tree-wide: disable doCheck and doInstallCheck where it fails#39463
tree-wide: disable doCheck and doInstallCheck where it fails#394637c6f434c merged 1 commit intoNixOS:stagingfrom
doCheck and doInstallCheck where it fails#39463Conversation
GrahamcOfBorg
added
6.topic: GNOME
|
For |
oxij
force-pushed
the
stdenv/docheckfalse-trivial
branch
from
5b3fefd to
7438083
Compare
GrahamcOfBorg
added
10.rebuild-darwin-stdenv
|
I think might be interested: @grahamc @Profpatsch @Ekleog Have tried to check that all the cleanups are pure layout changes (toward more uniformity). I think I have read every change; the cleanups are pure cleanups. Please remind me in a few days where few is more than a couple, if nobody merges it before. |
peti
left a comment
peti
left a comment
There was a problem hiding this comment.
This PR adds doCheck = false to GHC version 8.2.2, but not to any of the other versions we have even though they'll surely need the same setting. I suppose this PR cannot be expected to be "complete" since that would be a lot of packages to compile, but still it might be worthwhile to point out that switching to doCheck = true would break a lot of packages -- even after these changes.
|
Ugh, this leads to rebuilds because |
|
This PR would be a rebuild anyway: there are some cleanups in the layout of the shell script parts of some expressions.
|
Interestingly, no, at least not for the older ones. I compiled a bunch of them and only this one was broken for some reason.
This PR is maybe a third of all the changes. And even then I didn't try to recompile every single thing, so there are going to be errors even after (and if) the whole patchset is merged. Also I wouldn't recommend switch
In the perfect world |
|
This PR is maybe a third of all the changes. And even then I didn't try to recompile every single thing, so there are going to be errors even after (and if) the whole patchset is merged.
I think breakage level below the breakage of an allegedely minor bump in a highly-used library should be considered acceptable — obviously, as a proponent of tristates I say that the current PR is useful regardless of the eventual default value of doCheck.
In the perfect world `mkDerivation` would normalize shell scripts that that would be a noop too :)
At that point we have to just go full-Guix-style with a language for builders. I mean, the builder text is accessible to the build, so technically it is possible to copy it to $out…
|
|
@peti do I understand your comment as «not complete but an OK step in the general direction»? Looking at the feedback I am inclined to just merge it before we get a formatting-based merge conflict or something like that. |
|
@7c6f434c, yes, sure. I am happy with the direction of this PR. |
|
Nice. I updated #39464 with most of the infrastructure. See the OP there. |
8 participants
(the trivial part)
Motivation for this change
#33599. I implemented that issue for fun and giggles and now run a system that is "fully-tested". Turns out, not unexpectedly, an enormous number of packages fail to check. This is the first patch from that patchset. This patch disables many things that fail, but I couldn't (even partially) fix them. I.e. this patch contains only trivial
doCheck = falsechanges and a couple of cleanups. The rest is coming later.Note that I started documenting error messages too late, hence many changes here just disable tests without comments. I'm too lazy to repeat the effort to document the "doCheck = false; # fails" lines as rebuilding and fixing everything multiple times with 33599-patchset took almost two weeks.
Note: For most packages you won't be able to check why they actually fail with just
doCheck = true;without some more patches from the patchset because the patchset addscheckTragetautodetection to stdenv.Things done