[PATCH] Fix debian reproducibility test for tcsh

[josefs10]

tcsh 6.24.10-3 failed reproducibility tests in debian because of build variations with locale using the sort function. This patch switches LC_COLLATE with LC_ALL to fix that issue, based off this known reprotest issue. use-LC_ALL-with-sort.patch.txt

[suominen]

I did not find an explanation for the need to use LC_ALL instead of LC_COLLATE in the reprotest issue you linked to. Could you perhaps quote the relevant part in a comment here, please?

[josefs10]

You are right, the bug description is not very good, but it showed me what to look at. So reprotest in the second build overrides LC_ALL when varying locale. Here is an example with some of the lines from tc.const.c that get sorted.

$ LANG=en_US.UTF-8 LANGUAGE=en_US.UTF-8:en LC_COLLATE=C sort ./file.txt
STR0
STR1
STRBB
STR_
STRaddsuffix
STRbang

$ LC_ALL=et_EE.UTF-8 LANGUAGE=et_EE.UTF-8:fr LC_COLLATE=C sort ./file.txt
STR_
STR0
STR1
STRaddsuffix
STRbang
STRBB

Because reprotest overrides LC_ALL, the output of the sort command varies which causes the changes in tc.const.h and the binary differences. Build log for failed reprotest is below. reprotest-fail-build.txt

[suominen]

LC_ALL is an override for all LC_* variables, whereas LANG is a default for them. I can't help but feel like this is a bug in reprotest: it's expectation that setting LC_ALL does not change anything seems incorrect to me.

Maybe we should just be unsetting LC_ALL, although we'd be doing it just to pacify reprotest. I'd rather have reprotest fixed.

Or is there some important logic behind the behaviour of reportest that I should understand?

[josefs10]

I agree. Perhaps this is a reprotest issue (or reprotest feature that just causes problems here). This was my first time using reprotest, so I can talk to the maintainers to see if this can be fixed with some kind of override or if the default reprotest builds can be done without varying locale.

[thep]

LC_ALL is an override for all LC_* variables, whereas LANG is a default for them. I can't help but feel like this is a bug in reprotest: it's expectation that setting LC_ALL does not change anything seems incorrect to me.

I think I can understand why reprotest does this: its purpose is to simulate different build environments as much as possible, including when LC_ALL happens to be set, and the build should still reproduce the same result.

Maybe we should just be unsetting LC_ALL, although we'd be doing it just to pacify reprotest. I'd rather have reprotest fixed.

Unsetting LC_ALL is also fine, as long as all other locale categories have no effect on the output, which is the case here. I just think overriding LC_ALL here is simpler and safer.

I don't know the opinion of the reprotest maintainer, but I don't think it needs a fix for this.

[suominen]

@thep : I think you are right, it is safer to use LC_ALL, and now — having thought about it more — I think this is exactly the use case that LC_ALL was intended for.

Thank you both for taking the time to report the issue and to explain it to me. :)