kettle-dev / kettle-family / 37932715278

88%

push

github

🐛 Resume wave-dispatched commands by name, not position

template and test do not dispatch members positionally, so a --start-at
resume hint was wrong for them and silently dropped work.

Dispatch orderings differ per command:

    COMMAND        DISPATCH                        pending is suffix?
    template       gemspec-derived waves,           NO
                   concurrent + stop flag
    test           thread pool, stop flag           NO
    release        ReleaseWaves (config)            NO (already --only)
    lint/bup/      sequential, break on failure     YES
      check/docs/
      sync/etc.
    install        install_order, sequential        YES
    metadata       name-sorted, sequential          YES

Within a wave, template_results_for_wave uses a Queue and a thread pool,
and the first failure only sets stop = true, which prevents threads from
popping further work. Members already in flight still finish while later
waves never run (break memo unless wave_results.all?(&:ok?)). So the
pending set depends on thread scheduling and is interleaved through the
configured order rather than forming a suffix of it.

--start-at is positional: Selection#select_start_at drops everything
before the named member. Naming the failed member therefore drops every
pending member ordered ahead of it.

Measured on the kettle-dev family. nomono failed in template wave 1 while
kettle-dev, kettle-changelog, kettle-family and kettle-soup-cover were
pending at ordered positions 6, 8, 9 and 11, and kettle-drift at 7 and
kettle-wash at 10 had already succeeded:

    pending positions   [5, 7, 8, 10]
    suffix positions    [9, 10, 11, 12]   -> not a suffix

The emitted `--start-at nomono` resumed only nomono and token-resolver and
reported outcome success having never templated the other four. Confirmed
live, not just in specs:

    template --start-at nomono  -> kettle-dev, kettle-changelog marked "-"
    template --only nomono,...  -> same members marked "*"

Als... (continued)

2751 of 3533 branches covered (77.87%)

Branch coverage included in aggregate %.

15 of 15 new or added lines in 1 file covered. (100.0%)

6771 of 7258 relevant lines covered (93.29%)

110.02 hits per line