Testing GCC-4.5.3 on mips-sgi-IRIX6.5

Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Callum Massey]

On 24/08/2011 03:26, Callum Massey wrote:
>> hi all,
>> I've been trying to build GCC on irix and while I seem to have one 
>> that works it fails a great deal of the test suite, so many even that 
>> the mail is too big for the test suite mailing list and gets 
>> rejected, so, here it is.
>> http://sucs.org/~kais58/results
>> I was hoping that someone from this could see where I'm going wrong 
>> and missed something out somewhere.
>>
>> GNU as/binutils 2.21.1
>> SGI ld 7.4
>> Previous build of GCC-4.5.3 to compile it
>> mpfr-3.0.1
>> gmp-5.0.2
>> mpc-0.9
>> GNU make 3.8.2
>>
>> thanks,
>> Callum, ~kais58
>

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Rainer Orth]

Hi Callum,

>     I've been trying to build GCC on irix and while I seem to have one that
>     works it fails a great deal of the test suite, so many even that the
>     mail is too big for the test suite mailing list and gets rejected, so,
>     here it is.
>      http://sucs.org/~kais58/results
>     I was hoping that someone from this could see where I'm going wrong and
>     missed something out somewhere.

Those results look really bad, but I assume some systematic error.  To
investigate, one would need to look at e.g. gcc/testsuite/gcc/gcc.log to
see how exactly the tests are failing.

>     GNU as/binutils 2.21.1
>     SGI ld 7.4

Those should be ok, that what I'm using myself all the time.  Which
exact IRIX 6.5 build do you use (/bin/uname -R)?

>     Previous build of GCC-4.5.3 to compile it
>     mpfr-3.0.1
>     gmp-5.0.2
>     mpc-0.9

This might cause problems: I'd stay with the exact versions mentioned in
the install docs, since I've often seens serious problems with older
version.

>     GNU make 3.8.2

I'm staying with 3.81, but 3.82 isn't so much worse hopefully.

Another point: please try gcc 4.6.1 instead.  I've had serious problems
with the 4.5 branch, with bootstraps often aborting somewhere in the
middle with spurious SEGVs.  The same commands work fine if you restart
the bootstrap from there.  I'd usually attribute this to flakey
hardware, but the same machine is completely stable for me for 4.6 and
4.7 bootstraps.

Btw., which versions of tcl, tk and dejagnu do you use?  I have tcl 8.4,
expect 5.43.0, and dejagnu 1.4.4.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Callum Massey]

On 24/08/2011 17:02, Rainer Orth wrote:
> Hi Callum,
>
>>      I've been trying to build GCC on irix and while I seem to have one that
>>      works it fails a great deal of the test suite, so many even that the
>>      mail is too big for the test suite mailing list and gets rejected, so,
>>      here it is.
>>       http://sucs.org/~kais58/results
>>      I was hoping that someone from this could see where I'm going wrong and
>>      missed something out somewhere.
> Those results look really bad, but I assume some systematic error.  To
> investigate, one would need to look at e.g. gcc/testsuite/gcc/gcc.log to
> see how exactly the tests are failing.
https://sucs.org/~kais58/testsuite/ <https://sucs.org/%7Ekais58/testsuite/>
The entire testsuite directory is here
<https://sucs.org/%7Ekais58/testsuite/>
>
>>      GNU as/binutils 2.21.1
>>      SGI ld 7.4
> Those should be ok, that what I'm using myself all the time.  Which
> exact IRIX 6.5 build do you use (/bin/uname -R)?
/bin/uname -R
6.5 6.5.30m
Courtesy of SGI.
>>      Previous build of GCC-4.5.3 to compile it
>>      mpfr-3.0.1
>>      gmp-5.0.2
>>      mpc-0.9
> This might cause problems: I'd stay with the exact versions mentioned in
> the install docs, since I've often seens serious problems with older
> version.
I'll rebuild these and then use them to build gcc-4.6.1 now, do you mean 
the minimum required versions ie 2.4.2, 4.3.2, 0.8.1?
>>      GNU make 3.8.2
> I'm staying with 3.81, but 3.82 isn't so much worse hopefully.
>
> Another point: please try gcc 4.6.1 instead.  I've had serious problems
> with the 4.5 branch, with bootstraps often aborting somewhere in the
> middle with spurious SEGVs.  The same commands work fine if you restart
> the bootstrap from there.  I'd usually attribute this to flakey
> hardware, but the same machine is completely stable for me for 4.6 and
> 4.7 bootstraps.
>
> Btw., which versions of tcl, tk and dejagnu do you use?  I have tcl 8.4,
> expect 5.43.0, and dejagnu 1.4.4.
tcl/tk-8.5.10, expect-5.45, dejgnu-1.5 currently
>
> 	Rainer
>
Callum

