Building packages using ghc (6.8.2) results in inconsistent error assembler messages

[akrohit]

akrohit@illusion:~/tmp/ghc-6.8.2$ uname -a
Linux illusion 2.6.27-11-generic #1 SMP Thu Jan 29 19:28:32 UTC 2009 x86_64 GNU/Linux

akrohit@illusion:~/tmp/ghc-6.8.2$ gcc --version
gcc (GCC) 4.1.3 20080623 (prerelease) (Ubuntu 4.1.2-23ubuntu3)

akrohit@illusion:~/tmp/ghc-6.8.2$ ghc --version
The Glorious Glasgow Haskell Compilation System, version 6.8.2

Whenever installing any package using cabal or building packages locally results in some error messages of something like :

/tmp/ghc10273_0/ghc10273_0.s: Assembler messages:

/tmp/ghc10273_0/ghc10273_0.s:2502:0:
     Error: junk at end of line, first unrecognized character is `8'

/tmp/ghc10273_0/ghc10273_0.s:2830:0:
     Error: unknown pseudo-op: `.aligo'
<<ghc: 188196056 bytes, 25 GCs, 2710528/5779456 avg/max bytes residency (4 samples), 20M in use, 0.00 INIT (0.00 elapsed), 0.17 MUT (0.19 elapsed), 0.08 GC (0.10 elapsed) :ghc>>
make[2]: *** [Distribution/Compiler.o] Error 1
make[1]: *** [boot] Error 2
make: *** [stage1] Error 1

These error messages are not consistent. Above also results with gcc version 4.3.1

[igloo]

Hmm, curious. Are you able to test with 6.10.2?

If you try to compile a simple file, e.g.

main = putStrLn "Foo"

with

ghc -O -c test.hs

then do you get the error?

[akrohit]

Compiler output

[akrohit]

I just tested it with ghc-6.10.2 and is still reproducible.

The interesting part is that the bus does not occur with files containing simple programs like the above one (test.hs).

But while compiling any standard hackage package or any other complex program results in the error.

I am attaching a sample output of the bug which might help. (produced with ghc-6.10.2)

The lines marked with ****** are my comments

I am also attaching the portion of temporary file (as the original file is too big) that is reported in the error message. I have marked the first few error by giving a comment at the proper line no. To go to the error line just search the string "error" in file named ghc2061_0.s

Thought it might help.

[igloo]

OK, so some characters with least significant bit 0 are getting their least significant bit set to 1. We have d -> e, z -> { and : -> ;. I'd guess that this is the Ubuntu 6.8.2 and the 6.10.2 bindist, both of which work for other people, so it smells like a hardware problem to me. Have you tried running memtest or similar?

[akrohit]

Earlier I was using Fedora (Currently ubuntu 8.10) and the same errors were coming. So I guess it has got not much to do with any particular linux distro.

Moreover I just did a memtest and it passed it without any errors.

I have also used gcc for compiling other big large projects without any problem, moreover my system runs fine. I mean till yet I haven't seen any signs that might conclude to some hardware issues.

[igloo]

So just to check:

• You've only seen this on one computer
• You've seen it with 2 different OS installs
• You've seen it with GHC 6.8.2 and 6.10.2
• You've been using binary packages of GHC, not building GHC yourself

Is that all correct?

[akrohit]

Ya.

I tried building ghc on my system but faced the same errors so am using binary ghc packages.

Just one more thing to add. The errors I get is only when I try to create an executable. They all get loaded in the ghci (interpreter) correctly.

[simonmar]

Replying to igloo:

So just to check: * You've only seen this on one computer * You've seen it with 2 different OS installs * You've seen it with GHC 6.8.2 and 6.10.2 * You've been using binary packages of GHC, not building GHC yourself

Also

• The error is different each time, not repeatable

correct?

All the evidence points towards a hardware failure, I'm afraid. Do you have another machine you could try GHC on?

[akrohit]

Ya correct. The error messages are inconsistent.

I have another system on which everything runs fine.

[igloo]

OK, sounds like hardware problems, then.