You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Kamesh Jayachandran <ka...@collab.net> on 2006/09/18 10:28:48 UTC

[PATCH][merge-tracking]Only fs_fs backend should depend on sqlite(until libsvn_fs_util gets committed)

Hi All,
Find the attached patch and log.

With regards
Kamesh Jayachandran

Re: [PATCH][merge-tracking]Only fs_fs backend should depend on sqlite(until libsvn_fs_util gets committed)

Posted by Kamesh Jayachandran <ka...@collab.net>.
>> Max's note is fairly clear:
>>
>> # This is a workaround for a libtool bug, which manifests on Linux 
>> and similar
>> # ELF platforms, when linking to an installed Neon, and there are old
>> # Subversion libraries in the same directory as the installed Neon.
>>
>> As noted there, if you link to an already-installed Neon (i.e. you're
>> _not_ using an in-tree version), libtool will place the path to
>> the installed Neon before the path to the not-yet-installed copies
>> of the Subversion libraries.  Since the Subversion libraries are
>> usually installed in the same place as Neon, this causes the resulting
>> executables to pick up the installed Subversion libraries rather than
>> the just-compiled ones.
>>
>> This bug in libtool can be worked around by working out the transitive
>> dependencies ourselves, since that the only point where the bug occurs.
>>
>>  
>>> If it was the problem it should be the problem with every other 
>>> shared lib right?
>>> Like 'bdb' et al. we did not have any such dependency of 'bdb' with 
>>> each binary with a main function.
>>>     
>>
>> You're right, from the explanation above, it should also be a problem
>> with BDB.  I can't explain why it's not.
>>
>>  
>>> From my bare understanding of the issue, it might have got to do 
>>> with 'neon' dir under $abs_src_dir of subversion kind of build.
>>> As sqlite is never entertained from the $abs_srcdir of subversion we 
>>> need not bother about this.
>>>
>>>     
>>
>> I don't think that it necessarily involves a source build of Neon -
>> quite the opposite.
>>
>> Regards,
>> Malcolm
>>
>>   
> I could reproduce the same problem with my libtool 1.5.6.
> And I have a different solution to the issue, I am working on seeings 
> its effectiveness in different cases.
> Basically moving the libsvn_ra to end of libs fixes the issue.
> I am just experimenting with many such indirect 3rd party lib 
> references simultaneously by a 'exe'.
>
> With regards
> Kamesh Jayachandran
Even our current merge-tracking branch 'lt-svn' binary fails in 
convoluted setups as we currently don't have 'sqlite' in the libs line 
of 'svn'.(setup having 'sqlite' in the same dir as the old version of 
svn). I am proposing the different fix to the problem in a different thread.


With regards
Kamesh Jayachandran

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [PATCH][merge-tracking]Only fs_fs backend should depend on sqlite(until libsvn_fs_util gets committed)

Posted by Kamesh Jayachandran <ka...@collab.net>.
Malcolm Rowe wrote:
> On Tue, Sep 19, 2006 at 07:03:30PM +0530, Kamesh Jayachandran wrote:
>   
>> I failed to understand the neon issue here, I did not do any research on 
>> it yet.
>>     
>
> Max's note is fairly clear:
>
> # This is a workaround for a libtool bug, which manifests on Linux and similar
> # ELF platforms, when linking to an installed Neon, and there are old
> # Subversion libraries in the same directory as the installed Neon.
>
> As noted there, if you link to an already-installed Neon (i.e. you're
> _not_ using an in-tree version), libtool will place the path to
> the installed Neon before the path to the not-yet-installed copies
> of the Subversion libraries.  Since the Subversion libraries are
> usually installed in the same place as Neon, this causes the resulting
> executables to pick up the installed Subversion libraries rather than
> the just-compiled ones.
>
> This bug in libtool can be worked around by working out the transitive
> dependencies ourselves, since that the only point where the bug occurs.
>
>   
>> If it was the problem it should be the problem with every other shared 
>> lib right?
>> Like 'bdb' et al. we did not have any such dependency of 'bdb' with each 
>> binary with a main function.
>>     
>
> You're right, from the explanation above, it should also be a problem
> with BDB.  I can't explain why it's not.
>
>   
>> From my bare understanding of the issue, it might have got to do with 
>> 'neon' dir under $abs_src_dir of subversion kind of build.
>> As sqlite is never entertained from the $abs_srcdir of subversion we 
>> need not bother about this.
>>
>>     
>
> I don't think that it necessarily involves a source build of Neon -
> quite the opposite.
>
> Regards,
> Malcolm
>
>   
I could reproduce the same problem with my libtool 1.5.6.
And I have a different solution to the issue, I am working on seeings 
its effectiveness in different cases.
Basically moving the libsvn_ra to end of libs fixes the issue.
I am just experimenting with many such indirect 3rd party lib references 
simultaneously by a 'exe'.

