[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: gzip (i386) uses TEXTRELs



Hello,

I've just built gzip in a i386 chroot with the previously supplied
debdiff applied and checked with scanelf (from pax-utils) that the
textrel goes away.

# scanelf -t noasm/bin/* /bin/gzip 
TYPE   TEXTREL FILE 
 ET_DYN    -    noasm/bin/gzip 
 ET_DYN TEXTREL /bin/gzip 

Andreas Pommer: It would be great if you could verify that
gzip built with the patch you find at
https://bugs.debian.org/cgi-bin/bugreport.cgi?att=1;bug=890279;filename=gzip-no-asm.debdiff;msg=21
(ie. attached to #890279 ) makes it possible for you to reenable the
protection in the logrotate service file that Christian had to drop
for you. (Either by downgrading to previous logrotate version or
just systemctl edit and put the service changes back in again.)

m68k and ia64 porters: Please note that gzip seems to have asm code for
your architectures as well which gets disabled by the proposed patch.
Please speak up now if you think the asm for your architecture is better
and should be preferred over the C code.
Previous discussion quoted below to give you some background. See
#890279 for entire discussion.

Given the lack of feedback on this issue, I assume NMU will be the next
step here unless I get any feedback soon after this message.

Regards,
Andreas Henriksson


On Wed, Aug 29, 2018 at 12:11:20PM +0200, Andreas Henriksson wrote:
> Hi,
> 
> There are some useful info in the discussion at:
> https://launchpad.net/ubuntu/+source/gzip/+bug/49067
> 
> Apparently fedora (et.al.) has nowadays abandoned their patch in favour
> of using 'export DEFS="NO_ASM"' in the spec file, see
> https://src.fedoraproject.org/rpms/gzip/tree/master
> 
> Their changelog also comes with this interesting claim:
> "- don't use the asm code again as it's slower than the gcc compiled one"
> 
> Googling "gzip DEFS NO_ASM" gives a number of matches including
> openembedded, coreos, etc.
> It seems it's pretty common to disable the (i386) assembler code
> and given the benefits it seems to have, I think it would be
> useful for debian to also do that.
> 
> For potential convenience I'm attaching an untested debdiff which I
> hope someone can help test and verify on i386.
> 
> Regards,
> Andreas Henriksson
> 
> PS. Please also note fedora seems to be carrying a patch for glibc 2.28
> which might be useful in debian soon as well.
> https://src.fedoraproject.org/rpms/gzip/blob/master/f/gnulib.patch

> diff -Nru gzip-1.9/debian/changelog gzip-1.9/debian/changelog
> --- gzip-1.9/debian/changelog	2018-08-05 02:30:09.000000000 +0200
> +++ gzip-1.9/debian/changelog	2018-08-29 11:40:01.000000000 +0200
> @@ -1,3 +1,11 @@
> +gzip (1.9-2.1) UNRELEASED; urgency=medium
> +
> +  * Non-maintainer upload.
> +  * Disable usage of assembler code (Closes: #890279)
> +    - fedora claims gcc generates more optimized code anyway.
> +
> + -- Andreas Henriksson <andreas@fatal.se>  Wed, 29 Aug 2018 11:40:01 +0200
> +
>  gzip (1.9-2) unstable; urgency=medium
>  
>    * move to upstream's less-ugly fix for mingw compilation failure
> diff -Nru gzip-1.9/debian/rules gzip-1.9/debian/rules
> --- gzip-1.9/debian/rules	2018-08-05 02:30:09.000000000 +0200
> +++ gzip-1.9/debian/rules	2018-08-29 11:40:01.000000000 +0200
> @@ -24,6 +24,7 @@
>  endif
>  endif
>  
> +export DEFS=NO_ASM # Avoid TEXTRELs on i386
>  EXTRA_CFLAGS=		-Wall
>  EXTRA_CPPFLAGS=
>  


Reply to: