Abandoning this for now. I'll submit a MR on GitLab in a few days, when it's actually ready for review. I've copied all the comments, and I can paste those in the new MR to preserve context.

Abandoned in favor of https://gitlab.haskell.org/ghc/ghc/merge_requests/181.

@bgamari Almost there, sorry it's taking a long time. I'm going to submit the last PR (Cabal) later today. And the containers PR needs some more work. I'll work on getting everything merged by the end of this week.

Remove duplicate pragma.

Note that the build won't be totally warning free just yet because this patch doesn't suppress these warnings for the libraries.

Supress warnings.

Sorry, I was wrong. GHC does generate warnings because of this.

I looked at the stage1 and stage2 build logs, and GHC itself didn't generate any new warnings. I'll try to fix any warnings that other libraries in haskell/ might generate.

@RyanGlScott Oops, sorry. Fixed it :)

Bring back -ddump-tc.

Fixed some failing tests using 'traceTcRnForUser'.

(fixing the build)

@bgamari Should be ready for review soon :) I'm still working on rewriting the GCC parts.

@bgamari I started looking at DriverPipeline, and got distracted.
Also, even after this patch, '-ddump-mod-map' (used in Packages.mkPackageState) still prints the output to STDOUT,
because dumpPrefix is still Nothing. Does DriverPipeline.runPipeline not run before we get to mkPackageState?
On the other hand, it works ok if we use -ddump-file-prefix=blah by hand.

Fix other instances of when (dopt ...) print ....

As @RyanGlScott pointed out, there are a few other places where when (dopt ...) print ... is used directly. Fixing those now.

No, because those options will not have been processed by the RTS. Why would users need to do that?

@simonmar Oh, I see what you mean. Filtering these out would actually make it seem that they were in fact processed by the RTS,
which would be incorrect.

Start updating semantics per @simonmar's comment (TODO: remove GCC bits)

Thanks @simonmar. I'll update this patch in a couple days.

Ping :)

This is still a WIP patch for option 2 mentioned in comment:2 in Trac #15732 (extending the RTS to parse response files). I submitted it here to get some early feedback. And there are a few things that must be dealt with:

LGTM. Thanks @AndreasK :)

The ticket mentions one more thing which I have not implemented:

consider making liftA2 known-key for Traversable deriving.

Thanks @bgamari!

@bgamari @monoidal Does this look ready to be merged ?

(We should also merge this in the 8.6 branch)

Thanks @RyanGlScott. Fixed it.

Fix typo in the commit message.

You can grep for grep, some tests declare what to look for the in the dump. But in this example, I think it is hard to write a reliable grep command.

Get rid of the test

Oh right, I should've used -dsuppress-uniques. Would a comment in the test file help ? Otherwise, I'll update this patch to not include it.

@nomeata's suggestion

@simonpj's suggestions

It turns out that git rm PATH && git commit removes a sub-module from the git working tree but doesn't delete the relevant section from .gitmodules.

Still WIP, and did not intend to submit a patch.

In general I wish that :set weren't so destructive, but that is another issue altogether.

Does this look alright ?

Oh no, it was my fault. I'm AFK right now, but I'll submit another patch to update this hint tomorrow.

Oops. Sorry, I should've changed the status to "Changes Planned" or something. Should I file a trac issue to fix this "hint" (@RyanGlScott's comment) ? Or is OK in it's current form ?

That's a good point. If we decide to add this hint, it could maybe include a note about this effect.

Include an example that uses getArgsWithResponseFiles in the haddocks.

The build failure seems unrelated to the patch:
exception 'PhabricatorWorkerPermanentFailureException' with message 'Lease "PHID-DRYL-zbqyzktfang7i4ajamnt" never activated.'

Define getArgsWithResponseFiles in GHC.ResponseFile instead of System.Environment.

:)

Thanks :)

Mention the trac number in the comment.

Fixed the instance for Lifetime

@RyanGlScott I think most of the annotations should be correct. I've tagged some of the places which I wasn't sure about.

Fixed deriving instance Traversable Identity