[tlbuild] Broken tests for mendexk on arm

[Johannes Hielscher]

Am Sun, 21 Mar 2021 15:48:31 +0100
schrieb Mojca Miklavec <mojca.miklavec.lists at gmail.com>:

> Hi,
> 
> These changes broke mendexk tests or arm (Raspberry PI; armhf-linux &
> aarch64-linux):
>     http://tug.org/svn/texlive?view=revision&revision=58576
> 
> Sorting pages....done(27 comparisons).
> Making index file....done.
> 0 warnings, written in pprecA-0.ilg.
> Output written in pprecA-0.ind1.
> + diff ../../../texk/mendexk/tests/pprecA-0.ind pprecA-0.ind1
> 3c3
> <   \item entryA[[delim0]]i[[sfx 3p]], 1[[sfx 3p]], I[[sfx 3p]]
> ---
> >   \item entryA[[delim0]]i, ii, iii, 1, 2, 3, I, II, III  
> + exit 1
> FAIL tests/mendex.test (exit status: 1)
> 
> (I admit that I was fiddling with arm-related changes a bit in the
> build configuration during the last two days, but I hope that wasn't
> the reason for the failure.)
(No, it wasn't.) I can confirm test failure with the same symptoms
(aarch64-linux, not RPi hardware, binaries built on Debian and
OpenSuSE).

However, the changes from -r58576 are not causative. Takuji extended
upmendex/tests/upmendex.test, and uncovered by this that something had
been broken a long time already: my aarch64-linux binaries fail in the
same way since at least TL2019.

This means that it is unrelated to the recent source code changes, but
older. Or maybe not even within upmendex itself? I don't know. From a
superficial view, I don't see any CPU architecture specific code within
upmendex; it is strange that only ARM CPUs under Linux appear affected.
Kudos to Takuji (and Suzana Sagadin) for detecting that there is a bug,
and revealing it with a properly designed test. Yet, at the moment I
don't see how I can contribute much to further locating/resolving it.


@Takuji: I wonder if there is some reason to clean up "foo" files with
`rm -f foo.*`, but not the "pprec*.i*" files that might have been
created by previous page_precedence tests?


Best,
Johannes