Bug#884850: librsvg: FTBFS on (32-bit && !any-i386): masking-mask-01-b.svg: 661 pixels differ (with maximum difference of 2) from reference image

[Simon McVittie]

Source: librsvg
Version: 2.40.20-1
Severity: serious
Justification: fails to build from source (but built successfully in the past)

librsvg fails the same test in the same way on all the 32-bit
architectures, except for (hurd-)i386 which have a different test
failure, m68k and powerpcspe which (IMO unwisely) build with nocheck,
and kfreebsd-i386 which hasn't been attempted yet. This seems too
well-correlated to be coincidence.

https://buildd.debian.org/status/fetch.php?pkg=librsvg&arch=armel&ver=2.40.20-1&stamp=1513733747&raw=0
https://buildd.debian.org/status/fetch.php?pkg=librsvg&arch=armhf&ver=2.40.20-1&stamp=1513733897&raw=0
https://buildd.debian.org/status/fetch.php?pkg=librsvg&arch=mips&ver=2.40.20-1&stamp=1513733801&raw=0
https://buildd.debian.org/status/fetch.php?pkg=librsvg&arch=mipsel&ver=2.40.20-1&stamp=1513734074&raw=0
https://buildd.debian.org/status/fetch.php?pkg=librsvg&arch=hppa&ver=2.40.20-1&stamp=1513745188&raw=0
https://buildd.debian.org/status/fetch.php?pkg=librsvg&arch=powerpc&ver=2.40.20-1&stamp=1513733546&raw=0
https://buildd.debian.org/status/fetch.php?pkg=librsvg&arch=x32&ver=2.40.20-1&stamp=1513734144&raw=0
> # Storing test result image at /tmp/masking-mask-01-b-out.png
> # 661 pixels differ (with maximum difference of 2) from reference image
>
> # Storing test result image at /tmp/masking-mask-01-b-diff.png
> not ok 16 /rsvg/reftest/svg1.1/masking-mask-01-b.svg
> FAIL: rsvg-test 16 /rsvg/reftest/svg1.1/masking-mask-01-b.svg

i386 and hurd-i386 had a different test failure for which I've opened a
separate bug (wild guess: maybe it's to do with excess precision in the
387-compatible FPU, which is an i386 quirk).

Regards,
    smcv