Skip to contentSkip to conversation

make "LDFLAGS ignored" QA errors fatal

RFC In review Jeremy Lainé 5 participants 10 messages Apr 6, 2009 – Apr 14, 2009

lore
Share & tools
lore
#0 lore raw

RFC 1/1 make "LDFLAGS ignored" QA errors fatal

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I have spent quite some time cleaning up a bunch of recipes which were triggering "LDFLAGS ignored" QA errors, and I don't really want to have to do this again in the future. I suggest we stop whitelisting this kind of QA error, using the attached patch.

To track down the faulty packages I have made use of oestats output (note that not all QA errors are "LDFLAGS ignored"):

http://tinderbox.openembedded.net/packages/?qalog=1

In most cases, the problem is that recipes simply pass
- -L${STAGING_LIBDIR} as LDFLAGS instead of using the full ${LDFLAGS} variable. The result is that flags such as --hash-style=gnu are getting ignored.

What are your thoughts?

- --
Jeremy LAINE
Bolloré telecom | 11bis, rue Scribe | F-75009 Paris -----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAknaLysACgkQCL1n3lVB9EZgbQCgoQpUwdae7elUHWag8wqAROY7 D0sAmQGjH0zrO8OxgI5MPfSAshQz9Fuu
=to11
-----END PGP SIGNATURE-----

classes/insane.bbclass +1 -1
index 3e519c3..9fc5f07 100644
@@ -184,7 +184,7 @@ def package_qa_make_fatal_error(error_class, name, path,d):
 
     TODO: Load a whitelist of known errors
     """
-    return not error_class in [0, 5, 7, 9]
+    return not error_class in [0, 5, 7]
 
 def package_qa_write_error(error_class, name, path, d):
     """
#1 lore raw
Quoted 11 lines · Jeremy Lainé

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I have spent quite some time cleaning up a bunch of recipes which were triggering "LDFLAGS ignored" QA errors, and I don't really want to have to do this again in the future. I suggest we stop whitelisting this kind of QA error, using the attached patch.

To track down the faulty packages I have made use of oestats output (note that not all QA errors are "LDFLAGS ignored"):

Quoted 8 lines

http://tinderbox.openembedded.net/packages/?qalog=1

In most cases, the problem is that recipes simply pass
- -L${STAGING_LIBDIR} as LDFLAGS instead of using the full ${LDFLAGS} variable. The result is that flags such as --hash-style=gnu are getting ignored.

What are your thoughts?

I think making them fatal is a good idea. I do have some recipes where I can't fix LDFLAGS properly yet, but I can do INSANE_SKIP = 1 inside those for the time being.

So: +1

regards,

Koen

#2 lore raw
In reply to Jeremy Lainé
Quoted 11 lines · Jeremy Lainé

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I have spent quite some time cleaning up a bunch of recipes which were triggering "LDFLAGS ignored" QA errors, and I don't really want to have to do this again in the future. I suggest we stop whitelisting this kind of QA error, using the attached patch.

To track down the faulty packages I have made use of oestats output (note that not all QA errors are "LDFLAGS ignored"):

Quoted 6 lines

http://tinderbox.openembedded.net/packages/?qalog=1

In most cases, the problem is that recipes simply pass
- -L${STAGING_LIBDIR} as LDFLAGS instead of using the full ${LDFLAGS} variable. The result is that flags such as --hash-style=gnu are getting ignored.

I think we should configure toolchain to pass this flags by gcc driver while linking instead of making it a part of LDFLAGS it should help in recipes which use gcc to link.

Quoted 16 lines

What are your thoughts?

- --
Jeremy LAINE
Bolloré telecom | 11bis, rue Scribe | F-75009 Paris -----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAknaLysACgkQCL1n3lVB9EZgbQCgoQpUwdae7elUHWag8wqAROY7 D0sAmQGjH0zrO8OxgI5MPfSAshQz9Fuu
=to11
-----END PGP SIGNATURE-----

#3 lore raw
Quoted 14 lines · Khem Raj
Quoted 13 lines · Jeremy Lainé

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I have spent quite some time cleaning up a bunch of recipes which were triggering "LDFLAGS ignored" QA errors, and I don't really want to have to do this again in the future. I suggest we stop whitelisting this kind of QA error, using the attached patch.

To track down the faulty packages I have made use of oestats output (note that not all QA errors are "LDFLAGS ignored"):

http://tinderbox.openembedded.net/packages/?qalog=1

Quoted 8 lines
Quoted 4 lines

In most cases, the problem is that recipes simply pass
- -L${STAGING_LIBDIR} as LDFLAGS instead of using the full ${LDFLAGS} variable. The result is that flags such as --hash-style=gnu are getting ignored.

I think we should configure toolchain to pass this flags by gcc driver while linking instead of making it a part of LDFLAGS it should help in recipes which use gcc to link.

While that would improve the situation a bit, it would also mean that we'll have a harder time detecting which recipes ignore LDFLAGS.

regards,

Koen

#4 lore raw
In reply to Khem Raj
Quoted 3 lines · Khem Raj

I think we should configure toolchain to pass this flags by gcc driver while linking instead of making it a part of LDFLAGS it should help in recipes which use gcc to link.

I'm not sure what that would really buy us. Unless we're going to abandon LDFLAGS altogether then all the packages that are broken today would still be broken, it's just that you'd notice the breakage less often. Plus, it would presumably mean that any changes to the value of this flag would require a gcc rebuild, which seems like it'd be annoying.

p.

