GeezerVison

I am an OpenAI Codex agent working through the contributor's account. This was found during a review of Git metadata APIs used by development tools: I checked whether blame results for working-tree changes could be consumed through the same datetime properties as committed results.

Both blame parsers currently populate the author and committer timestamps but omit their timezone offsets. For uncommitted lines, Git reports a null object ID, so reading authored_datetime or committed_datetime tries to load that nonexistent object and raises BadObject. Populate the two existing timezone fields from Git's porcelain output, using the existing offset conversion helper. Committed entries also avoid an unnecessary lazy object read.

The regression fixture creates a real repository with distinct +0530 author and -0400 committer offsets, then modifies its file. The test covers ordinary and incremental blame for both HEAD and working-tree changes, compares the datetime offsets to Git, and rejects unexpected lazy metadata loads.

Validation on Windows, CPython 3.13.5, Git 2.54.0 and GixPython 0.3.0:

• Final regression: four failing subtests on the exact original source, then passes with the fix.
• Complete non-network test_repo.py and test_commit.py: 175 passed, 8 skipped, 1 xfailed and 68 subtests passed with each of the CLI and Gix backend selections. Blame remains a CLI fallback with either backend.
• Ruff, mypy, configured basedpyright, shellcheck, Bash syntax and applicable file checks pass.
• Independent AI review exercised six API combinations and 24 returned entries, including positive/negative offsets and uncommitted lines.

The full repository suite, network-marked tests, other OS/Python combinations, documentation build and complete pre-commit wrapper were not run. GixPython's unmodified official source distribution was built locally because no Windows wheel is published; initial dependency build timeouts are retained in the work log. No Gix API or fallback policy changes are included.