Tests for #2731 (Ignore dependencies that are not Buildable) #3039
Conversation
|
LGTM. Have you run tests, etc. for each individual commit? |
|
/cc @kosmikus I think this can be merged now. |
|
@BardurArantsson I tested each commit with GHC 7.10.2, but I can push the commits one at a time to build them with Travis. I'd like to make sure that the two commits from #2731 build. |
|
Fair enough, just checking :). I wonder if there's something that could be done to get Travis to always build each commit in a series... |
When doing the index conversion prior to dependency solving, we now consider the "Buildable" flag for package components. In particular, if the "Buildable" flag of a component is "True" only under certain conditions, then all build dependencies of that component will be placed under the same conditions.
When configuring a package, the condition trees in the package descriptions are evaluated according to the known configuration and flag assignment. During this process, it becomes also known whether a component has its "Buildable" flag set to True or False. We now disregard all dependencies of non-buildable components.
…r DSL. 'DSL.exResolve' now takes a 'Maybe [Extension]' for supported extensions and a 'Maybe [Language]' for supported languages. 'Nothing' means that extensions/languages are not checked by the solver, and 'Just []' means that no extensions/languages are allowed.
grayjay
force-pushed
the
buildable-tests
branch
from
cd65e96 to
928850d
Compare
|
https://docs.travis-ci.com/user/triggering-builds/ might be of help. |
|
I was hoping for something which could be set up once and for all for the whole repository. (Oh, how I long for a proper review + CI system...) |
|
Well, we do have a Phabricator instance :) (Though I don't think builds are setup.) |
|
No direct experience with Phab, but doesn't it require some sort of weird custom command line tool instead of just using standard git commands? (When reviewing and pushing patches to review, obviously.) Anyway, this is probably enough off-topic spam for this thread... |
Tests for #2731 (Ignore dependencies that are not Buildable)
|
Looks like they all succeeded. Merged |
|
Thanks! |
I added a package test for the Cabal component and several unit tests for the solver component. This PR also resolves a conflict between #2731 and #2732. I had to make a few changes to the solver DSL:
Allow a flag to be used multiple times in a package. Reusing a flag previously caused this error:
Represent the Buildable field. Now both sides of a conditional can be either not buildable, or buildable with a list of dependencies.
Allow a compiler to support zero languages or extensions. I changed the signature of
DSL.exResolveto distinguish between not checking languages/extensions and not allowing any languages/extensions.I thought of two other improvements, but I left them for separate issues or PRs. I didn't see an easy way to check flag choices directly in the DSL. I instead had to check the dependencies that the flags introduced. I also noticed that the default flag value is false in the DSL and true in cabal-install/Cabal. Some of these tests rely on the default, so it should probably be consistent.