That's pretty unusual. I think that line breaks are determined using
libunistring's implementation of Unicode's linebreaking algorithm.
It's strange that it would act differently on Debian testing. We have
our own code of libunistring internally; maybe we're somehow using the
one that comes with Debian instead, and perhaps that's a slightly
different version that treats periods differently.
Is this the only failure you see there?
I don't think LC_ALL will affect this because the tests already set it
internally (the same way you did).
On Fri, Nov 22, 2024 at 11:17 AM Friedrich Beckmann
<friedrich.beckm...@posteo.de> wrote:
>
> I have a regression failure on debian testing which is related to the
> linebreaks in testing mode. The failure is:
>
> # -*- compilation -*-
> 315. evaluate.at:4361: testing expressions - TIME.DAYS ...
> ## --- ##
> ## OPT ##
> ## --- ##
> ../../pspp/tests/evaluate.at:4376: pspp --testing-mode evaluate.sps
> --- - 2024-11-22 19:44:23.140796466 +0100
> +++ /home/fritz/pspp/build/tests/testsuite.dir/at-groups/315/stdout
> 2024-11-22 19:44:23.135189554 +0100
> @@ -4,8 +4,8 @@
>
> time.days(0.5) => 43200.00
>
> -evaluate.sps:6.17-6.30: error: DEBUG EVALUATE: Type mismatch invoking TIME.
> -DAYS(number) as time.days(string).
> +evaluate.sps:6.17-6.30: error: DEBUG EVALUATE: Type mismatch invoking
> +TIME.DAYS(number) as time.days(string).
> 6 | DEBUG EVALUATE /time.days('x').
> | ^~~~~~~~~~~~~~
>
> 315. evaluate.at:4361: 315. expressions - TIME.DAYS (evaluate.at:4361):
> FAILED (evaluate.at:4376)
>
> So in my testsuite pspp makes a linebreak after „TIME.“ while the test
> expects to put „TIME.DAYS(number)“ in a new line. Any idea why this happens?
> I set the locale via „export LC_ALL=C“ before running the test but it does
> not make a difference.
