Walk through re-imported history behind a root commit

[rumblefrog]

Follow-up to #31, fixing a regression it introduced for SourceMod.

Problem

SourceMod has three root commits. The mainline (first parent from HEAD) starts at 251cced (2008-03-30, "Spring Cleaning"). Its plugins/include is identical to 3a14544 (2008-03-28) in the older SVN history. That older history only reconnects as the second parent of the 2008-09 "Eliminated head" merges.

After #31 the walk followed first parents only, so it stopped at the snapshot:

• 1,581 symbols were "created" on 2008-03-30. For example, PrintToChat was really added 2007-06-01 (6c17b9f), GetClientName 2007-02-27, and HookEvent 2007-03-04.
• 1,017 symbols had "last updated" set to the snapshot instead of their real earlier change.

Change

• When the first-parent walk reaches a root commit, it looks for the latest commit elsewhere in HEAD's history, no newer than the root, whose files matching the patterns are identical to the root's. It then keeps walking first parents from there.
• Each commit is diffed against the previous commit walked. A re-imported root is unchanged from its predecessor, so it produces no changes.
• A root with no such predecessor is still diffed against an empty tree.
• The rev-list count is recomputed at the root. SourceMod counts restart there (e.g. PrintToChat is created at count 872, before 251cced at count 1), because that is what git rev-list --count reports for those commits.

Verification

• Two new walker tests: re-imported history, and an unrelated root that still starts from an empty tree. All 4 history tests pass.
• Ran the 11 upstream repos from the parser corpus (update-corpus.sh include patterns) and compared every symbol, nested ones included, against a reference that uses the git CLI. All 5,252 symbols match on created, last updated and count, with no missing or extra symbols. SourceMod now walks 1,133 changing commits instead of 713.
• PrintToChat, GetClientName and HookEvent are back to their 2007 commits. GetClientAuthId stays at the #147 merge (count 4830).
• No measurable slowdown: SourceMod goes from 5.2s to 6.1s because it walks about 420 more commits. The other repos are unchanged.

Bundles generated after #31 should be regenerated from scratch (without -b).

🤖 Generated with Claude Code

https://claude.ai/code/session_01UmDZ8MBYNvgeQ52tpsJe3t

---

Generated by Claude Code