Evil Mangler hanging while building 5.04.3-1 on RedHat 8.0, 9.0?

Simon Marlow simonmar at microsoft.com
Tue Apr 22 05:30:11 EDT 2003


 
> This problem arose for me on RedHat 8.0, and now it's doing 
> it on RedHat 
> 9.0.  However, it only ever arose on the same machine (my 
> RedHat 8.0 laptop 
> has no trouble compiling GHC).  The fact that it always 
> happens with the 
> same files regardless
> 
> When compiling ghc/compiler/absCSyn/AbsCUtils.lhs in stage-1, and 
> ghc/compiler/parser/Parser.hs and 
> libraries/base/Foreign/C/Error.hs in 
> stage-2 (i.e., with ghc-inplace), my machine hangs (it's not 
> thrashing, it 
> hangs: no change after an hour, no cursor blinking, no 
> response to any 
> stimuli; it even takes a few pushes of the reset button to 
> have any effect some times).

Wow!  "GHC makes my machine hang"?  Surely this is a Linux problem, not
a GHC problem?

> (There may be other files downstream that exhibit this 
> behavior, but I 
> haven't gotten around to pushing the process past 
> libraries/base/Foreign/C/Error.hs yet.)
> 
> I suspect the Evil Mangler because I can successfully compile 
> each of these 
> modules using -Onot and setting -tmpdir to be a local 
> directory rather than 
> /tmp.  This guarantees that the NCG is used (*), and so the 
> Evil Mangler doesn't come into play.

Using -fasm forces the NCG too.

> I've heard of other isolated cases of this behavior, but 
> never heard an 
> explanation.  Does anyone know what's going on?  Has anyone else 
> experienced this behavior?

Nothing springs to mind, I'm afraid.  One possibility is an accidental
DOS (such as a fork bomb), but usually the limits imposed by the OS
prevent this sort of thing.

> (*) Why is ghc/compiler/parser/Parser.hs compiled with -fviaC set 
> explicitly?  The NCG seems to cope just fine.  Is this a 
> hangover from bad 
> old days when the NCG couldn't handle Happy-generated files?

Possibly a hangover, yes.  We support building the latest GHC using any
GHC >= 4.08, which had some areas of incompleteness in the NCG, so
removing these requires testing against 4.08.

Cheers,
	Simon



More information about the Glasgow-haskell-bugs mailing list