Discussion for PMS TODO: Should use conditionals be allowed in RESTRICT?
Better you provide a reason they shouldn't; they've been in since (roughly) .51_pre17, mainly to allow for restricting fetch, although they're also useful for the misc pkgs that need to disable tests for certain flags enabled (lapack comes to mind).
Well, the questions are a) whether anyone's using them, b) whether they've been supported long enough to be considered EAPI-0 material and c) whether ebuild(5) is wrong when it says they're not allowed.
svn conversion of portage mildly screwed things up, but can find reference to it from 10/05; iow, it's been out on releases for a long while. Re: usage, atm not seeing anything, although was used at it's introduction.
Nothing in the tree uses them, and ebuild(5) and devmanual claim they're illegal. WONTFIXing, at least for EAPI 0.
ear mark it for EAPI-1 is better ... ferringb provided some good examples of where it'd be useful
Devmanual claiming something doesn't mean it's right (as you like to state, difference between policy and base format req). Bluntly, devmanual not stating it is an oversight on their part- it was in prior to that doc being written. Meanwhile, dev-python/numarray-1.5.2-r1 restrict="lapack? ( test )" So yep, it's used.
in use in tree and is useful and is supported by portage -> add to EAPI-0
I'd add that it is useful for one additional reason if we go the route of adding src_test as a mandatory phase. Specifically to be able to conditionally RESTRICT based on keyword, e.g. RESTRICT="ppc64? ( test )" for test suites that otherwise work but are known to fail on certain architectures (for endian or bitness reasons or whatever) where the desire or capability to fix the test suite does not exist.
c87c7a650743196266577f1c07acc4e669b5517e