On 10/03/2020 11:02, Jan-Marek Glogowski wrote:
But honestly: why do we want to have subsequentcheck at all, and not just run all tests after the build. This would get rid of all the use_more_fonts and whatever over hacks currently exist. The machine should be busy enough building LO. Eventually it might even lead to more stable test runs ;-)
There are some honest unit tests around that have narrow, well-defined dependencies, and I think it's perfectly fine to keep them that way and have them run as soon as all their dependencies are available. For some CompilerTests it even makes sense to run them as early as possible during the build, to verify that the assumptions of how the build will be conducted are met.
But yes, as I outlined before, we could get rid of subsequentcheck in the sense of having all *Tests lumped under one check target and have each *Test spell out its dependencies (as coarse or as fine grained as practical) individually.
(And then maybe have a machinery to add specific *Tests also to additional targets, like your javacheck target at <https://gerrit.libreoffice.org/c/core/+/88833> "Introduce javacheck target and move Java tests", or Rene's demand for running certain JUnitTests against a non-instdir installation.)