LANGUAGE pragma fails if it falls on a 1024-byte boundary

[Deewiant]

Putting too much whitespace and/or comment text before a LANGUAGE pragma seems to cause parsing it to fail. In the following, the closing } of the comment is byte 1025. Perhaps the whole thing is expected to fit into the first 1024 bytes or be wholly inside an aligned 1024-byte range?

-- xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
-- xxxxxxxxxxxxxxxxxxxxxxxx
-- xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
-- xxxxxxxxxxxxxxxxxxxxxxx
-- xxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--
--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

{-# LANGUAGE ScopedTypeVariables #-}
main = return ()

Removing absolutely anything prior to the start of the LANGUAGE pragma will make it work: one of the exes, an entire line, any whitespace, whatever.

[Deewiant]

I tried adding more comments before it and that also made it work. Adding still more breaks it again as soon as the 2048th byte is among the contents of the {-# #-}, so this is indeed an alignment issue.

The error message which I managed to forget putting into the report:

arst.hs:21:13:
    cannot parse LANGUAGE pragma: comma-separated list expected

[Deewiant]

The culprit is getOptions' in compiler/main/HeaderInfo.hs. The function has a mechanism of indicating that it wants more input but it isn't doing it properly in this case. I'm not familiar enough with the system to be able to fix it, but it seems to me that modifying the definition of parseLanguage (in the where clause) somehow should suffice.

Note that the same function parses OPTION pragmas, which do work correctly even if they fall on a 1024-byte boundary.

[simonmar]

Fixed:

Thu Mar 12 14:11:03 GMT 2009  Simon Marlow <marlowsd@gmail.com>
  * FIX #3079, dodgy parsing of LANGUAGE pragmas
  I ended up rewriting this horrible bit of code, using (yikes) lazy I/O
  to slurp in the source file a chunk at a time.  The old code tried to
  read the file a chunk at a time, but failed with LANGUAGE pragmas
  because the parser for LANGUAGE has state and the state wasn't being
  saved between chunks.  We're still closing the Handle eagerly, so
  there shouldn't be any problems here.

[igloo]

Merged, and test read068 added.