Re: Bug#814702: ghostscript: FTBFS on sparc64 due to mismatched symbols file

[Chris Liddell <cjl@shadesoft.net>]

The memory hdr_id is disabled on SPARC because it relies on memory 
accesses that aren't sufficiently aligned for SPARC - IIRC, it's aligned 
to sizeof(void *) which doesn't work fully on SPARC systems, but I could 
be wrong (I'm not looking at the sources right now).

If the failure is due to it being enable on x86 (and most other 
platforms) and not on SPARC, either Debian could opt to disable the 
hdr_id stuff for all platforms, or I can look into making it work 
correctly on SPARC.

I have to confess, even though I added it to Ghostcript, I was at the 
the time, and still am, unconvinced of its utility....

Chris


On 14/02/16 11:25, John Paul Adrian Glaubitz wrote:
> Source: ghostscript
> Version: 9.16~dfsg-2.1
> Severity: normal
> User: debian-sparc@lists.debian.org
> Usertags: sparc64
>
> Hello!
>
> ghostscript currently fails to build from source on sparc64 due to a mismatched
> symbols file:
>
> dpkg-gensymbols: warning: debian/libgs9/DEBIAN/symbols doesn't match completely debian/libgs9.symbols
> --- debian/libgs9.symbols (libgs9_9.16~dfsg-2.1_sparc64)
> +++ dpkg-gensymbolsIeNygU 2016-02-04 01:22:29.643520921 +0000
> @@ -3332,7 +3332,7 @@
>    gxht_thresh_planes@Base 9.05~
>    hc_put_code_proc@Base 8.61.dfsg.1
>    hc_put_last_bits_proc@Base 8.61.dfsg.1
> - hdr_id@Base 9.10~dfsg
> +#MISSING: 9.16~dfsg-2.1# hdr_id@Base 9.10~dfsg
>    height@Base 8.61.dfsg.1
>    ht_order_procs_table@Base 8.61.dfsg.1
>   #MISSING: 9.00~dfsg-1# iWrite2@Base 8.61.dfsg.1
> @@ -4004,6 +4004,7 @@
>    pixel_image_params@Base 8.61.dfsg.1
>    pixmap_high_level_pattern@Base 9.10~dfsg
>    plane_device_init@Base 8.61.dfsg.1
> + png_push_fill_buffer@Base 9.16~dfsg-2.1
>    pop_estack@Base 8.61.dfsg.1
>    pprintd1@Base 8.61.dfsg.1
>    pprintd2@Base 8.61.dfsg.1
> dh_makeshlibs: failing due to earlier errors
>
> The simple and obvious fix would be to mark the "hdr_id" symbol on sparc64 as
> missing which I have already done for the package in the 'unreleased' suite,
>
> (arch!=sparc64) hdr_id@Base 9.10~dfsg
>
> which works fine. However, I think it would be better to figure out first why
> this particular symbol is missing on sparc64 and whether this poses any problem.
>
> So far, the patched version seems to work without issues and we therefore
> might just go ahead and patch the symbols file as suggested, but it would
> be still be good if someone could have a deeper look at this. Maybe it's
> any of the (transitive) build dependencies of ghostscript which need to be
> rebuilt?
>
> Cheers,
> Adrian