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Rainer Orth]

Hi Callum,

>> Those results look really bad, but I assume some systematic error.  To
>> investigate, one would need to look at e.g. gcc/testsuite/gcc/gcc.log to
>> see how exactly the tests are failing.
> https://sucs.org/~kais58/testsuite/ <https://sucs.org/%7Ekais58/testsuite/>
> The entire testsuite directory is here
> <https://sucs.org/%7Ekais58/testsuite/>

thanks.  I see that the fast majority of failures is with

compiler exited with status -1
output is:
spawn failed

I saw them myself at a time, and under par found

11260mS          expect(517760): open("/dev/ptmx", O_RDWR, 037777777777) = 8
11261mS          expect(517760): stat("/dev/pts/2", 0x7ffae0f0) errno = 2 (No such file or directory)

For some reason, ptys weren't created on demand, but had to be created
manually since /dev/MAKEDEV had no support for that.  I never found out
what was wrong, but the problem hasn't bitten me since.

>>>      Previous build of GCC-4.5.3 to compile it
>>>      mpfr-3.0.1
>>>      gmp-5.0.2
>>>      mpc-0.9
>> This might cause problems: I'd stay with the exact versions mentioned in
>> the install docs, since I've often seens serious problems with older
>> version.
> I'll rebuild these and then use them to build gcc-4.6.1 now, do you mean
> the minimum required versions ie 2.4.2, 4.3.2, 0.8.1?

Exactly.  At least on Tru64 UNIX, newer versions even fail their own
tests, and we had similar reports e.g. on SPARC.

>> Btw., which versions of tcl, tk and dejagnu do you use?  I have tcl 8.4,
>> expect 5.43.0, and dejagnu 1.4.4.
> tcl/tk-8.5.10, expect-5.45, dejgnu-1.5 currently

Those should be fine.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Callum Massey]

On 08/24/2011 06:00 PM, Rainer Orth wrote:
> thanks.  I see that the fast majority of failures is with
>
> compiler exited with status -1
> output is:
> spawn failed
>
> I saw them myself at a time, and under par found
>
> 11260mS          expect(517760): open("/dev/ptmx", O_RDWR, 037777777777) = 8
> 11261mS          expect(517760): stat("/dev/pts/2", 0x7ffae0f0) errno = 2 (No such file or directory)
>
> For some reason, ptys weren't created on demand, but had to be created
> manually since /dev/MAKEDEV had no support for that.  I never found out
> what was wrong, but the problem hasn't bitten me since.
>
>> I'll rebuild these and then use them to build gcc-4.6.1 now, do you mean
>> the minimum required versions ie 2.4.2, 4.3.2, 0.8.1?
> Exactly.  At least on Tru64 UNIX, newer versions even fail their own
> tests, and we had similar reports e.g. on SPARC.
I've built 4.6.1 with these and uploaded the testsuite directory again.
http://sucs.org/~kais58/testsuite/
> Those should be fine.
>
> 	Rainer
>
Callum

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Rainer Orth]

