perf(gitpr): fetch the merge plan in one request, and show progress

[MCamner]

Follow-up to #180: wire ui_spinner into the PR-merge tool. Measuring the calls first changed what the fix should be.

The wait was nine round trips, not one slow one

gitpr-merge-safe.sh read nine PR fields with nine separate gh pr view calls, one per field:

9 separate pr_field calls: 3.44s
1 batched call:            0.47s

A spinner makes a wait legible. Not waiting is better. So the batching is the fix and the spinner covers what is left. Two more redundant calls went with it: a separate gh pr view --json number existence check that said what the plan read already says by failing, and gh pr checks running twice — once to print, once to decide.

Measured end to end against live gh: the whole plan phase now takes 0.99s, spinner included.

The delimiter is load-bearing

The batched fields are joined on U+001F, not a tab. Tab counts as IFS whitespace, so read collapses a run of them and one empty field shifts every later field left:

$ printf 'a\tb\t\tc\n' | { IFS=$'\t' read -r w x y z; echo "w=[$w] x=[$x] y=[$y] z=[$z]"; }
w=[a] x=[b] y=[c] z=[]        # <- c landed in y, z is gone

reviewDecision is empty on any PR nobody reviewed, which is most of them — so a tab-separated read would have printed the URL as the review decision. Confirmed against live gh on #180, where reviewDecision is in fact empty:

  Review:   none
  URL:      https://github.com/MCamner/macos-scripts/pull/180

join also refuses to join null, so nulls map to "" first.

Sourcing

The git menu launches this script as its own process (( cd "$CURRENT_REPO" && "$merge_script" )), it never sources it — so the script loads mq-ui.sh itself and falls back to a no-op ui_spinner when the library is unreachable. A missing spinner must not cost you the merge tool. Step 6 pins that.

Test coverage

First coverage this script has ever had; gitmerge-safe-smoke.sh covers the local-merge sibling, not this one. It matters here more than most — this is the one script in the repo that can change a remote branch.

The fake gh records every invocation, so "one call, not nine" is asserted against behaviour rather than against how the source happens to read. Also pinned: the no-TTY guardrail, that the confirmation can be declined, and that declining means gh pr merge was never called.

$ MACOS_SCRIPTS_HOME="$PWD" MQ_NO_TUI=1 ./tools/scripts/test-all.sh
[PASS] Shell lint passed at warning severity (197 files)
[PASS] All selftest checks passed.

🤖 Generated with Claude Code