[csw-devel] SF.net SVN: gar:[21040] csw/mgar/pkg/lang-python/pyqt/trunk/Makefile

Maciej Bliziński maciej at opencsw.org
Sat May 11 10:54:12 CEST 2013


On Fri, May 10, 2013 at 07:48:22PM +0200, Peter FELECAN wrote:
> Right. I wondered a long time about this, even asked for help on the
> maintainers list, and didn't find a reasonable explanation even though
> now it is crystal clear...
> 
> However, the explanation that I found at
> http://wiki.opencsw.org/checkpkg-error-tags#toc10 is a little bit
> misleading, isn't it?
> 
> "binary-architecture-does-not-match-placement
> 
> On the sparc architecture, binaries in /opt/csw/bin need to be at most a
> sparcv8 binary on Solaris 9 and at most a sparcv8+ binary on Solaris
> 10. A sparcv9 binary must be placed under a subdirectory,
> e.g. /opt/csw/bin/sparcv9.
> 
> A typical failure mode happens when CFLAGS from GAR are ignored by the
> build system, which causes Solaris Studio to produce sparcv8+ binaries,
> instead of sparcv8.
> 
> Recommended fix: Make sure that CFLAGS are propagated properly.
> When to override: Do not override. When in doubt, ask on the maintainers
> mailing list."

Yes, this check was written to catch something entirely different. By
coincidence, the binary in question was a binary for which we check if
it is in the right subdirectory. Otherwise the check wouldn't have
fired. So it was a coincidence, what we're concerned about here is not
what the check was actually looking at.

We don't currently have filesystem layout checks in place. If we had,
this kind of check would fire too.

Checks need some sort of algorithmic formula. We would need to define
what are good paths for files. For example, good paths would be:

/opt/csw/bin/foo
/opt/csw/bin/sparcv9/foo
/opt/csw/libexec/projname/bar
/opt/csw/projname/share/doc/README
/opt/csw/lib/projname/bar

Bad paths:

/usr/bin/foo
/opt/csw/bin/wibble/baz
/opt/csw/wibble/baz
/opt/csw/projname/wibble/baz

Debatable:

/opt/csw/projname/lib/libfoo.so.1

(Because shared libraries usually don't need to be isolated thanks to
the soname mechanism.)

Should we implement a check like that?

Maciej
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.opencsw.org/pipermail/devel/attachments/20130511/8967bb4a/attachment.asc>


More information about the devel mailing list