[PATCH] glibc: install correct stubs (fixes midori crash)

[PATCH] glibc: install correct stubs (fixes midori crash)

[Stanislav Brabec]

Attached patch prevents overwriting of a real stub include file by a
dummy stub file. Patch is tested only with the latest glibc from the
angstrom.

Dummy stub include causes mis-compilation of undefined number of
packages, increment of DISTRO_PR may be considered as useful.

I am not sure, why this line was there. Correct stub file is generated
by make install, at least in the current version of glibc.

--- a/recipes/glibc/glibc-stage.inc
+++ b/recipes/glibc/glibc-stage.inc
@@ -8,7 +8,6 @@ do_stage() {
 	install -d ${STAGING_INCDIR}/gnu \
 		   ${STAGING_INCDIR}/bits \
 		   ${STAGING_INCDIR}/rpcsvc
-	install -m 0644 ${S}/include/gnu/stubs.h ${STAGING_INCDIR}/gnu/
 	install -m 0644 ${B}/bits/stdio_lim.h ${STAGING_INCDIR}/bits/
 	install -m 0644 misc/syscall-list.h ${STAGING_INCDIR}/bits/syscall.h
 	for r in ${rpcsvc}; do

Review of compilation logs shows that at least following packages are
referring stub functions: glib-2.0, coreutils, module-init-tools,
openssh, rsync, e2fsprogs, libpcap, pth, python, rxvt-unicode, tar,
tcpdump.

Example of mis-compiled package:

midori fails due to GIO mis-compilation. GIO (glib-2.0) incorrectly
thinks, that inotify_init1 is implemented. But it is not true on systems
compiled with kernel headers < 2.6.27. GIO fails to initialize change
notification, which normally never happens.

Note:

Kernel added several new functions since 2.6.23 (e. g. inotify_init1).
It would be nice to reflect it somehow: update linux-libc-headers and
either keep minimal supported kernel in the build as low as possible
(makes a bit slower glibc), move affected packages from platform to
machine feeds or create different repos with different minimal kernel
version. What do you think about it?


________________________________________________________________________
Stanislav Brabec
http://www.penguin.cz/~utx/zaurus

Re: [PATCH] glibc: install correct stubs (fixes midori crash)

[Koen Kooi]

On 14-07-09 21:24, Stanislav Brabec wrote:
> Attached patch prevents overwriting of a real stub include file by a
> dummy stub file. Patch is tested only with the latest glibc from the
> angstrom.
>
> Dummy stub include causes mis-compilation of undefined number of
> packages, increment of DISTRO_PR may be considered as useful.
>
> I am not sure, why this line was there. Correct stub file is generated
> by make install, at least in the current version of glibc.
>
> --- a/recipes/glibc/glibc-stage.inc
> +++ b/recipes/glibc/glibc-stage.inc
> @@ -8,7 +8,6 @@ do_stage() {
>   	install -d ${STAGING_INCDIR}/gnu \
>   		   ${STAGING_INCDIR}/bits \
>   		   ${STAGING_INCDIR}/rpcsvc
> -	install -m 0644 ${S}/include/gnu/stubs.h ${STAGING_INCDIR}/gnu/
>   	install -m 0644 ${B}/bits/stdio_lim.h ${STAGING_INCDIR}/bits/
>   	install -m 0644 misc/syscall-list.h ${STAGING_INCDIR}/bits/syscall.h
>   	for r in ${rpcsvc}; do
>
> Review of compilation logs shows that at least following packages are
> referring stub functions: glib-2.0, coreutils, module-init-tools,
> openssh, rsync, e2fsprogs, libpcap, pth, python, rxvt-unicode, tar,
> tcpdump.
>
> Example of mis-compiled package:
>
> midori fails due to GIO mis-compilation. GIO (glib-2.0) incorrectly
> thinks, that inotify_init1 is implemented. But it is not true on systems
> compiled with kernel headers<  2.6.27. GIO fails to initialize change
> notification, which normally never happens.
>
> Note:
>
> Kernel added several new functions since 2.6.23 (e. g. inotify_init1).
> It would be nice to reflect it somehow: update linux-libc-headers and
> either keep minimal supported kernel in the build as low as possible
> (makes a bit slower glibc), move affected packages from platform to
> machine feeds or create different repos with different minimal kernel
> version. What do you think about it?

Apply the patch, bump glibc PR and angstrom distro PR.

regards,

Koen

Re: [PATCH] glibc: install correct stubs (fixes midori crash)

[Michael Smith]

Stanislav Brabec wrote:
 > Attached patch prevents overwriting of a real stub include file by a
 > dummy stub file. Patch is tested only with the latest glibc from the
 > angstrom.

Hi Stanislav,

Patch 2/3 here fixes it:

http://patchwork.openembedded.org/bundle/msmith%40cbnco.com/glibc_tweaks/

Also cleans some of the other lines that aren't needed.

Mike

Re: [PATCH] glibc: install correct stubs (fixes midori crash)

[Stanislav Brabec]

Michael Smith wrote:
> Stanislav Brabec wrote:
>  > Attached patch prevents overwriting of a real stub include file by a
>  > dummy stub file. Patch is tested only with the latest glibc from the
>  > angstrom.
> 
> Hi Stanislav,
> 
> Patch 2/3 here fixes it:
> 
> http://patchwork.openembedded.org/bundle/msmith%40cbnco.com/glibc_tweaks/
> 
> Also cleans some of the other lines that aren't needed.

Your patch is better.

And INC_PR seems to be a good idea for any .inc file. I have been
thinking about a generic feature setting revision to
inc_revision+bb_revision.

Well, I am not a core team member, but you have my ACK. The bug is
apparent and debugging of subsequent errors take a lot of time.

Acked-by: Stanislav Brabec <utx@penguin.cz>


________________________________________________________________________
Stanislav Brabec
http://www.penguin.cz/~utx/zaurus