Adjust integrate output for ghc-9

ghc-9 introduces a regression(?) in FP precision (See ghc ticket #19569). While this is unfortunate we want nofib to be in sync with ghc.

So for the moment I will just accept the output produced by ghc-head for nofib.

changed the description

mentioned in merge request !44 (merged)

I'm not sure whether we should merge this. ghc#19569 (closed) is clearly a bug: We shouldn't risk backwards compatibility with floating point expression compiled by older compilers willy nilly.

So I'd rather say we should keep on passing --keep-going for the time being. Or maybe refine the --keep-going flag so that it takes a Shake target.

I'm not sure whether we should merge this. ghc#19569 (closed) is clearly a bug: We shouldn't risk backwards compatibility with floating point expression compiled by older compilers willy nilly.

In that case we should make this a ghc test instead of a benchmark. I opened a ticket about this on the ghc tracker. As for nofib I think I will just delete the stdout file so it passes for old and new versions of ghc.

Yes, that makes sense.

added 25 commits

• d1da8d8a...24a179b1 - 24 commits from branch master
• 7a32d6b5 - Adjust integrate output for ghc-9

Compare with previous version

approved this merge request

added 8 commits

• 7a32d6b5...e68b056d - 7 commits from branch master
• 53f24b38 - Adjust integrate output for ghc-9

Compare with previous version

added 1 commit

• 2596125d - Adjust integrate output for ghc-9

Compare with previous version

I've hid the numeric result by default in the benchmark and also replaced the random tabs in the benchmark with spaces in the process.

The testsuite of ghc will check for stability in the future so there is no need to check for the exact result. I also checked the new code path still forces the result properly.

added 3 commits

• 2596125d...2341139a - 2 commits from branch master
• bca01968 - Adjust integrate output for ghc-9

Compare with previous version

merged