Re: [Gm2] setjmp/longjmp Issues on Debian/Alpha

[Gaius Mulley]

Michael Lambert <address@hidden> writes:

> Hi Gaius,
>
> The test iso/run/pass/builtinlj.mod fails on Debian/Alpha with
>
> -----
> call longjmp
> Segmentation fault
> -----
>
> My first thought was that the buffer for setjmp/longjmp was too small,
> so I replaced
>
> env: ARRAY [0..5] OF WORD ;
>
> by
>
> env: ARRAY [0..32767] OF CHAR ;
>
> with the same result.
>
> Then I thought it could be an alignment issue, so instead of calling
> setjmp and longjmp with an argument of ADR(env) I tried ADR(env[i])
> for 0 <= i < 8.  That still gave exactly the same results.
>
> Any suggestions for tracking this down?
>
> Regards,
>
> Michael

Hi Michael,

This regression test also fails on x86_64 Debian Etch as well.  One
way to find the bug might be to binary chop the cvs checkout dates
since this test passed.  (All tests passed under Debian Etch x86_64 on
2008-11-04).  If this is true for the alpha then binary chopping the
cvs checkouts and then finally feeding patches in one at a time should
narrow it down - I suspect..

regards,
Gaius