fix(fixtures): pin the workspace_flow branch to main, and ignore t/ trash directories

[mizchi]

Fixes #191.

Two small t/ hygiene changes.

1. Pin the fixture's branch (#191)

fixtures/workspace_flow/bootstrap.sh created its four repositories with a bare git init, so the branch was whatever the host git defaulted to. Two tests built on that fixture name main, and fail wherever init.defaultBranch is unset and git still defaults to master (git 2.43 here):

	before	after
t9014-workspace-hub-flow.sh	0 of 2	2 of 2
t90010-workspace-git-compat.sh	5 of 7	7 of 7

t9014 broke in its setup — git checkout main found no such pathspec, and bit hub pr create --base refs/heads/main then reported Target branch not found. t90010 broke on git push -u origin main.

 git init -q
+git symbolic-ref HEAD refs/heads/main
 git config user.email "fixture@example.com"

Setting HEAD while it is still unborn rather than passing git init -b: the latter needs git 2.28, and the script is otherwise plain POSIX sh with no version floor.

Verification

With init.defaultBranch unset, all four repositories the fixture creates now report refs/heads/main, and every test built on it was run:

t9009   4/5   (1 pre-existing failure, see below)
t9011   4/4     t9015   4/4
t9012   5/5     t9016   7/7
t9013   5/5     t9017   6/6
t9014   2/2  ←  t9018   5/5
t90010  7/7  ←

The seven tests other than t9014 and t90010 never name a branch, and are unchanged either way.

Running the whole test-subdir task confirms one repair and no regression:

             passed  failed
before          4       7
after           5       6     (t90010 moved; nothing else changed)

2. Ignore the trash directories

t/test-lib.sh removes TRASH_DIRECTORY only when the whole file passes, so every red run leaves one behind. Six accumulated while triaging the t900 failures, and untracked scratch repositories sitting next to the tests are easy to commit by accident.

Verified by running a still-failing test and confirming git status stays clean.

What this does not fix

Six tests in the test-subdir task are red on main for reasons unrelated to branch names, and stay red. I triaged each and filed them individually: #193, #194, #195, #196, #197, #198.

They come down to two causes, not six:

• branch-name assumptions in the test files themselves, the same class as this PR — t9001-subdir-push: all four failures come from git push origin main on a master repo #193, part of t9004-workspace-routing: bit branch output is grepped for main on a master repo #194, t9005-workspace-commit-push: setup pushes to main on a master repo #195;
• implicit command translation not firing: bit status inside a workspace returns plain git status instead of the workspace form, while the explicit bit ws status works — t9004-workspace-routing: bit branch output is grepped for main on a master repo #194, t9007-workspace-nested-translation: repo status from a nested path does not fail as a plain repo command #196, t9008-workspace-flow-cache: implicit bit status after a flow run does not show workspace status #197, t9009-workspace-flow-fixture: implicit bit status in a fixture workspace does not show workspace status #198.

This matters for the plan sketched in #190, where I suggested #191 was the blocker for wiring t/ into CI. It is not the only one; that wiring needs the six above resolved first.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VDVUJHu38YtVAr6sKZtRDB