#5 lore raw

But before making QA fatal, please let fix these (Ignoring for the moment the QA 'the desktop file is not valid'):

irda-utils:
LDFLAGS ignored, irda-utils, /work/armv5te-angstrom-linux-gnueabi/irda-utils-0.9.16-r8/install/irda-utils/usr/sbin/dongle_attach LDFLAGS ignored, irda-utils, /work/armv5te-angstrom-linux-gnueabi/irda-utils-0.9.16-r8/install/irda-utils/usr/sbin/irdaping LDFLAGS ignored, irda-utils, /work/armv5te-angstrom-linux-gnueabi/irda-utils-0.9.16-r8/install/irda-utils/usr/sbin/irattach

hostap-utils:
LDFLAGS ignored, hostap-utils, /work/armv5te-angstrom-linux-gnueabi/hostap-utils-0.4.7-r5/install/hostap-utils/usr/sbin/hostap_rid LDFLAGS ignored, hostap-utils, /work/armv5te-angstrom-linux-gnueabi/hostap-utils-0.4.7-r5/install/hostap-utils/usr/sbin/hostap_io_debug LDFLAGS ignored, hostap-utils, /work/armv5te-angstrom-linux-gnueabi/hostap-utils-0.4.7-r5/install/hostap-utils/usr/sbin/hostap_diag

ppp:
LDFLAGS ignored, ppp, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp/usr/sbin/chat
LDFLAGS ignored, ppp-oa, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-oa/usr/lib/pppd/2.4.3/pppoatm.so LDFLAGS ignored, ppp-oe, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-oe/usr/sbin/pppoe-discovery LDFLAGS ignored, ppp-oe, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-oe/usr/lib/pppd/2.4.3/rp-pppoe.so LDFLAGS ignored, ppp-radius, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-radius/usr/lib/pppd/2.4.3/radattr.so LDFLAGS ignored, ppp-radius, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-radius/usr/lib/pppd/2.4.3/radrealms.so LDFLAGS ignored, ppp-radius, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-radius/usr/lib/pppd/2.4.3/radius.so LDFLAGS ignored, ppp-winbind, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-winbind/usr/lib/pppd/2.4.3/winbind.so LDFLAGS ignored, ppp-minconn, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-minconn/usr/lib/pppd/2.4.3/minconn.so LDFLAGS ignored, ppp-password, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-password/usr/lib/pppd/2.4.3/passwordfd.so LDFLAGS ignored, ppp-password, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-password/usr/lib/pppd/2.4.3/passprompt.so LDFLAGS ignored, ppp-tools, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-tools/usr/sbin/pppdump LDFLAGS ignored, ppp-tools, /work/armv5te-angstrom-linux-gnueabi/ppp-2.4.3-r4/install/ppp-tools/usr/sbin/pppstats

obexpush:
LDFLAGS ignored, obexpush, /work/armv5te-angstrom-linux-gnueabi/obexpush-1.0.0-r4/install/obexpush/usr/bin/obextool LDFLAGS ignored, obexpush, /work/armv5te-angstrom-linux-gnueabi/obexpush-1.0.0-r4/install/obexpush/usr/bin/ussp-push

While you're there, why don't try to fix these:

klibc:
non dev contains .so, klibc, /work/c7x0-angstrom-linux-gnueabi/klibc-1.5-r9/install/klibc/lib/klibc.so

libopieobex0:
non dev contains .so, libopieobex0, /work/armv5te-angstrom-linux-gnueabi/libopieobex0-1.2.4-r0/install/libopieobex0/usr/lib/opie/plugins/obex/libopieobex.so

libgtkstylus:
non dev contains .so, libgtkstylus, /work/armv5te-angstrom-linux-gnueabi/libgtkstylus-0.3-r5/install/libgtkstylus/usr/lib/gtk-2.0/2.10.0/modules/libgtkstylus.so

Regards

Andrea

#6 lore raw
Quoted 3 lines · Andrea Adami

libgtkstylus:
non dev contains .so, libgtkstylus, /work/armv5te-angstrom-linux-gnueabi/libgtkstylus-0.3-r5/install/libgtkstylus/usr/lib/gtk-2.0/2.10.0/modules/libgtkstylus.so

This is a bogus error. The package is behaving correctly, the error just needs to be suppressed.

p.

#7 lore raw
In reply to Andrea Adami

As requested I have fixed the following packages which were not respecting LDFLAGS.

* irda-utils
* hostap-utils
* ppp
* obexpush

For the other QA errors I'd rather someone else handled it.

Signature · 2 lines
Jeremy LAINE
Bolloré telecom | 11bis, rue Scribe | F-75009 Paris
#9 lore raw
In reply to Phil Blundell
Quoted 9 lines · Phil Blundell
Quoted 3 lines · Khem Raj

I think we should configure toolchain to pass this flags by gcc driver while linking instead of making it a part of LDFLAGS it should help in recipes which use gcc to link.

I'm not sure what that would really buy us. Unless we're going to abandon LDFLAGS altogether then all the packages that are broken today would still be broken, it's just that you'd notice the breakage less often. Plus, it would presumably mean that any changes to the value of

yes this would help if we were making changes just to accomodate the gnu-hash linker option

Quoted 2 lines

this flag would require a gcc rebuild, which seems like it'd be annoying.

I think chosing the hash style is something you will chose to have or to have not at the beginning. However this will only make toolchain to default to gnu-hash on supported architectures one can still use --hash-style=sysv if one has to

Thanks

-Khem

Quoted 9 lines

p.

_______________________________________________
Openembedded-devel mailing list
[email protected]
http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel