Soname bump

Soname bump

[Andrew Haley]

There has been a soname bump to libffi.so.6.  Why was this?  Will older
packages linked against libffi.so.5 still work?  Can I perhaps just symlink
libffi.so.6 -> libffi.so.5 ?

Andrew.

Re: Soname bump

[Matthias Klose]

On 08/05/2011 03:32 PM, Andrew Haley wrote:
> There has been a soname bump to libffi.so.6.  Why was this?  Will older
> packages linked against libffi.so.5 still work?  Can I perhaps just symlink
> libffi.so.6 -> libffi.so.5 ?

symbols were removed from 5, so I assume the soname change is rectified.
However you should be able to just rebuild packages with 6, afaics the API
didn't change.

btw, is there any chance to see a final release of 3.0.10?

  Matthias

Re: Soname bump

[Anthony Green]

Andrew Haley <aph@redhat.com> writes:

> There has been a soname bump to libffi.so.6.  Why was this?  Will older
> packages linked against libffi.so.5 still work?  Can I perhaps just symlink
> libffi.so.6 -> libffi.so.5 ?

Some debug functions that have should never have been exported were
removed from production builds a few months ago (src/debug.c).  It's
highly unlikely anybody was using them, but I wanted to stick with the
rules defined in libtool-version.  Here's the ChangeLog entry...

2011-02-11  Anthony Green  <green@moxielogic.com>

	* libtool-version: Update.
	* Makefile.am (nodist_libffi_la_SOURCES): Add src/debug.c if
	FFI_DEBUG.
	(libffi_la_SOURCES): Remove src/debug.c
	(EXTRA_DIST): Add src/debug.c
	* Makefile.in: Rebuilt.

AG



>
> Andrew.

Re: Soname bump

[Andrew Haley]

On 08/07/2011 04:28 PM, Matthias Klose wrote:
> On 08/05/2011 03:32 PM, Andrew Haley wrote:
>> There has been a soname bump to libffi.so.6.  Why was this?  Will older
>> packages linked against libffi.so.5 still work?  Can I perhaps just symlink
>> libffi.so.6 -> libffi.so.5 ?
> 
> symbols were removed from 5, so I assume the soname change is rectified.

I don't understand what you mean.

Andrew.

Re: Soname bump

[Andrew Haley]

On 08/08/2011 04:48 AM, Anthony Green wrote:
> Andrew Haley <aph@redhat.com> writes:
> 
>> There has been a soname bump to libffi.so.6.  Why was this?  Will older
>> packages linked against libffi.so.5 still work?  Can I perhaps just symlink
>> libffi.so.6 -> libffi.so.5 ?
> 
> Some debug functions that have should never have been exported were
> removed from production builds a few months ago (src/debug.c).  It's
> highly unlikely anybody was using them, but I wanted to stick with the
> rules defined in libtool-version.  Here's the ChangeLog entry...
> 
> 2011-02-11  Anthony Green  <green@moxielogic.com>
> 
> 	* libtool-version: Update.
> 	* Makefile.am (nodist_libffi_la_SOURCES): Add src/debug.c if
> 	FFI_DEBUG.
> 	(libffi_la_SOURCES): Remove src/debug.c
> 	(EXTRA_DIST): Add src/debug.c
> 	* Makefile.in: Rebuilt.

Oh, I see.  This is a PITA because we need a new libffi for ARM Fedora 15 --
the old libffi doesn't work -- but we can't change the soname in the
Fedora package.  I'm not sure exactly what to do.

Andrew.

Re: Soname bump

[Anthony Green]

Andrew Haley <aph@redhat.com> writes:

> On 08/08/2011 04:48 AM, Anthony Green wrote:
>> Andrew Haley <aph@redhat.com> writes:
>> 
>>> There has been a soname bump to libffi.so.6.  Why was this?  Will older
>>> packages linked against libffi.so.5 still work?  Can I perhaps just symlink
>>> libffi.so.6 -> libffi.so.5 ?
>> 
>> Some debug functions that have should never have been exported were
>> removed from production builds a few months ago (src/debug.c).  It's
>> highly unlikely anybody was using them, but I wanted to stick with the
>> rules defined in libtool-version.  Here's the ChangeLog entry...
>> 
>> 2011-02-11  Anthony Green  <green@moxielogic.com>
>> 
>> 	* libtool-version: Update.
>> 	* Makefile.am (nodist_libffi_la_SOURCES): Add src/debug.c if
>> 	FFI_DEBUG.
>> 	(libffi_la_SOURCES): Remove src/debug.c
>> 	(EXTRA_DIST): Add src/debug.c
>> 	* Makefile.in: Rebuilt.
>
> Oh, I see.  This is a PITA because we need a new libffi for ARM Fedora 15 --
> the old libffi doesn't work -- but we can't change the soname in the
> Fedora package.  I'm not sure exactly what to do.

What's the issue with bumping the soname?  Can't you just rebuild
dependencies?

AG

Re: Soname bump

[Andrew Haley]

On 08/15/2011 04:36 PM, Anthony Green wrote:
> Andrew Haley <aph@redhat.com> writes:
> 
>> On 08/08/2011 04:48 AM, Anthony Green wrote:
>>> Andrew Haley <aph@redhat.com> writes:
>>>
>>>> There has been a soname bump to libffi.so.6.  Why was this?  Will older
>>>> packages linked against libffi.so.5 still work?  Can I perhaps just symlink
>>>> libffi.so.6 -> libffi.so.5 ?
>>>
>>> Some debug functions that have should never have been exported were
>>> removed from production builds a few months ago (src/debug.c).  It's
>>> highly unlikely anybody was using them, but I wanted to stick with the
>>> rules defined in libtool-version.  Here's the ChangeLog entry...
>>>
>>> 2011-02-11  Anthony Green  <green@moxielogic.com>
>>>
>>> 	* libtool-version: Update.
>>> 	* Makefile.am (nodist_libffi_la_SOURCES): Add src/debug.c if
>>> 	FFI_DEBUG.
>>> 	(libffi_la_SOURCES): Remove src/debug.c
>>> 	(EXTRA_DIST): Add src/debug.c
>>> 	* Makefile.in: Rebuilt.
>>
>> Oh, I see.  This is a PITA because we need a new libffi for ARM Fedora 15 --
>> the old libffi doesn't work -- but we can't change the soname in the
>> Fedora package.  I'm not sure exactly what to do.
> 
> What's the issue with bumping the soname?  Can't you just rebuild
> dependencies?

Yes, but that will mean that packages will be broken by the libffi update,
so the Fedora update mechanism will have to push new versions to hundreds
of thousands (millions?) of users.  If I change the libffi package I don't
think there's any way to confine the soname bump to ARM, even though only
ARM needs it.

Andrew.

Re: Soname bump

[Anthony Green]

Andrew Haley <aph@redhat.com> writes:

> Yes, but that will mean that packages will be broken by the libffi update,
> so the Fedora update mechanism will have to push new versions to hundreds
> of thousands (millions?) of users.  If I change the libffi package I don't
> think there's any way to confine the soname bump to ARM, even though only
> ARM needs it.

Andrew and I just discussed this on IRC, the result of which is that I
just reversed the ABI changing fix from the git libffi repo, and will
try to push out 3.0.10 this week in order to publish a Fedora 15 update.

Then I'll immediately re-apply this patch to the 3.0.11 train for F16.

This also means I'll be going through all of the libffi mail I haven't
answered in recent months.  Thank you for your continued patience!

#libffi on freenode will be a good place to find me this week!

AG

Re: Soname bump

[Matthias Klose]

On 08/15/2011 06:51 PM, Andrew Haley wrote:
> On 08/15/2011 04:36 PM, Anthony Green wrote:
>> Andrew Haley <aph@redhat.com> writes:
>>
>>> On 08/08/2011 04:48 AM, Anthony Green wrote:
>>>> Andrew Haley <aph@redhat.com> writes:
>>>>
>>>>> There has been a soname bump to libffi.so.6.  Why was this?  Will older
>>>>> packages linked against libffi.so.5 still work?  Can I perhaps just symlink
>>>>> libffi.so.6 -> libffi.so.5 ?
>>>>
>>>> Some debug functions that have should never have been exported were
>>>> removed from production builds a few months ago (src/debug.c).  It's
>>>> highly unlikely anybody was using them, but I wanted to stick with the
>>>> rules defined in libtool-version.  Here's the ChangeLog entry...
>>>>
>>>> 2011-02-11  Anthony Green  <green@moxielogic.com>
>>>>
>>>> 	* libtool-version: Update.
>>>> 	* Makefile.am (nodist_libffi_la_SOURCES): Add src/debug.c if
>>>> 	FFI_DEBUG.
>>>> 	(libffi_la_SOURCES): Remove src/debug.c
>>>> 	(EXTRA_DIST): Add src/debug.c
>>>> 	* Makefile.in: Rebuilt.
>>>
>>> Oh, I see.  This is a PITA because we need a new libffi for ARM Fedora 15 --
>>> the old libffi doesn't work -- but we can't change the soname in the
>>> Fedora package.  I'm not sure exactly what to do.
>>
>> What's the issue with bumping the soname?  Can't you just rebuild
>> dependencies?
> 
> Yes, but that will mean that packages will be broken by the libffi update,
> so the Fedora update mechanism will have to push new versions to hundreds
> of thousands (millions?) of users.  If I change the libffi package I don't
> think there's any way to confine the soname bump to ARM, even though only
> ARM needs it.

You should be able to rebuild the packages; I didn't see any necessary source
changes. However not bumping the soname can result in undefined symbols in
existing binaries.

  Matthias

Re: Soname bump

[Matthias Klose]

On 08/15/2011 07:35 PM, Anthony Green wrote:
> Andrew Haley <aph@redhat.com> writes:
> 
>> Yes, but that will mean that packages will be broken by the libffi update,
>> so the Fedora update mechanism will have to push new versions to hundreds
>> of thousands (millions?) of users.  If I change the libffi package I don't
>> think there's any way to confine the soname bump to ARM, even though only
>> ARM needs it.
> 
> Andrew and I just discussed this on IRC, the result of which is that I
> just reversed the ABI changing fix from the git libffi repo, and will
> try to push out 3.0.10 this week in order to publish a Fedora 15 update.
> 
> Then I'll immediately re-apply this patch to the 3.0.11 train for F16.

and having two libffi5's with a different set of symbols?

Re: Soname bump

[Anthony Green]

Matthias Klose <doko@ubuntu.com> writes:

> On 08/15/2011 07:35 PM, Anthony Green wrote:
>> Andrew Haley <aph@redhat.com> writes:
>> 
>>> Yes, but that will mean that packages will be broken by the libffi update,
>>> so the Fedora update mechanism will have to push new versions to hundreds
>>> of thousands (millions?) of users.  If I change the libffi package I don't
>>> think there's any way to confine the soname bump to ARM, even though only
>>> ARM needs it.
>> 
>> Andrew and I just discussed this on IRC, the result of which is that I
>> just reversed the ABI changing fix from the git libffi repo, and will
>> try to push out 3.0.10 this week in order to publish a Fedora 15 update.
>> 
>> Then I'll immediately re-apply this patch to the 3.0.11 train for F16.
>
> and having two libffi5's with a different set of symbols?

That's not what I'm saying. libffi 3.0.10 would have the same set of
symbols as 3.0.9 and also have the .5 so version number. 

Matthias and I just discussed this on #libffi.  He pointed out that 361
packages in Ubuntu depend on libffi and he's been building them with the
libffi 3.0.10 release candidates for a while.  Forcing him back to a .5
soname means forcing a rebuild of all of those packages, so...  I'll
make a 3.0.11 on the heals of 3.0.10 with just the debug symbol / soname
change so that Ubuntu can pick this up and avoid a major rebuild.  I'll
see about putting 3.0.11 into Fedora 16 as well.

AG

Re: Soname bump

[Matthias Klose]

On 08/15/2011 08:09 PM, Anthony Green wrote:
> Matthias Klose <doko@ubuntu.com> writes:
> 
>> On 08/15/2011 07:35 PM, Anthony Green wrote:
>>> Andrew Haley <aph@redhat.com> writes:
>>>
>>>> Yes, but that will mean that packages will be broken by the libffi update,
>>>> so the Fedora update mechanism will have to push new versions to hundreds
>>>> of thousands (millions?) of users.  If I change the libffi package I don't
>>>> think there's any way to confine the soname bump to ARM, even though only
>>>> ARM needs it.
>>>
>>> Andrew and I just discussed this on IRC, the result of which is that I
>>> just reversed the ABI changing fix from the git libffi repo, and will
>>> try to push out 3.0.10 this week in order to publish a Fedora 15 update.
>>>
>>> Then I'll immediately re-apply this patch to the 3.0.11 train for F16.
>>
>> and having two libffi5's with a different set of symbols?
> 
> That's not what I'm saying. libffi 3.0.10 would have the same set of
> symbols as 3.0.9 and also have the .5 so version number. 
> 
> Matthias and I just discussed this on #libffi.  He pointed out that 361
> packages in Ubuntu depend on libffi and he's been building them with the
> libffi 3.0.10 release candidates for a while.  Forcing him back to a .5
> soname means forcing a rebuild of all of those packages, so...  I'll
> make a 3.0.11 on the heals of 3.0.10 with just the debug symbol / soname
> change so that Ubuntu can pick this up and avoid a major rebuild.  I'll
> see about putting 3.0.11 into Fedora 16 as well.

thanks!  I didn't notice that the removal of the debug symbols would be reverted
too.

  Matthias