Framework failures on windows with plugins

On Windows, when the load is heavy (i.e. when doing a full sh validate I get a bunch of framework failures in the plugins/ testsuite directory

Framework failures:
   plugins/plugins07.run             plugins07 [normal] (pre_cmd failed: 2)
   plugins/T10420.run                T10420 [normal] (pre_cmd failed: 2)
   plugins/T11244.run                T11244 [normal] (pre_cmd failed: 2)
   plugins/plugin-recomp-pure.run    plugin-recomp-pure [normal] (pre_cmd failed: 2)
   plugins/plugin-recomp-impure.run  plugin-recomp-impure [normal] (pre_cmd failed: 2)
   plugins/plugin-recomp-flags.run   plugin-recomp-flags [normal] (pre_cmd failed: 2)

I sent a full log to Tamar who said: Thanks for the log, that does give a clue. The command fails with

setup.exe: 'C:/code/HEAD/inplace/bin/ghc-pkg.exe' exited with an error:
...
rule-defining-plugin-0.1: cannot find any of
["libHSrule-defining-plugin-0.1-GxqqrdsQ5NRK9hAhEkvz8Z.a","libHSrule-defining-plugin-0.1-GxqqrdsQ5NRK9hAhEkvz8Z.p_a","libHSrule-defining-plugin-0.1-GxqqrdsQ5NRK9hAhEkvz8Z-ghc8.5.20180616.so","libHSrule-defining-plugin-0.1-GxqqrdsQ5NRK9hAhEkvz8Z-ghc8.5.20180616.dylib","HSrule-defining-plugin-0.1-GxqqrdsQ5NRK9hAhEkvz8Z-ghc8.5.20180616.dll"]
on library path (use --force to override)

make[2]: *** [Makefile:18: package.T10420] Error 1

so it's the setup install from https://github.com/ghc/ghc/blob/c2783ccf545faabd21a234a4dfc569cd856082b9/testsuite/tests/plugins/rule-defining-plugin/Makefile failing.

Unfortunately, all those tests run with -v0 which is annoying because now the verbosity of the testsuite doesn't control that of these tests. I'm not sure why these commands fail under heavy load though. I'll need to dive into the source of ghc-pkg to figure out what's happening.

Notice that all the framework failures are these plugin tests which modify a package database. A wild guess is that ghc-pkg tries to take a lock on all package-databases or something when it's mutating one. But I'm not intimately familiar with the package store and this doesn't explain why it doesn't happen on Linux.

For now one solution I can propose is to create a ticket to track these and mark these tests as cpu multirace on Windows.

Trac metadata

Trac field	Value
Version	8.4.3
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture