directory outside a Git
checkout. The overlapping full runs are not a controlled elapsed-time
comparison. Repository-wide Ruff lint/format, all pinned pre-commit hooks,
`mypy` (47 files), `basedpyright` (zero errors or warnings), the Sphinx HTML
build and `git diff --check` pass.

The Windows Python 3.14 job for PR #2274 failed in
`TestSubmodule.test_base_rw` when forced module-only removal called
`rmtree`: Windows reported `WinError 32` for `git/ext/gitdb`.
The legacy test retains that checkout's `Repo` through `csm.repo` and
starts its persistent `cat-file` process while reading parent commits.
Closing only the fresh handle created by `Submodule.remove()` cannot
release this caller-owned process.

Close the retained test-owned parent before deleting its checkout. Keep
strict removal and all existing happy-path assertions, and remove the
obsolete commented-out skip that claimed removal ignored this error.
The original test predates the PR's base, so record this as a follow-up
rather than rewriting shared history.

The focused `test_base_rw` passed locally with both CLI and Gix.
Relevant pre-commit hooks passed. Hosted Windows validation remains
part of the PR CI loop.

Review item #2274 (comment)
reports that "reading them raises `AttributeError`" when native decoding has
no tagger, and requests preserving initialized fields through a CLI fallback.
Four SHA-1/SHA-256 regressions reproduce the native failure. The two cases
with a message also expose CLI loss of the first message line, so falling
back alone would not restore complete metadata.

Adapt `Tag.tagger()` returning `None` to the existing empty `Actor` and zero
date/offset defaults through `parse_actor_and_date("")`. Gix already exposes
the absence correctly; no new Git semantics or fallback is required. Make
the CLI parser recognize the optional tagger header and preserve the entire
message. Git 2.54.0's `tag.c::parse_tag_buffer()` accepts an absent tagger
and sets its date to zero. The existing GIX-28 strict writer gap remains open.

All four regressions now pass on both backends; the native cases forbid
`Git.execute()` during metadata access. Reference and CLI object/index tests
pass on CLI (105 tests, 24 subtests) and Gix (105 tests, 24 subtests).
Ruff, ShellCheck and mypy pass.

Review item #2274 (comment)
reports that "subsequent iteration or lookup silently accepts that truncated
tree" after an entry fails validation. Stored trees containing a `.git`
entry reproduce this on both backends: the first iteration rejects the name,
but the second accepts an empty tree or the valid prefix.

Build the entries in a local list and attach `_cache` only after every record
passes parsing and validation. Failed reads leave the lazy cache unset, so
later accesses retry and raise the same error instead of exposing partial data.
No backend behavior or tree validation policy changes. Git 2.54.0's `mktree`
and `ls-tree` accept and emit the fixture's `.git` entry; GitPython applies
its existing metadata-path guard while consuming that output.

Four regressions cover invalid entries first and last in SHA-1/SHA-256 trees.
All fail before the fix and pass afterwards on CLI and native Gix `ls_tree`.
All 97 tree tests pass on each backend. Ruff, ShellCheck and mypy pass.

Review item #2274 (comment)
reports that "the returned head points to `refs/heads/topic`, which may be
missing or identify a different branch" after renaming to the short branch
name `refs/heads/topic`. A repository with an existing `topic` at commit A
and a source branch at commit B reproduces the incorrect returned ref on
both backends.

Compute the destination by prepending `refs/heads/` to the literal name
passed to `git branch`. Validate that same destination and retain it only
after Git successfully renames the branch. Git 2.54.0's
`builtin/branch.c::copy_or_rename_branch()` uses `refs.c::check_branch_ref()`,
which always prepends that namespace. The general `to_full_path()` helper
continues to normalize already-qualified references for its other callers.

The SHA-1/SHA-256 regressions fail before this fix and pass afterwards.
They check the returned path, name and commit, current `HEAD`, and that the
original `topic` still points to commit A. All 50 reference tests and 24
subtests pass on each backend, including ordinary and forced renames.
Ruff, ShellCheck, mypy and basedpyright pass.

Review item #2274 (comment)
reports that "reading them raises `AttributeError`" when native decoding has
no tagger, and requests preserving initialized fields through a CLI fallback.
Four SHA-1/SHA-256 regressions reproduce the native failure. The two cases
with a message also expose CLI loss of the first message line, so falling
back alone would not restore complete metadata.

