bug#70104: "FAIL: test-canonicalize" coreutils v9.5 on musl libc

[Bruno Haible]

Adept's Lab wrote:
> >> test-canonicalize.c:411: assertion 'strcmp (result1, "//") == 0' failed

Thanks for the report. I reproduce it with gnulib testdirs
  $ rm -rf ../testdir1; ./gnulib-tool --create-testdir --dir=../testdir1 
--single-configure canonicalize-lgpl
  $ rm -rf ../testdir2; ./gnulib-tool --create-testdir --dir=../testdir2 
--single-configure canonicalize
when run on Alpine Linux 3.9, 3.14, and 3.19.

History of these failures:
  - For many years, one of the tests was failing on Alpine Linux.
  - On 2021-01-17, I added a fix, that consists of a realpath() override [1].
  - On 2022-07-31, the bug was reported again [2], against the sed-4.8 release
    which predates the 2021-01-17 fix [3].
  - On 2023-09-04, Paul changed the behaviour of the unit tests on musl libc,
    without giving an explanation [4].
  - Now, the unit tests fail again, as reported above.

I'm therefore partially reverting Paul's change from 2023-09-04 (attached).

Paul Eggert wrote:
> Does this behavior (of whether / and // are the same directory) depend 
> on musl version, or on how musl is configured?

I think you must ask this to yourself:
  - What caused you to change the unit tests on 2023-09-04?
  - How is the musl version that you used on that date configured?
    What I use is Alpine Linux, as I said in versions 3.9, 3.14, 3.19.
  - Did you work with gnulib testdirs at that time, or did you use
    a package in which some modules are --avoid ed from import?

Bruno

[1] https://lists.gnu.org/archive/html/bug-gnulib/2021-01/msg00215.html
[2] https://lists.gnu.org/archive/html/bug-sed/2022-07/msg00003.html
[3] https://ftp.gnu.org/gnu/sed/
[4] https://lists.gnu.org/archive/html/bug-gnulib/2023-09/msg00011.html

0001-canonicalize-lgpl-tests-Fix-test-failure-on-musl-lib.patch
Description: Text Data