I believe we need to do b) Revert commit 07da52ce2d in the ghc-7.10 branch and continue to use llvm-3.5 for the 7.10 series. I think the plan is to require llvm-3.6 or a later version for 7.12.1 and later Ghc 7.10.1 only supports llvm 3.5 as stated in https://ghc.haskell.org/trac/ghc/ticket/9929 I filed a doc bug on this yesterday, https://ghc.haskell.org/trac/ghc/ticket/10302 On Wed, Apr 15, 2015 at 7:12 AM, Erik de Castro Lopo wrote:

Erik de Castro Lopo wrote:

...

At this point we need to decide whether:

a) Require llvm-3.6 for 7.10.2 and later.

I've done some testing on the ghc-7.10 branch (with a couple of patches cherry picked from master) with llvm-3.6 on amd64-linux. WIth this, there are 17 unexpected failures during validate:

deriving/should_fail T7959 [stderr mismatch] (normal) deriving/should_fail drvfail005 [stderr mismatch] (normal) deriving/should_fail drvfail009 [stderr mismatch] (normal) indexed-types/should_fail SimpleFail14 [stderr mismatch] (normal) simplCore/should_compile spec-inline [stderr mismatch] (optasm) typecheck/should_compile T10195 [exit code non-0] (normal) typecheck/should_fail T2994 [stderr mismatch] (normal) typecheck/should_fail T3540 [stderr mismatch] (normal) typecheck/should_fail T7778 [stderr mismatch] (normal) typecheck/should_fail tcfail036 [stderr mismatch] (normal) typecheck/should_fail tcfail057 [stderr mismatch] (normal) typecheck/should_fail tcfail063 [stderr mismatch] (normal) typecheck/should_fail tcfail134 [stderr mismatch] (normal) typecheck/should_fail tcfail146 [stderr mismatch] (normal) typecheck/should_fail tcfail147 [stderr mismatch] (normal) typecheck/should_fail tcfail151 [stderr mismatch] (normal) typecheck/should_fail tcfail209 [stderr mismatch] (normal)

All bit one oft these all seem to be stderr mismatches rather than actual problems.

My validate run on armhf is still running.

Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/ _______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs

yes wrt b) making a one-character fix of 3.6 to 3.5 on the ghc-7.10 branch sound like the right thing to me. wrt a) my understanding is that for 7.10.2 we will also use llvm 3.5 as in 7.10.1 For 7.12.1 and forward we will also have one supported llvm version per release which right now for 7.12.1 is 3.6 but could become 3.7 The plan for llvm for 7.12.1and later is given in https://ghc.haskell.org/trac/ghc/wiki/ImprovedLLVMBackend. On Wed, Apr 15, 2015 at 9:55 AM, Reid Barton wrote:

On Wed, Apr 15, 2015 at 2:04 AM, Erik de Castro Lopo

...

wrote:

...

At this point we need to decide whether:

a) Require llvm-3.6 for 7.10.2 and later.

Surely we're not going to do this.

...

or

b) Revert commit 07da52ce2d in the ghc-7.10 branch and continue to use llvm-3.5 for the 7.10 series.

Why do we need to revert anything, can't we just make a one-character fix of 3.6 to 3.5 on the ghc-7.10 branch?

Regards, Reid Barton

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs

Erik de Castro Lopo writes:

Reid Barton wrote:

...

Why do we need to revert anything, can't we just make a one-character fix of 3.6 to 3.5 on the ghc-7.10 branch?

Even with that one character change the LLVM detection is not complete or correct. Specifically, building the compiler will enforce the right llvm version, but the binary-dist configure script just uses "llc" and "opt".

Furthermore, I am pretty sure that #10292 is causes by llvm-3.5 and that if we use llvm-3.6 buildng on Arm works. Unfortunately building and testing on Arm takes a long time and I am still in the process of verifying this.

While I would love to see LLVM 3.6 rolled out soon as a GHC target (it is substantially more stable on ARM in my experience) I agree that doing this for 7.10.2 would be a bit of a large change for a minor release. If nothing else it would be surprising for users to find that their previously working LLVM/GHC combination needed to be upgraded upon moving to 7.10.2. In my experience LLVM 3.5 works reasonably well on ARM; it seems like this should be sufficient to carry us through to 7.12. Cheers, - Ben