Bug#1038252: librsvg: FTBFS on i386: transform::tests::parses_transform_list, transform::tests::parses_valid_transform failed

[Simon McVittie]

Control: severity -1 important

On Sun, 18 Jun 2023 at 12:42:12 +0100, Simon McVittie wrote:
> On Fri, 16 Jun 2023 at 21:11:21 +0100, Simon McVittie wrote:
> > On Fri, 16 Jun 2023 at 21:00:13 +0200, Sebastian Ramacher wrote:
> > > https://buildd.debian.org/status/fetch.php?pkg=librsvg&arch=i386&ver=2.54.5%2Bdfsg-2&stamp=1686829500&raw=0
> > > 
> > > ---- transform::tests::parses_transform_list stdout ----
> > > thread 'transform::tests::parses_transform_list' panicked at 'assertion failed: t1.y0.approx_eq(t2.y0, (epsilon, 1))', src/transform.rs:716:9
> > > 
> > > ---- transform::tests::parses_valid_transform stdout ----
> > > thread 'transform::tests::parses_valid_transform' panicked at 'assertion failed: t1.y0.approx_eq(t2.y0, (epsilon, 1))', src/transform.rs:716:9
> > 
> > This might be i387 excess precision resulting in rounding errors on i386
> > that differ from what happens on other architectures, or it might be
> > something unrelated.
> 
> Rebuilding 2.54.5+dfsg-1 in bookworm has similar behaviour, which means
> this is a behaviour change in build-dependency packages, rather than a
> result of a change in librsvg. One possible trigger is the upgrade of rustc
> 1.61 to 1.63.

I've uploaded 2.54.5+dfsg-3 with those two tests skipped on i386. It's
still a bug that we have to do that, and investigation by someone who
knows i386 quirks and/or Rust would be useful, but skipping those tests
makes this non-RC.

    smcv