Inconsistent sort behavior

[odysei]

While doing some sorting with new sort (v 0.8.0) that came with Ubuntu 26.04, I have noticed an inconsistent behavior.
Attaching example sort1.txt and sort2.txt files that:

• lines 6&7 got randomly assigned using sort;
• stay invariant if you run over sort (again).

Expected behavior:

1. files sort1.txt and sort2 are identical;
2. sort produces the same output for inputs from sort1.txt or sort2.txt

sort1.txt
sort2.txt

[ksgk1]

I am currently looking into this, I currently worked on a minimal example, that shows that this is indeed an issue. Two files containing:

01
0_1
02

and

0_1
01
02

will produce:

01
0_1
0_1
01
02
02

Looking into the compare code, when setting the global_settings.precomputed.fast_lexicographic = true; it will produce:

01
01
02
02
0_1
0_1

Using --sort=<general-numeric,human-numeric> also produces a well sorted list.

[ksgk1]

I created a PR to fix this issue. #12013

[collinfunk]

@sylvestre This bug is quite severe. I imagine it will need patching in Ubuntu.

The issue is all the programs with localization implemented (e.g. sort, join, expr) use AlternateHandling::Shifted. This leads to some obviously incorrect behavior, for example:

$ expr 'USA' = 'U.S.A.'
0
$ uu_expr 'USA' = 'U.S.A.'
1
$ printf '_\n.\n\n' | sort -u

.
_
$ printf '_\n.\n\n' | uu_sort -u
_

@pixelb FYI, I'll have a look at adding some tests with punctuation later.