With regards
Kamesh Jayachandran

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [PATCH][merge-tracking]Only fs_fs backend should depend on sqlite(until libsvn_fs_util gets committed)

Posted by Malcolm Rowe <ma...@farside.org.uk>.
On Tue, Sep 19, 2006 at 07:03:30PM +0530, Kamesh Jayachandran wrote:
> I failed to understand the neon issue here, I did not do any research on 
> it yet.

Max's note is fairly clear:

# This is a workaround for a libtool bug, which manifests on Linux and similar
# ELF platforms, when linking to an installed Neon, and there are old
# Subversion libraries in the same directory as the installed Neon.

As noted there, if you link to an already-installed Neon (i.e. you're
_not_ using an in-tree version), libtool will place the path to
the installed Neon before the path to the not-yet-installed copies
of the Subversion libraries.  Since the Subversion libraries are
usually installed in the same place as Neon, this causes the resulting
executables to pick up the installed Subversion libraries rather than
the just-compiled ones.

This bug in libtool can be worked around by working out the transitive
dependencies ourselves, since that the only point where the bug occurs.

> If it was the problem it should be the problem with every other shared 
> lib right?
> Like 'bdb' et al. we did not have any such dependency of 'bdb' with each 
> binary with a main function.

You're right, from the explanation above, it should also be a problem
with BDB.  I can't explain why it's not.

> From my bare understanding of the issue, it might have got to do with 
> 'neon' dir under $abs_src_dir of subversion kind of build.
> As sqlite is never entertained from the $abs_srcdir of subversion we 
> need not bother about this.
> 

I don't think that it necessarily involves a source build of Neon -
quite the opposite.

Regards,
Malcolm

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [PATCH][merge-tracking]Only fs_fs backend should depend on sqlite(until libsvn_fs_util gets committed)

Posted by Kamesh Jayachandran <ka...@collab.net>.
Malcolm Rowe wrote:
> On Mon, Sep 18, 2006 at 03:58:48PM +0530, Kamesh Jayachandran wrote:
>   
>> * build.conf
>>   (svnadmin): Should not have a direct dependency with sqlite.
>>   (svndumpfilter): Should not have a direct dependency with sqlite.
>>   (svnlook): Should not have a direct dependency with sqlite.
>>   (libsvn_fs): Should not have a direct dependency with sqlite.
>>   (libsvn_fs_base): Should not have a direct dependency with sqlite.
>>     
>
> I assume those extra sqlite dependencies aren't there to avoid a sqlite
> version of the 'neon problem' that's documented early in build.conf?
>
> Regards,
> Malcolm
>
>   
I failed to understand the neon issue here, I did not do any research on 
it yet.
If it was the problem it should be the problem with every other shared 
lib right?
Like 'bdb' et al. we did not have any such dependency of 'bdb' with each 
binary with a main function.
 From my bare understanding of the issue, it might have got to do with 
'neon' dir under $abs_src_dir of subversion kind of build.
As sqlite is never entertained from the $abs_srcdir of subversion we 
need not bother about this.

With regards
Kamesh Jayachandran

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [PATCH][merge-tracking]Only fs_fs backend should depend on sqlite(until libsvn_fs_util gets committed)

Posted by Kamesh Jayachandran <ka...@collab.net>.
Malcolm Rowe wrote:
> On Mon, Sep 18, 2006 at 03:58:48PM +0530, Kamesh Jayachandran wrote:
>   
>> * build.conf
>>   (svnadmin): Should not have a direct dependency with sqlite.
>>   (svndumpfilter): Should not have a direct dependency with sqlite.
>>   (svnlook): Should not have a direct dependency with sqlite.
>>   (libsvn_fs): Should not have a direct dependency with sqlite.
>>   (libsvn_fs_base): Should not have a direct dependency with sqlite.
>>     
>
> I assume those extra sqlite dependencies aren't there to avoid a sqlite
> version of the 'neon problem' that's documented early in build.conf?
>
> Regards,
> Malcolm
>
>   
Thanks for the pointer, I have a tough time in understanding it :).

With regards
Kamesh Jayachandran

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [PATCH][merge-tracking]Only fs_fs backend should depend on sqlite(until libsvn_fs_util gets committed)

Posted by Malcolm Rowe <ma...@farside.org.uk>.
On Mon, Sep 18, 2006 at 03:58:48PM +0530, Kamesh Jayachandran wrote:
> * build.conf
>   (svnadmin): Should not have a direct dependency with sqlite.
>   (svndumpfilter): Should not have a direct dependency with sqlite.
>   (svnlook): Should not have a direct dependency with sqlite.
>   (libsvn_fs): Should not have a direct dependency with sqlite.
>   (libsvn_fs_base): Should not have a direct dependency with sqlite.

I assume those extra sqlite dependencies aren't there to avoid a sqlite
version of the 'neon problem' that's documented early in build.conf?

Regards,
Malcolm

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org