Re: llvm35/clang35 validation requires @rpath acceptance in shlib validation

On 11/8/14, 5:23 PM, Alexander Hansen wrote:
> On 11/8/14, 3:33 PM, David Fang wrote:
>> On Sat, 8 Nov 2014, Alexander Hansen wrote:
>>
>>> On 11/8/14, 1:03 PM, David Fang wrote:
>>>>  Alex,
>>>>
>>>> >  On Nov 4, 2014, at 11:07 AM, David Fang <[email protected]>
>>>> wrote:
>>>> > > >  Hi all,
>>>> > >       Sitting in my experimental tree is a draft of llvm35's
>>>> packaging
>>>> > >  that bootstraps (much like its predecessor llvm34), but fails
>>>> Shlib
>>>> > >  validation because the install_names for shlibs all start with
>>>> @rpath
>>>> > >  (or
>>>> > >  on older darwin @executable_path or @loader_path).  This was a
>>>> > >  decision by
>>>> > >  upstream build maintainers that allows this package to be
>>>> > >  binary-relocatable, even though fink makes no such guarantees
>>>> about > >  its
>>>> > >  packages -- fink's remain in fixed install locations.
>>>> > >  dmacks and akh began to look into a patch that would allow this
>>>> to > >  pass
>>>> > >  validation.  Can you guys comment on the status/issues/obstacles?
>>>> > >  Such a fix would unblock llvm35 from hitting dists.
>>>> > > > >  Thanks!
>>>> > > > >  Fang
>>>> > > > >  --
>>>> > >  David Fang
>>>> > >  http://www.csl.cornell.edu/~fang/
>>>> > > > > >  Here?s the change I injected thus far:
>>>> > >  diff --git a/perlmod/Fink/Validation.pm
>>>> b/perlmod/Fink/Validation.pm
>>>> >  index 29fc4ab..a5010dd 100644
>>>> >  --- a/perlmod/Fink/Validation.pm
>>>> >  +++ b/perlmod/Fink/Validation.pm
>>>>> @@ -2198,12 +2198,12 @@ sub _validate_dpkg {
>>>> >                                         <OTOOL>; # skip first line
>>>> >                                         my ($libname,
>>>> $compat_version)
>>>> >  = <OTOOL> =~ /^\s*(\S+)\s*\(compatibility version ([\d\.]+)/;
>>>> >                                         close (OTOOL);
>>>> >  -                                       if ($libname !~ /^\//) {
>>>> >  +                                       if (($libname !~ /^\//) and
>>>> >  ($libname !~ /\@.*path\//)) {
>>>> >                                                 print "Error: package
>>>> >  contains the shared library\n";
>>>> >                                                 print "
>>>> >  $dylib\n";
>>>> >                                                 print "       but the
>>>> >  corresponding install_name\n";
>>>> >                                                 print "
>>>> >  $libname\n";
>>>> >  -                                               print "       is not
>>>> >  an absolute pathname.\n";
>>>> >  +                                               print "       is not
>>>> >  an absolute pathname or runtime path.\n";
>>>> >                                                 $looks_good = 0;
>>>> >                                         } elsif (not exists
>>>> >  $deb_shlibs->{$libname}) {
>>>> >                                                 $libname =~
>>>> >  s/^$basepath/%p/;
>>>> > >  And the validator moves on:
>>>> > >  Validating .deb file
>>>> >  /sw/fink/debs/polly35-shlibs_3.5.0-0_darwin-x86_64.deb...
>>>> >  Error: package contains the shared library
>>>> >           /sw/opt/llvm-3.5/lib/libPolly.dylib
>>>> >        but the corresponding install_name and compatibility_version
>>>> >           @rpath/libPolly.dylib 0.0.0
>>>> >        are not listed in the Shlibs field.  See the packaging manual.
>>>>
>>>>  Thanks for working on this patch.
>>>>  I'm a little puzzled about the libPolly error, since my .info file
>>>> has:
>>>>
>>>>       Package: polly35-shlibs
>>>>       ...
>>>>           Files: <<
>>>>                   opt/llvm-3.5/lib/LLVMPolly.so
>>>>                   opt/llvm-3.5/lib/libPolly*.dylib
>>>>           <<
>>>>           DocFiles: tools/polly/LICENSE.txt tools/polly/CREDITS.txt
>>>>           Shlibs: <<
>>>> !% p/opt/llvm-3.5/lib/LLVMPolly.so
>>>> !% p/opt/llvm-3.5/lib/libPolly.dylib
>>>>           <<
>>>>
>>>>  Am I missing some obvious fix?
>>>>
>>>>  Fang
>>>>
>>>
>>> I assumed that it was because Shlibs entries refer to the
>>> install_name, which happens not to map onto the filename when using
>>> @rpath and friends.  Maybe
>>>
>>> Shlibs: <<
>>>     !@ rpath/LLVMPolly.so
>>>     !@ rpath/LLVMPolly.dylib
>>> <<
>>>
>>> would do the job.  I'll check that.
>>
>> Did any of the other split-offs trigger that diagnostic?  All shlib
>> split-offs follow the same form.
>>
>
> Yeah, they all did.
>

Now, having put @rpath/foo.dylib in the relevant places, I've been able 
to find the next place where there's a problem:

Fiona-2:upstream_fink hansen$ fink validate 
/sw/fink/debs/clang35-shlibs_3.5.0-0_darwin-x86_64.deb
Validating .deb file 
/sw/fink/debs/clang35-shlibs_3.5.0-0_darwin-x86_64.deb...
Error: Shlibs field specifies file '@rpath/libclang.3.5.dylib', but it 
does not exist!
Error: Shlibs field specifies file '@rpath/libiomp5.dylib', but it does 
not exist!

Maybe in the case where we have @rpath or other such variable we'll want 
simply to find the dylib, since the whole point of @rpath is that the 
libraries don't need to know where they are except relative to the 
executable, right?
-- 
Alexander Hansen, Ph.D.
Fink User Liaison
My package updates: http://finkakh.wordpress.com/

------------------------------------------------------------------------------
_______________________________________________
Fink-devel mailing list
[email protected]
List archive:
http://news.gmane.org/gmane.os.apple.fink.devel
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-devel