Qiskit / coheriq / 40980589058
86%
main: 93%

LAST BUILD BRANCH: gh-readonly-queue/main/pr-52-97a9f9f411b6aec7d08423c39d92b179d3d568b5
DEFAULT BRANCH: main
Ran
Jobs 1
Files 5
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 86.239%. Remained the same
40980589058

push

github

web-flow 
Run the unit test suite on Windows via spawn-based test isolation (#35)

* Park a prototype for spawn-based per-test isolation

Exploratory only; nothing here is wired into tox or CI.  `testpaths` covers
only src/ and test/, so tools/ is never collected and the deliberately
failing demo tests cannot leak into the suite.

The requirement in this repo is not fork, it is a process whose domain/engine
registry is empty.  A spawned interpreter satisfies that by construction,
where a fork only satisfies it as long as the parent has not already touched
the registry.  So the Windows gap and the pytest-forked maintenance problem
are the same problem, and the plugin here runs the suite green on pytest
9.2.0, the version whose report serialization broke pytest-forked and forced
the pytest<9.2 pin in e21e0ba.

Not adopted: --forked works on the platforms we test, and Windows already
gets real dispatch coverage through the in-process notebook and ragged envs.
The trigger to revisit is the pin, not Windows support.

Parked mainly for two findings that outlive the prototype.  First, the cost:
about 0.17s per test, dominated by importing pytest, which parallelizes away
almost entirely (4.5s serial to 0.5s at 16-way on 26 tests) but scales with
import cost, so it needs remeasuring before being trusted on a package with
heavy imports.  Second, the trap: the first version reported a segfaulting
test as nothing at all, a silent pass, because a child killed by a signal
writes no report and absence read as success.  Treating a missing call phase
as failure is the fix.  That failure mode belongs to any process-isolation
scheme, not just this one.

Assisted-by: Claude Opus 5

* Make the spawn prototype work under coverage

Still exploratory; nothing is wired into tox or CI.

Coverage now reports 81%, byte-identical to the --forked baseline: same
per-file percentages, same missing-line lists, 27 passed either way.

This was the open question most likely to sink the approach, ... (continued)

188 of 218 relevant lines covered (86.24%)

0.86 hits per line

Jobs
ID Job ID Ran Files Coverage
1 40980589058.1 5
86.24
GitHub Action Run
Source Files on build 40980589058