Callum,

> On 08/24/2011 06:00 PM, Rainer Orth wrote:
>> thanks.  I see that the fast majority of failures is with
>>
>> compiler exited with status -1
>> output is:
>> spawn failed
>>
>> I saw them myself at a time, and under par found
>>
>> 11260mS          expect(517760): open("/dev/ptmx", O_RDWR, 037777777777) = 8
>> 11261mS          expect(517760): stat("/dev/pts/2", 0x7ffae0f0) errno = 2 (No such file or directory)
>>
>> For some reason, ptys weren't created on demand, but had to be created
>> manually since /dev/MAKEDEV had no support for that.  I never found out
>> what was wrong, but the problem hasn't bitten me since.
>>
>>> I'll rebuild these and then use them to build gcc-4.6.1 now, do you mean
>>> the minimum required versions ie 2.4.2, 4.3.2, 0.8.1?
>> Exactly.  At least on Tru64 UNIX, newer versions even fail their own
>> tests, and we had similar reports e.g. on SPARC.
> I've built 4.6.1 with these and uploaded the testsuite directory again.
> http://sucs.org/~kais58/testsuite/

Results look as bad as before, although it's far easier to see from the

        make mail-report.log

output.

Did you create the /dev/pts/* devices before the testsuite run?

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Callum Massey]

On 08/25/2011 03:07 PM, Rainer Orth wrote:
> Callum,
>
>> On 08/24/2011 06:00 PM, Rainer Orth wrote:
>>> thanks.  I see that the fast majority of failures is with
>>>
>>> compiler exited with status -1
>>> output is:
>>> spawn failed
>>>
>>> I saw them myself at a time, and under par found
>>>
>>> 11260mS          expect(517760): open("/dev/ptmx", O_RDWR, 037777777777) = 8
>>> 11261mS          expect(517760): stat("/dev/pts/2", 0x7ffae0f0) errno = 2 (No such file or directory)
>>>
>>> For some reason, ptys weren't created on demand, but had to be created
>>> manually since /dev/MAKEDEV had no support for that.  I never found out
>>> what was wrong, but the problem hasn't bitten me since.
>>>
>>>> I'll rebuild these and then use them to build gcc-4.6.1 now, do you mean
>>>> the minimum required versions ie 2.4.2, 4.3.2, 0.8.1?
>>> Exactly.  At least on Tru64 UNIX, newer versions even fail their own
>>> tests, and we had similar reports e.g. on SPARC.
>> I've built 4.6.1 with these and uploaded the testsuite directory again.
>> http://sucs.org/~kais58/testsuite/
> Results look as bad as before, although it's far easier to see from the
>
>          make mail-report.log
>
> output.
>
> Did you create the /dev/pts/* devices before the testsuite run?
No, I wasn't certain how to do it. Do I just need to run /dev/MAKEDEV?
> 	Rainer
>
Callum

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Rainer Orth]

Callum,

>> Did you create the /dev/pts/* devices before the testsuite run?
> No, I wasn't certain how to do it. Do I just need to run /dev/MAKEDEV?

unfortunately not: those devices are supposed to be created by
/usr/sbin/mkpts, as described in getpty(3).  I have no idea why this
doesn't happen, but you'll have to do it manually as root, using mknod:

# cd /dev/pts
# mknod <N> c 15 <N>
# chmod 620 <N>
# chown <user> sys

for all missing N (I've created them up to 127).

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Callum Massey]

On 25/08/2011 15:21, Rainer Orth wrote:
> Callum,
>
>>> Did you create the /dev/pts/* devices before the testsuite run?
>> No, I wasn't certain how to do it. Do I just need to run /dev/MAKEDEV?
> unfortunately not: those devices are supposed to be created by
> /usr/sbin/mkpts, as described in getpty(3).  I have no idea why this
> doesn't happen, but you'll have to do it manually as root, using mknod:
>
> # cd /dev/pts
> # mknod<N>  c 15<N>
> # chmod 620<N>
> # chown<user>  sys
>
> for all missing N (I've created them up to 127).
>
> 	Rainer
>
Made the pts' and re-ran make check, results here, 
http://sucs.org/~kais58/results

Callum

Re: Testing GCC-4.5.3 on mips-sgi-IRIX6.5

[Rainer Orth]

Callum,

> Made the pts' and re-ran make check, results here,
> http://sucs.org/~kais58/results

those results are mostly in line with my own, with one exception: you
shouldn't try testing the O32 ABI (-mabi=32) any longer.  Support for
that multilib has been removed in GCC 4.6.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University