ghci segmentation fault when computation is interrupted

[fft1976]

This is using Ubuntu 8.04. Both source-compiled and binary (libedit2) show this behavior.

GHCi, version 6.10.2: http://www.haskell.org/ghc/  :? for help
Loading package ghc-prim ... linking ... done.
Loading package integer ... linking ... done.
Loading package base ... linking ... done.
Prelude> let a = a
Prelude> a

Then I press Ctrl-C

Segmentation fault
$ 

Potentially relevant:

$ dpkg -l | grep libedit
ii  libedit-dev                                2.9.cvs.20050518-4                                   BSD editline and history libraries (developm
ii  libedit2                                   2.9.cvs.20050518-4                                   BSD editline and history libraries

[fft1976]

Actually

let a = a
a

isn't necessary. Ctrl-C causes Segfault regardless.

I should also add that the installation didn't go very smoothly: rts and other error messages. See follow-up messages in #3151

[int-e]

If this is on x86, libedit is responsible for the segfault. It installs its own SIGINT signal handler using signal() and then reinstalls the RTS signal handler using signal again. This causes the SA_SIGINFO flag to be lost. As a result, the RTS' signal handler does not receive the siginfo_t data that it needs. I have a patch for editline, but I haven't reported this anywhere yet.

N.B. This does not happen in x86_64 because there the siginfo_t is always passed to the signal handler.

[int-e]

Patch against libedit-20090111-3.0 from  http://www.thrysoee.dk/editline/; it applies to Debian's 20050518 version as well (using patch -p2 in the libedit-20050518 subdirectory), but I haven't tested the result.

[benl]

Pressing Ctrl-C in ghci on sparc/solaris also causes a segfault.

[igloo]

6.12 won't use libedit, so we only need to worry about this if we do a 6.10.3 release.

[cmcq]

See #2812. I don't get a segfault in 6.10.3.