Adapt `Tag.tagger()` returning `None` to the existing empty `Actor` and zero
date/offset defaults through `parse_actor_and_date("")`. Gix already exposes
the absence correctly; no new Git semantics or fallback is required. Make
the CLI parser recognize the optional tagger header and preserve the entire
message. Git 2.54.0's `tag.c::parse_tag_buffer()` accepts an absent tagger
and sets its date to zero. The existing GIX-28 strict writer gap remains open.

All four regressions now pass on both backends; the native cases forbid
`Git.execute()` during metadata access. Reference and CLI object/index tests
pass on CLI (105 tests, 24 subtests) and Gix (105 tests, 24 subtests).
Ruff, ShellCheck and mypy pass.

Review item #2274 (comment)
reports that "subsequent iteration or lookup silently accepts that truncated
tree" after an entry fails validation. Stored trees containing a `.git`
entry reproduce this on both backends: the first iteration rejects the name,
but the second accepts an empty tree or the valid prefix.

Build the entries in a local list and attach `_cache` only after every record
passes parsing and validation. Failed reads leave the lazy cache unset, so
later accesses retry and raise the same error instead of exposing partial data.
No backend behavior or tree validation policy changes. Git 2.54.0's `mktree`
and `ls-tree` accept and emit the fixture's `.git` entry; GitPython applies
its existing metadata-path guard while consuming that output.

Four regressions cover invalid entries first and last in SHA-1/SHA-256 trees.
All fail before the fix and pass afterwards on CLI and native Gix `ls_tree`.
All 97 tree tests pass on each backend. Ruff, ShellCheck and mypy pass.

Review item #2274 (comment)
reports that "the returned head points to `refs/heads/topic`, which may be
missing or identify a different branch" after renaming to the short branch
name `refs/heads/topic`. A repository with an existing `topic` at commit A
and a source branch at commit B reproduces the incorrect returned ref on
both backends.

Compute the destination by prepending `refs/heads/` to the literal name
passed to `git branch`. Validate that same destination and retain it only
after Git successfully renames the branch. Git 2.54.0's
`builtin/branch.c::copy_or_rename_branch()` uses `refs.c::check_branch_ref()`,
which always prepends that namespace. The general `to_full_path()` helper
continues to normalize already-qualified references for its other callers.

The SHA-1/SHA-256 regressions fail before this fix and pass afterwards.
They check the returned path, name and commit, current `HEAD`, and that the
original `topic` still points to commit A. All 50 reference tests and 24
subtests pass on each backend, including ordinary and forced renames.
Ruff, ShellCheck, mypy and basedpyright pass.

Review item #2274 (comment)
reports that "reading them raises `AttributeError`" when native decoding has
no tagger, and requests preserving initialized fields through a CLI fallback.
Four SHA-1/SHA-256 regressions reproduce the native failure. The two cases
with a message also expose CLI loss of the first message line, so falling
back alone would not restore complete metadata.

Adapt `Tag.tagger()` returning `None` to the existing empty `Actor` and zero
date/offset defaults through `parse_actor_and_date("")`. Gix already exposes
the absence correctly; no new Git semantics or fallback is required. Make
the CLI parser recognize the optional tagger header and preserve the entire
message. Git 2.54.0's `tag.c::parse_tag_buffer()` accepts an absent tagger
and sets its date to zero. The existing GIX-28 strict writer gap remains open.

All four regressions now pass on both backends; the native cases forbid
`Git.execute()` during metadata access. Reference and CLI object/index tests
pass on CLI (105 tests, 24 subtests) and Gix (105 tests, 24 subtests).
Ruff, ShellCheck and mypy pass.

Review item #2274 (comment)
reports that "subsequent iteration or lookup silently accepts that truncated
tree" after an entry fails validation. Stored trees containing a `.git`
entry reproduce this on both backends: the first iteration rejects the name,
but the second accepts an empty tree or the valid prefix.

Build the entries in a local list and attach `_cache` only after every record
passes parsing and validation. Failed reads leave the lazy cache unset, so
later accesses retry and raise the same error instead of exposing partial data.
No backend behavior or tree validation policy changes. Git 2.54.0's `mktree`
and `ls-tree` accept and emit the fixture's `.git` entry; GitPython applies
its existing metadata-path guard while consuming that output.

Four regressions cover invalid entries first and last in SHA-1/SHA-256 trees.
All fail before the fix and pass afterwards on CLI and native Gix `ls_tree`.
All 97 tree tests pass on each backend. Ruff, ShellCheck and mypy pass.

