[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: SVN broken since the newest version

From: Stefan Sperling <stsp_at_elego.de>
Date: Mon, 5 Jul 2010 15:01:01 +0200

On Mon, Jul 05, 2010 at 03:42:35PM +0300, Sjors Gielen wrote:
>
> Op 5 jul 2010, om 15:26 heeft Stefan Sperling het volgende geschreven:
>
> > The Subversion project itself does not maintain the Fink package.
> > So I'm afraid we're probably not the right people to file this issue with.
> > I suspect an error in the packaging, rather than a bug in Subversion.
> > Please talk to the Fink package maintainers.
>
> Hello Stefan,
>
> As said in the original message, I did talk to the Fink package
> maintainers, which recommended me to run the test suite, which failed.
> A test suite does not normally fail because of a simple packaging
> problem.

Packaging is very complicated because it involves linking
Subversion to the right libraries at build and run-time.
It is easy to overlook subtle problems during build and installation
which cause unexpected behaviour (such as random segfaults).

> Fink patches the libtool file to fix compilation, and it uses
> extensive configure flags. None of this should make so much of the
> test suite fail, let alone corrupt repositories. None of this has been
> changed, as far as I know, in the past versions; this bug also does
> not appear with some of the other SVN users on Fink I've spoken to.

If other fink users don't see it, that makes it even more
likely that you have a local problem on your machine.

> Please check the bug report itself,

I did check the bug report.

The paste-bin post linked from the bug report (http://pastebin.com/RmFVqZYd)
contains obvious failures such as:

sjors_at_smbp:~/Projecten/ruwave/svn/trunk/server$ svn status
Bus error
sjors_at_smbp:~/Projecten/ruwave/svn/trunk/server$ svn up
Segmentation fault

This means something is broken with the Subversion build and/or
installation. Most likely a problem with incompatible APR versions
being used at build time and at run time.

Subversion developers are running the test suite on MacOSX before
release. And we don't release with failing tests.

> it contains some more information, including a lead on what might be the cause... :)

You mean the hypothetical off-by-one bug you mentioned?
Or this trace?

Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00000000
0x0010108a in svn_io_get_dirents2 ()
(gdb) bt
#0 0x0010108a in svn_io_get_dirents2 ()
#1 0x000a64f6 in get_dir_status ()
#2 0x000a6e07 in close_edit ()
#3 0x00061ae7 in svn_client_status4 ()
#4 0x00011ce4 in svn_cl__status ()
#5 0x0000c312 in main ()
(gdb) quit

This is very likely a problem with the APR library, because it is
used to do i/o. If Subversion cannot even read dirents, I'm wondering
why any of the regression tests pass for you at all (maybe it works
half the time?).

I'd like to help you figure out what the problem is.
For starters, let's find out what APR your Subversion binary runs with.
Please post the output of:
  ldd `which svn`

Thanks,
Stefan
Received on 2010-07-05 15:02:05 CEST

This is an archived mail posted to the Subversion Dev mailing list.