On 13 September 2004 17:03, Volker Stolz wrote:
> I looked into the SF-bug [ 1012521 ] getUserEntryForID dies where a
> reentrant
> function from libc would cause an illegal instruction. First, I
> thought the
> maybe we'd forgotten to pass -D_POSIX_PTHREAD_SEMANTICS, but it's
> there after all.
> It looks like something doesn't work right when Solaris wraps the
> reentrant
> functions in the following way:
>
> <pwd.h> does something along the lines of
>
> static int
> getpwuid_r(__uid, __pwd, __buf, __len, __res)
> uid_t __uid;
> struct passwd *__pwd;
> char *__buf;
> int __len;
> struct passwd **__res;
> {
> extern int __posix_getpwuid_r();
> return (__posix_getpwuid_r(__uid, __pwd, __buf, __len,
> __res)); }
Oh bum. The header file is inserting ordinary function definitions in
the code, which means we have to tell the mangler about these. I
believe there are already examples of this - check out the
i386-*-solaris2 case just above the code for "Warning - retaining
unknown function..." in ghc-asm.lprl.
You need to ensure that these functions just get passed through the
mangler unharmed, i.e. no prologue/epilogue processing. I'm not sure
how best to do this - the mangler currently seems to do
prologue/epilogue processing for *all* code blocks on sparc. You might
have to use a new code chunk type for these functions, and avoid the
prologue/epilogue processing for those chunks.
A better solution would be to avoid all this mess somehow. Perhaps use
C wrappers for getpwuid_r, or something?
Cheers,
Simon
_______________________________________________
Cvs-ghc mailing list
[EMAIL PROTECTED]
http://www.haskell.org/mailman/listinfo/cvs-ghc