Review item #2274 (comment)
reports that "the returned head points to `refs/heads/topic`, which may be
missing or identify a different branch" after renaming to the short branch
name `refs/heads/topic`. A repository with an existing `topic` at commit A
and a source branch at commit B reproduces the incorrect returned ref on
both backends.

Compute the destination by prepending `refs/heads/` to the literal name
passed to `git branch`. Validate that same destination and retain it only
after Git successfully renames the branch. Git 2.54.0's
`builtin/branch.c::copy_or_rename_branch()` uses `refs.c::check_branch_ref()`,
which always prepends that namespace. The general `to_full_path()` helper
continues to normalize already-qualified references for its other callers.

The SHA-1/SHA-256 regressions fail before this fix and pass afterwards.
They check the returned path, name and commit, current `HEAD`, and that the
original `topic` still points to commit A. All 50 reference tests and 24
subtests pass on each backend, including ordinary and forced renames.
Ruff, ShellCheck, mypy and basedpyright pass.

Since 212573b (gitpython-developers#2274), GitPython requires Git 2.52 or newer for
library-managed operations, and the tests require it too. The README
says that Git must be "available in `PATH` or selected with
`GIT_PYTHON_GIT_EXECUTABLE`", which lets a developer whose default
`git` is older point GitPython at a newer one without changing
`PATH`.

tox did not support the second way. Its environments only receive the
variables tox passes by default plus those in `pass_env`, which listed
just `SSH_*`. So `GIT_PYTHON_GIT_EXECUTABLE` set in the shell was
silently dropped, and `tox` ran the tests against whatever `git` came
first in `PATH`, even when that was too old.

Add `GIT_PYTHON_GIT_EXECUTABLE` to `pass_env` in `[testenv]`. Every
environment inherits it, including `gix` and the `py38` through
`py310` import smoke checks, none of which override `pass_env`.
`GIT_PYTHON_BACKEND` is still not passed through, since `[testenv]`
and `gix` each set it explicitly.

Checked locally with tox 4.67.1 by running an environment with its
command overridden to print the variable, with
`GIT_PYTHON_GIT_EXECUTABLE` set in the calling shell. Before this
change it printed `None`; after it, the value that was set.
`tox config` shows the variable in `pass_env` for `py311`, `gix`, and
`mypy`.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

The tox `ruff`, `format`, and `misc` environments only run tools on
the source tree: `ruff check`, `ruff format --check`, and the
remaining pre-commit hooks. But they did not set `skip_install`, so
they inherited the default of installing GitPython with
`extras = test` from `[testenv]`. That has always been unneeded work,
but it became a real obstacle when 212573b (gitpython-developers#2274) added GixPython to
the `test` extra for Python 3.11 and later. GixPython 0.3.0 publishes
wheels only for macOS, so on Linux and Windows installing it builds
its source distribution, which needs Rust 1.89 or newer and a linker.
So running Ruff through tox on those platforms now needed a Rust
toolchain.

Set `skip_install = true` in all three environments. `ruff` and
`format` already list `ruff` in `deps`. `misc` got `pre-commit` from
the `test` extra, so list it in `deps` instead. Like the `test`
extra's entry, it is unversioned. The pre-commit hooks themselves are
installed by pre-commit into its own environments as before.

Run locally with tox 4.67.1, each environment now installs only
`ruff` or `pre-commit` and passes. `ruff` reports "All checks
passed!", `format` reports 112 files already formatted, and `misc`
runs every non-Ruff hook successfully. `ruff` and `format` each took
about 1.5 seconds, compared with about 12 seconds for `ruff` before,
most of which went to installing the `test` extra.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Since 212573b (gitpython-developers#2274), GitPython requires Git 2.52 or newer for
library-managed operations, and the tests require it too. The README
says that Git must be "available in `PATH` or selected with
`GIT_PYTHON_GIT_EXECUTABLE`", which lets a developer whose default
`git` is older point GitPython at a newer one without changing
`PATH`.

tox did not support the second way. Its environments only receive the
variables tox passes by default plus those in `pass_env`, which listed
just `SSH_*`. So `GIT_PYTHON_GIT_EXECUTABLE` set in the shell was
silently dropped, and `tox` ran the tests against whatever `git` came
first in `PATH`, even when that was too old.

Add `GIT_PYTHON_GIT_EXECUTABLE` to `pass_env` in `[testenv]`. Every
environment inherits it, including `gix` and the `py38` through
`py310` import smoke checks, none of which override `pass_env`.
`GIT_PYTHON_BACKEND` is still not passed through, since `[testenv]`
and `gix` each set it explicitly.

