[Bug localedata/23421] New: Strange collation rules for A and space with UTF-8 locale when other characters appended

["b.cama at kerlink dot fr" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=23421

            Bug ID: 23421
           Summary: Strange collation rules for A and space with UTF-8
                    locale when other characters appended
           Product: glibc
           Version: 2.28
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: localedata
          Assignee: unassigned at sourceware dot org
          Reporter: b.cama at kerlink dot fr
                CC: libc-locales at sourceware dot org
  Target Milestone: ---

Created attachment 11136
  --> https://sourceware.org/bugzilla/attachment.cgi?id=11136&action=edit
A and space collation test case

Hi,
I stumbled against a strange string ordering bug, and managed to reduce it to
the attached test case. I *think* it comes from the locale data only, as I can
change the difference values slightly (but not the ordering) compared to an
older localedata, while still having the exact same behavior between 2.24 libc
and latest master (as of two days ago).

Here is the output with git master:

% ./testrun.sh ../collate_a_space
setlocale(LC_COLLATE,"C") = 361286057
strcoll("A", " ") = 33
strcoll("AB", " B") = 33
strcoll("B", " ") = 34
strcoll("BB", " B") = 34
setlocale(LC_COLLATE,"en_US.UTF-8") = 380448880
strcoll("A", " ") = 1
strcoll("AB", " B") = -13
strcoll("B", " ") = 1
strcoll("BB", " B") = 1

(the result is exactly the same when not using testrun.sh and only setting
LOCPATH)

And with my stock libc (2.24):

% ../collate_a_space
setlocale(LC_COLLATE,"C") = 1774630437
strcoll("A", " ") = 33
strcoll("AB", " B") = 33
strcoll("B", " ") = 34
strcoll("BB", " B") = 34
setlocale(LC_COLLATE,"en_US.UTF-8") = 1082470992
strcoll("A", " ") = 1
strcoll("AB", " B") = -1
strcoll("B", " ") = 1
strcoll("BB", " B") = 1

Note the second strcoll test, which gives an opposite result with an UTF-8
locale compared to raw C one. This only happen with letter “A” (or even “a”),
but no other one, hence the “B” test for comparison. And the ordering is
correct when comparing lone “A” (or any letter) and “ ” (space), with nothing
appended.

Note that this is with 100% ASCII characters.

-- 
You are receiving this mail because:
You are on the CC list for the bug.