polserver / polserver / 36239996799
82%

DEFAULT BRANCH: master
Ran
Jobs 1
Files 856
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: 81.935% (+0.006%) from 81.929%
36239996799

push

github

web-flow 
Keep test runs from colliding, and stop setup failures cascading (#985)

* Name the shard directory the shared resource it is

shard_test_1, shard_test_2, shard_test_roundtrip, unittest_pol and
ecompile_watch_test all write to coretest. Fixtures ordered only the
shard tests, so ctest -jN could start one of the others on top of a
running shard test. They now share RESOURCE_LOCK coretest.

* Compare the escript output without starting a git to do it

compareresult() ran git diff --no-index once per script per pass, 84
times for escript_string alone. CMake compares the files now, ignoring
trailing whitespace and a missing final newline as git did. git still
draws the diff when they differ.

escript_string: 7.7s -> 6.4s. All escript tests at -j8: 51s -> 37s.

* Offer the whole suite as one parallel ctest run

The five test targets run one after another. test_all runs the same
tests in one ctest -j, so the escript suite overlaps the mostly idle
shard test. It is opt-in; existing targets and CI are unchanged.

- DEPENDS keeps the order the coretest family needs.
- COST starts shard_test_1 first. PROCESSORS charges it two slots.
- shard_test_1 gets a ctest TIMEOUT, so a wedged shard reports a
  timeout.

On a 16-CPU box it is no faster, 285s against 280s. It is meant for the
4-vCPU CI runners, where it is not yet measured.

* Stop one package's failed setup from failing the next package

tests.src skipped a package's cleanup when its setup failed, which left
accounts and objects behind for later packages. A failed setup now
still runs the cleanup. IGNORE_TEST still skips both.

- The client cleanup copes with a setup that never connected.
- The boat package removes the probe NPC corpses it leaves on 100,100,
  where the house tests place a multi. Only the client package's corpse
  test cleared them before.

* Move the test client's control port out of the dynamic range

The test client listened on 50000, inside the Windows dynamic port
range. Windows reserve... (continued)

60259 of 73545 relevant lines covered (81.93%)

625094.58 hits per line

Coverage Regressions

Lines Coverage ∆ File
2
63.41
0.0% pol-core/pol/mobile/ufacing.cpp
Jobs
ID Job ID Ran Files Coverage
1 36239996799.1 856
81.93
GitHub Action Run
Source Files on build 36239996799