Fix warning parsing emitted by newer version of GHC

[EduardSergeev]

Fixes #98

[dramforever]

Unfortunately this breaks the error_warnings handling below, IIUC

Maybe something like (?:\[GHC-.*\] )?\[(-W.+)\]?

[EduardSergeev]

This should only affect warning handling. Could you provide a simple example of a code which it breaks?

[dramforever]

okay sorry i misunderstood, this only matches warnings printed in a single line

[EduardSergeev]

Yes, but newer GHC versions emit multi-line warnings which in version 0.2.3 are captured as single_line_warning (cutting the rest of the warning off, see #98 for example).
Example of a warning which would be captured by the previous version of single_line_warning:

warning: [GHC-38417] [-Wmissing-signatures]
    Top-level binding with no type signature: test :: p -> p

With this change it is correctly handled as a multi-line warning.

[dramforever]

My concern is mainly that these warnings (in the error_warnings list below) should be shown as errors:

                '-Wdeferred-type-errors',
                '-Wdeferred-out-of-scope-variables',
                '-Wtyped-holes'

It would be fine as long as none of these could be emitted in a single line in GHC 9.6+. However if any of these could then the regex would match the [GHC-*] instead of the [-W*] part, and it would no longer match the list.

[Hexadecimaaal]

hi! i'm poking around on the pr branch with ghc 9.6.4, and what @dramforever mentioned above does happen:

seems like it only happens if we have ghc output like this:

ghci | .../test.hs:1:1-4: warning: [GHC-38417] [-Wmissing-signatures]
ghci |     Top-level binding with no type signature: mlem :: Num a => p -> a

with the [GHC-nnnnn] part. the defined but not used warning does not have this part:

ghci | .../test.hs:1:6-9: warning: [-Wunused-matches]
ghci |     Defined but not used: ‘blep’

ok i think mine was not rebuilt completely, i can confirm the fix does work for me. i'm looking into some minor issues, as the typed holes and deferred type errors should be severity error.

[Hexadecimaaal]

one issue here is that if i enable -fdefer-typed-holes (as is by default), the behavior should be upgrading a hole to a error severity in our diagnosis. this requires us to actually match for the \[(-W.+)\] part: (two lines below)

    warning: /^warning:(?: \[GHC-.*\])? \[(-W.+)\]$/

this work as intended on my setup.

for the single line warning we do rely on the group (error_warnings.indexOf(res_sl_warning[1]) >= 0 ?). however i could not really get any single line warning in the first place, perhaps we need older ghc? or do you have some examples for a single line warning that does hit this regex?

[EduardSergeev]

i could not really get any single line warning in the first place, perhaps we need older ghc? or do you have some examples for a single line warning that does hit this regex?

Personally I cannot recall seeing single-line warning from GHC myself. Maybe the author of this line can provide an example of the code which emit such warning in some version of GHC?

[dramforever]

As an example, GHC 8.10 prints -Wunrecognised-pragmas as a single line warning

{-# what #-}
module What where

[1 of 1] Compiling What             ( test.hs, test.o )

test.hs:1:1: warning: [-Wunrecognised-pragmas] Unrecognised pragma
  |
1 | {-# what #-}
  | ^^^

[dramforever]

9.6.4 does print it as a multiline, so ithink it's changed so that it's not possible to get a single-line error/warning with [GHC-*] (edit i was wrong), although as shown earlier 9.6 also has unnumbered warnings

[1 of 1] Compiling What             ( test.hs, test.o )

test.hs:1:1: warning: [GHC-42044] [-Wunrecognised-pragmas]
    Unrecognised pragma: WHAT
  |
1 | {-# what #-}
  | ^^^^^^^^

edit no it looks like single line is still possible if we turn up the available columns for the pretty printer -dppr-cols=150, which is imo fair because like, i have that many columns on the terminal! it's also probably a good idea to handle this to cover all bases

test.hs:1:1: warning: [GHC-42044] [-Wunrecognised-pragmas] Unrecognised pragma: WHAT
  |
1 | {-# what #-}
  | ^^^^^^^^

[dramforever]

Thanks for the help. I don't think i'll be able to try it out and figure out how to integrate it atm.

I'll look into the ci fails myself later.

[EduardSergeev]

I'll look into the ci fails myself later.

The build is already fixed. With this and this commits.
Fixed by a previous commit so both commits from this PR were no longer needed and have been removed by rebasing on the latest master.

[Hexadecimaaal]

after some experimentation and discussions, we've decided to go with the /^warning:(?: \[GHC-.*\])? \[(-W.+)\]$/, as it preserves the original behavior. i'm closing this in favor of #100. thank you for your help!