Repository navigation
test: split conflict-prone count pins and shared tables so concurrent PRs touch disjoint lines #2615
Description
Independent PRs keep conflicting in the same few test lines. Each new registry, generator family or dispatch table bumps a shared count pin, or appends to a shared tuple or dict. So every concurrent pair conflicts, and every rebase re-derives the numbers by hand. Seen in the rebases of #2595, #2612 and #2603:
test_dispatch_registry_unit(test_cases_cover_every_table_named_in_the_issue): a scalar total plus a per-family dict.- The four const count pins:
test_const_enum_builtin_parity,test_const_enum_get,test_const_enum_lookupandtest_const_registry_protocol. examples/generators/options.py'sFAMILIES, where every family appends at the tail.
Plan. Restructure these, and any other test hotspot that the merge history shows, so that independent changes touch disjoint files or lines:
- derive totals rather than pinning them;
- move one-entry-per-item data into per-item files or sorted one-per-line manifests;
- split monolithic modules by protocol where that is the conflict.
Each guard keeps its strength: it must still fail on the same regressions, which mutation checks will show.
Activity
JarryShaw commentedon Nov 2, 2026
Owner ruling, paraphrased: for the test files that keep causing heavy conflicts, I have the liberty and autonomy to split them so they stop conflicting.
This answers the question of how far the restructure may go: no further sign-off is needed on the structure chosen. A worker is on it. The change is test-only (
tests/andexamples/generators/), and every restructured guard must still fail on the same regressions, which mutation checks will prove. #2595 and #2607 edit some of the same pins, so whichever of them lands after this one rebases onto the new layout.JarryShaw commentedon Nov 2, 2026
The PR is #2644.
test: split the conflict-prone count pins and generator tables (#2615)
test: split the conflict-prone count pins and generator tables (#2615) (