Checked locally with tox 4.67.1 by running an environment with its
command overridden to print the variable, with
`GIT_PYTHON_GIT_EXECUTABLE` set in the calling shell. Before this
change it printed `None`; after it, the value that was set.
`tox config` shows the variable in `pass_env` for `py311`, `gix`, and
`mypy`.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

The tox `ruff`, `format`, and `misc` environments only run tools on
the source tree: `ruff check`, `ruff format --check`, and the
remaining pre-commit hooks. But they did not set `skip_install`, so
they inherited the default of installing GitPython with
`extras = test` from `[testenv]`. That has always been unneeded work,
but it became a real obstacle when 212573b (gitpython-developers#2274) added GixPython to
the `test` extra for Python 3.11 and later. GixPython 0.3.0 publishes
wheels only for macOS, so on Linux and Windows installing it builds
its source distribution, which needs Rust 1.89 or newer and a linker.
So on those platforms, running Ruff through tox needed a Rust
toolchain whenever 3.10 was not found but 3.11 or 3.12 was, since tox
then picked one of those for these environments.

Set `skip_install = true` in all three environments. `ruff` and
`format` already list `ruff` in `deps`. `misc` got `pre-commit` from
the `test` extra, so list it in `deps` instead. Like the `test`
extra's entry, it is unversioned. The pre-commit hooks themselves are
installed by pre-commit into its own environments as before.

Run locally with tox 4.67.1, each environment now installs only
`ruff` or `pre-commit` and passes. `ruff` reports "All checks
passed!", `format` reports 112 files already formatted, and `misc`
runs every non-Ruff hook successfully. `ruff` and `format` each took
about 1.5 seconds, compared with about 12 seconds for `ruff` before,
most of which went to installing the `test` extra.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Since 212573b (gitpython-developers#2274), GitPython requires Git 2.52 or newer for
library-managed operations, and the tests require it too. The README
says that Git must be "available in `PATH` or selected with
`GIT_PYTHON_GIT_EXECUTABLE`", which lets a developer whose default
`git` is older point GitPython at a newer one without changing
`PATH`.

tox did not support the second way. Its environments only receive the
variables tox passes by default plus those in `pass_env`, which listed
just `SSH_*`. So `GIT_PYTHON_GIT_EXECUTABLE` set in the shell was
silently dropped, and `tox` ran the tests against whatever `git` came
first in `PATH`, even when that was too old.

Add `GIT_PYTHON_GIT_EXECUTABLE` to `pass_env` in `[testenv]`. Every
environment inherits it, including `gix` and the `py38` through
`py310` import smoke checks, none of which override `pass_env`.
`GIT_PYTHON_BACKEND` is still not passed through, since `[testenv]`
and `gix` each set it explicitly.

Checked locally with tox 4.67.1 by running an environment with its
command overridden to print the variable, with
`GIT_PYTHON_GIT_EXECUTABLE` set in the calling shell. Before this
change it printed `None`; after it, the value that was set.
`tox config` shows the variable in `pass_env` for `py311`, `gix`, and
`mypy`.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

The tox `ruff`, `format`, and `misc` environments only run tools on
the source tree: `ruff check`, `ruff format --check`, and the
remaining pre-commit hooks. But they did not set `skip_install`, so
they inherited the default of installing GitPython with
`extras = test` from `[testenv]`. That has always been unneeded work,
but it became a real obstacle when 212573b (gitpython-developers#2274) added GixPython to
the `test` extra for Python 3.11 and later. GixPython 0.3.0 publishes
wheels only for macOS, so on Linux and Windows installing it builds
its source distribution, which needs Rust 1.89 or newer and a linker.
So on those platforms, running Ruff through tox needed a Rust
toolchain whenever 3.10 was not found but 3.11 or 3.12 was, since tox
then picked one of those for these environments.

Set `skip_install = true` in all three environments. `ruff` and
`format` already list `ruff` in `deps`. `misc` got `pre-commit` from
the `test` extra, so list it in `deps` instead. Like the `test`
extra's entry, it is unversioned. The pre-commit hooks themselves are
installed by pre-commit into its own environments as before.

Run locally with tox 4.67.1, each environment now installs only
`ruff` or `pre-commit` and passes. `ruff` reports "All checks
passed!", `format` reports 114 files already formatted, and `misc`
runs every non-Ruff hook successfully. `ruff` and `format` each took
about 1.5 seconds, compared with about 12 seconds for `ruff` before,
most of which went to installing the `test` extra.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>