- User Since
- Dec 18 2014, 1:25 PM (123 w, 3 d)
This now validates on 8.0. https://phabricator.haskell.org/B15572
- Use a flag instead of checking for NoParent
Why do you want to revert this? I think this is useful to have.
I think this will validate now but is it better than before? Maybe. It seemed that to fix the tests I had to progressively make things more ugly.
- Fail if there is no parent on lookup
- Pass the right thing to be printed into IncorrectParent
- Remove tracing functions + comment
- Lookup using GHCi lookup function for overloaded case as well
- T13622 fix
- Add LookupSub test
- Prefer disambiguated occurences to ambiguous occurences
- Fix test failures
Fri, Apr 28
- Fix inverted logic in overloaded renamer
Thu, Apr 13
Wed, Apr 12
- New base commit
This is the first step in a larger refactoring I plan in this module.
- Fixup the broken tests, I think it will be better to fix this wrinkle about returning Just an arbritrary identifier
Mon, Apr 10
It is failing because the base commit is D3435 which I still need to sort out.
Sat, Apr 8
Fri, Apr 7
Tue, Apr 4
Mon, Apr 3
Is this patch still waiting to be reviewed @osa1 ?
- Update determinism001
Mar 29 2017
Mar 27 2017
Mar 25 2017
Mar 24 2017
Yes, this is a good idea but the implementation of the function could be improved.
Mar 23 2017
- updated comments
Mar 21 2017
Ok the code looks good now but still some error message wibbles I think.
Mar 20 2017
Perhaps it is because you are forcing instances to be loaded which are not imported by them module you are testing.
Try changing the error message to print out all possible instances and then seeing which ones are different before and after your patch.
Mar 19 2017
It seems the error messages now contain redundant information when a binding is both "relevant" and "very relevant" (type correct).
Mar 17 2017
The paper this links to is not finished but at least the first half is fairly complete and explains the purpose of the strictness analyser.
Mar 16 2017
Mar 14 2017
Is this actually faster given the cost of constructing a set each time? I don't have good intuition for this but I have thought about doing it before.
Mar 12 2017
Not the right place. The stderr output of ./validate is quite different on my machine to harbormaster which makes this difficult to debug.
Mar 11 2017
Mar 8 2017
What are these tests meant to test? Why does qq006 depend on parsec, that doesn't seem very robust.
Mar 6 2017
Mar 3 2017
Mar 1 2017
Gipeda graph which will hopefully catch up to this point - https://perf.haskell.org/ghc/#graph/tests/alloc/T12234;hl=55efc9718b520ef354e32c15c4b49cdfecce412f
Feb 26 2017
It also appears the repository was set incorrectly. Using arc diff is much easier in my experience!
Did you create this using the web interface? It is much better if you can create the diff using arc so that CI runs.
@Nolan You can update revision by using arc diff --update D3212 in future to avoid having to create a new differential.
Feb 25 2017
Adding Reid who has previously expressed an opinion about the amount of compiler output.
The build log js around 1200 lines after this change.
Put the flag in the right place..
Feb 24 2017
- Fix build
Feb 22 2017
@austin tells me that this should now be fixed.
Feb 21 2017
Yes you are right.
It is really hard to see what is going on here. Perhaps you could put each commit up as a separate diff?