Mapping over PairRDD using Tuple2 results in GHC bug - makeStatic

[dzhus]

When trying to map a function that swaps Tuple2 members over a PairRDD a b to get PairRDD b a, I'm seeing this:

$ stack --nix build && stack --nix exec -- sparkle package sparkle-example-rddops && stack --nix exec -- spark-submit --master 'local[1]' sparkle-example-rddops.jar

...

["The","quick","brown","fox","jumps","over","the","lazy","dog"]
"dog lazy the over jumps fox brown quick The"
["ehT","kciuq","nworb","xof","spmuj","revo","eht","yzal","god"]
["The","quick","brown"]
["fox","The","jumps","quick","lazy","dog","over","brown","the"]
True
sparkle-worker: GHC bug - makeStatic: Unresolved static form at line 107, column 1.
CallStack (from HasCallStack):
  error, called at libraries/base/GHC/StaticPtr/Internal.hs:26:5 in base:GHC.StaticPtr.Internal
$ uname -a
Linux osboxes 4.8.0-36-generic #36~16.04.1-Ubuntu SMP Sun Feb 5 09:39:57 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux

Here's a minimal example: MailOnline@4191d4f

If I add StaticPointers pragma to PairRDD module, it works as expected (MailOnline@6708a2a). Any ideas as to what is exactly happening here? Is this a combination of https://ghc.haskell.org/trac/ghc/ticket/14204 + withStatic in PairRDD?

[mboes]

Yes, that's right. I got hit myself by this earlier last week. If you'd like to push your change as a PR, I can cut a new point release pronto. Thanks for the bug report!

FWIW, GHC 8.4 should have much better diagnostics.

[mboes]

I should clarify - in my case it was a missing LANGUAGE pragma in Language.Java (now fixed). It's unclear why this is showing up now though, seeing as these LANGUAGE pragmas must have been missing for a long time. Might be due to a new compiler version. There might be a few more of these lurking around, since the GHC bug you mention turns what ought to be a compile time error into a runtime error.

[dzhus]

Thanks, I opened a PR. There's no rush to make a new release – I'm fine using Git version, plus I have more changes coming.