subject:"Bug#377507\: Please unblock testing migration of postgresql\-7.4"

Bug#377507: Please unblock testing migration of postgresql-7.4

2006-07-20 Thread Thiemo Seufer

Thiemo Seufer wrote:
 Thiemo Seufer wrote:
 [snip]
Thank you for trying this out. However, I'm sceptical: postgresql-8.1
still gives the same error when built with gcc-4.1:

  
http://buildd.debian.org/fetch.php?pkg=postgresql-8.1ver=8.1.4-4arch=mipsstamp=1152713470file=logas=raw
  
http://buildd.debian.org/fetch.php?pkg=postgresql-8.1ver=8.1.4-4arch=mipselstamp=1152713545file=logas=raw

So maybe the buildds run some different package versions.
   
   I did a rebuild on unstable/mips, no failures, packages install fine,
   postmaster process runs. So the failure might be related to the buildd
   environment.
  
  It seems to be kernel related, a build on a 2.6.12 64bit kernel shows
  bus errors.
 
 The bus error happens immediately at startup when writing a double to
 unaligned address. I'll have to test if the binary was miscompiled
 (due to kernel differences?!), or if the kernel is supposed to handle
 this better.

This is a kernel bug related to FP emulation. It is fixed in 2.6.16.
Once the buildds are updated and the relevant postgres packages add
mips/mipsel again everything should build fine. For package upgrades
it might be advisable to check the version of the running kernel.


Thiemo


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]

Bug#377507: Please unblock testing migration of postgresql-7.4

2006-07-17 Thread Martin Pitt

Hi Thiemo!

Thiemo Seufer [2006-07-16 21:42 +0100]:
FWIW, I tried a build with today's unstable on mips, after enabling
the server packages again. There was one failure in the testsuite:

...
test geometry ... ok
test horology ... failed (ignored)

That's normal in 7.4, there is little I can do about it. It was fixed
in 8.0, and it's nothing to worry about.

The resulting sever packages installed fine, no obvious failure
(but OTOH I don't know what a serious test would be).

Merely passing the test suite is fine, so far it has failed right when
trying to start postmaster.

In addition, I built with gcc-4.1. Since some (mostly FFI related) ABI
bugs were fixed between 3.3 and 4.0, this can explain the difference.

I recommend to try a build without any mips/mipsel specific workarounds
(the horology failure appears to be normal, it happens also on other
architectures).

Thank you for trying this out. However, I'm sceptical: postgresql-8.1
still gives the same error when built with gcc-4.1:

http://buildd.debian.org/fetch.php?pkg=postgresql-8.1ver=8.1.4-4arch=mipsstamp=1152713470file=logas=raw

http://buildd.debian.org/fetch.php?pkg=postgresql-8.1ver=8.1.4-4arch=mipselstamp=1152713545file=logas=raw

So maybe the buildds run some different package versions. Also, bug
357603 has indicated that this was a problem with binutils, not with
gcc (if using gcc-3.3 the problem occurs, too, and in Sarge it worked
well with gcc-3.3).

I'll try another upload of p-7.4 with mipsen re-enabled, I can always
disable it quickly again if it still fails.

Thanks for your investigations!

Martin

--
Martin Pitthttp://www.piware.de
Ubuntu Developer http://www.ubuntu.com
Debian Developer http://www.debian.org

In a world without walls and fences, who needs Windows and Gates?

signature.asc
Description: Digital signature

Bug#377507: Please unblock testing migration of postgresql-7.4

2006-07-17 Thread Thiemo Seufer

Martin Pitt wrote:
Hi Thiemo!