2:22 p.m.
Simon Josefsson <jas(a)pdc.kth.se> writes:
Perhaps I should start by seeing what problems there might be
compiling it with gcc then? :-) I'll do that now.
This is Max Matveev describing the invariant his code is expecting.
Does that hold when compiling with gcc?
1999-12-28 Max Matveev <max(a)melbourne.sgi.com>
* unexelfsgi.c (unexec): Change the way we decide which segment
should be extended.
Assumption that .bss section should be outside the PT_LOADable
segment. On IRIX with version 6.2 and above, .bss (or .sbss, if
it's present) is inside the 'data' segment. This would fail the
test which was used to find a segment to grow and cover new
heap. Instead of this assumption, I created another one - on IRIX
the segment to grow should start below .bss and it's address
should extent above the end of .bss. Once this segment is
identified, it's grown to accommodate the new heap and new
zero-length .bss section is added at the end of .data2.
4:12 p.m.
New subject: cannot build xemacs-21.1.9 on mips-sgi-irix6.5
>>>> "JV" == Jan Vroonhof
<vroonhof(a)math.ethz.ch> writes:
JV> Simon Josefsson <jas(a)pdc.kth.se> writes:
> Perhaps I should start by seeing what problems there might be
> compiling it with gcc then? :-) I'll do that now.
JV> This is Max Matveev describing the invariant his code is expecting.
JV> Does that hold when compiling with gcc?
Sounds like a question of which _linker_ is used, not which compiler.
Which one gcc uses is determined by the --with-gnu-ld flag used when
configuring gcc. The system ld comes for free, so folks are more
likely to install gcc than gnu ld (part of binutils).
JV> 1999-12-28 Max Matveev <max(a)melbourne.sgi.com>
JV> * unexelfsgi.c (unexec): Change the way we decide which segment
JV> should be extended.
JV> Assumption that .bss section should be outside the PT_LOADable
JV> segment. On IRIX with version 6.2 and above, .bss (or .sbss, if
JV> it's present) is inside the 'data' segment. This would fail the
JV> test which was used to find a segment to grow and cover new
JV> heap. Instead of this assumption, I created another one - on IRIX
JV> the segment to grow should start below .bss and it's address
JV> should extent above the end of .bss. Once this segment is
JV> identified, it's grown to accommodate the new heap and new
JV> zero-length .bss section is added at the end of .data2.
1:02 a.m.
New subject: cannot build xemacs-21.1.9 on mips-sgi-irix6.5
Martin Buchholz writes:
:> >>>>> "JV" == Jan Vroonhof <vroonhof(a)math.ethz.ch>
writes:
:>
:> JV> Simon Josefsson <jas(a)pdc.kth.se> writes:
:> >> Perhaps I should start by seeing what problems there might be
:> >> compiling it with gcc then? :-) I'll do that now.
:>
:> JV> This is Max Matveev describing the invariant his code is expecting.
:> JV> Does that hold when compiling with gcc?
:>
:> Sounds like a question of which _linker_ is used, not which compiler.
:> Which one gcc uses is determined by the --with-gnu-ld flag used when
:> configuring gcc. The system ld comes for free, so folks are more
:> likely to install gcc than gnu ld (part of binutils).
FWIW, I can build 21.1.9 on Irix with MIPSpro C and system ld. So,
unless gcc does something interesting, I don't see how using gcc
front-end could cause memory corruption. How about elfdump-ing the
failing xemacs executable?
max
8:19 a.m.
New subject: cannot build xemacs-21.1.9 on mips-sgi-irix6.5
On Sunday 12 Mar 00, Martin Buchholz writes:
>>>>> "JV" == Jan Vroonhof
<vroonhof(a)math.ethz.ch> writes:
JV> This is Max Matveev describing the invariant his code is expecting.
JV> Does that hold when compiling with gcc?
Sounds like a question of which _linker_ is used, not which compiler.
Which one gcc uses is determined by the --with-gnu-ld flag used when
configuring gcc. The system ld comes for free, so folks are more
likely to install gcc than gnu ld (part of binutils).
Hi,
I'm the one who originally reported this. I should add that the gcc
build attempt (which failed) and the cc build (which succeeded) *both*
used the system ld.
However, I have found an old version of gnu ranlib 2.8 on this system.
It would have been applied during both builds. I am currently
"negotiating" with our sysadmin to have it removed, at which point
I'll try again to build with gcc, and report back.
(Not sure whether an old ranlib would be a factor, but I would like to
rule it out before wasting anyone else's time.)
Regards,
David
-------------------------------------------------------------------
David Starks-Browning | starksb(a)ebi.ac.uk
EMBL Outstation -- |
The European Bioinformatics Institute |
Wellcome Trust Genome Campus | tel: +44 (1223) 494 616
Hinxton, Cambridge, CB10 1SD, UK | fax: +44 (1223) 494 468
-------------------------------------------------------------------
1:12 p.m.
New subject: cannot build xemacs-21.1.9 on mips-sgi-irix6.5
Hi, just thought I should follow up with new information.
On Tuesday 14 Mar 00, David Starks-Browning writes:
Hi,
I'm the one who originally reported this. I should add that the gcc
build attempt (which failed) and the cc build (which succeeded) *both*
used the system ld.
However, I have found an old version of gnu ranlib 2.8 on this system.
It would have been applied during both builds. I am currently
"negotiating" with our sysadmin to have it removed, at which point
I'll try again to build with gcc, and report back.
(Not sure whether an old ranlib would be a factor, but I would like to
rule it out before wasting anyone else's time.)
I've since ruled out the old ranlib as a factor. Even after removing
the suspect ranlib, the gcc-2.95.2 build fails on mips-sgi-irix6.5.
Regards,
David
-------------------------------------------------------------------
David Starks-Browning | starksb(a)ebi.ac.uk
EMBL Outstation -- |
The European Bioinformatics Institute |
Wellcome Trust Genome Campus | tel: +44 (1223) 494 616
Hinxton, Cambridge, CB10 1SD, UK | fax: +44 (1223) 494 468
-------------------------------------------------------------------
9067
days inactive
9077
days old
4 comments
4 participants
participants (4)
-
David Starks-Browning -
Jan Vroonhof -
Martin Buchholz -
Max Matveev