mbsnlen: Fix bug in mcel-based implementation

[Bruno Haible]

The new u*-vasnprintf tests that I added today succeed on all
platforms. But on Solaris 11.4, in a testdir that contains also
the 'mcel-prefer' module, they fail:


FAIL: unistdio/test-ulc-vasnprintf3.sh
======================================

../../gltests/unistdio/test-ulc-vasnprintf3.c:226: assertion 'strcmp (result, " 
      h\303\251t\303\251rog\303\251n\303\251it\303\251 33") == 0' failed
Stack trace:
0x40c59d print_stack_trace
        ../../gllib/abort-debug.c:40
0x40c59d rpl_abort
        ../../gllib/abort-debug.c:94
0x40c0f8 test_function
        ../../gltests/unistdio/test-ulc-vasnprintf3.c:226
0x40c487 test_vasnprintf
        ../../gltests/unistdio/test-ulc-vasnprintf3.c:266
0x40c4be main
        ../../gltests/unistdio/test-ulc-vasnprintf3.c:276
../../gltests/unistdio/test-ulc-vasnprintf3.sh: line 20: 6427: Abort(coredump)
../../build-aux/test-driver: line 113: 6425: Abort(coredump)
FAIL unistdio/test-ulc-vasnprintf3.sh (exit status: 262)


The cause, it turns out, is that
  mbsnlen ("h\251", 1)
returns 2. Which is obvious nonsense: mbsnlen (s, count) should always return
a value ≤ count.

So, I wrote unit tests for mbsnlen, and the mcel-based implementation fails
these tests in many places:

../../gltests/test-mbsnlen.c:35: assertion 'mbsnlen ("", 1) == 1' failed
../../gltests/test-mbsnlen.c:36: assertion 'mbsnlen ("\0", 2) == 2' failed
../../gltests/test-mbsnlen.c:38: assertion 'mbsnlen ("H", 0) == 0' failed
../../gltests/test-mbsnlen.c:40: assertion 'mbsnlen ("H", 2) == 2' failed
../../gltests/test-mbsnlen.c:42: assertion 'mbsnlen ("Hello", 0) == 0' failed
../../gltests/test-mbsnlen.c:43: assertion 'mbsnlen ("Hello", 1) == 1' failed
../../gltests/test-mbsnlen.c:44: assertion 'mbsnlen ("Hello", 2) == 2' failed
../../gltests/test-mbsnlen.c:46: assertion 'mbsnlen ("Hello", 6) == 6' failed
../../gltests/test-mbsnlen.c:50: assertion 'mbsnlen ("\303\244\303\266", 0) == 
0' failed
../../gltests/test-mbsnlen.c:51: assertion 'mbsnlen ("\303\244\303\266", 1) == 
1' failed
../../gltests/test-mbsnlen.c:52: assertion 'mbsnlen ("\303\244\303\266", 2) == 
1' failed
../../gltests/test-mbsnlen.c:53: assertion 'mbsnlen ("\303\244\303\266", 3) == 
2' failed
../../gltests/test-mbsnlen.c:55: assertion 'mbsnlen ("\303\244\303\266", 5) == 
3' failed
../../gltests/test-mbsnlen.c:57: assertion 'mbsnlen ("7\342\202\254", 0) == 0' 
failed
../../gltests/test-mbsnlen.c:58: assertion 'mbsnlen ("7\342\202\254", 1) == 1' 
failed
../../gltests/test-mbsnlen.c:59: assertion 'mbsnlen ("7\342\202\254", 2) == 2' 
failed
../../gltests/test-mbsnlen.c:60: assertion 'mbsnlen ("7\342\202\254", 3) == 2' 
failed
../../gltests/test-mbsnlen.c:62: assertion 'mbsnlen ("7\342\202\254", 5) == 3' 
failed
../../gltests/test-mbsnlen.c:64: assertion 'mbsnlen ("\360\237\220\203", 0) == 
0' failed
../../gltests/test-mbsnlen.c:65: assertion 'mbsnlen ("\360\237\220\203", 1) == 
1' failed
../../gltests/test-mbsnlen.c:66: assertion 'mbsnlen ("\360\237\220\203", 2) == 
1' failed
../../gltests/test-mbsnlen.c:67: assertion 'mbsnlen ("\360\237\220\203", 3) == 
1' failed
../../gltests/test-mbsnlen.c:69: assertion 'mbsnlen ("\360\237\220\203", 5) == 
2' failed
../../gltests/test-mbsnlen.c:73: assertion 'mbsnlen ("\342\202", 2) == 1' failed
../../gltests/test-mbsnlen.c:74: assertion 'mbsnlen ("\360\237\220", 3) == 1' 
failed

The two attached patches make the tests pass. At this point, I don't have
time to deal with the MEE vs. SEE behaviour differences; that has to wait
for later.


2024-06-18  Bruno Haible  <bruno@clisp.org>

        mbsnlen: Add tests.
        * tests/test-mbsnlen.c: New file.
        * tests/test-mbsnlen.sh: New file, based on tests/test-mbsspn.sh.
        * modules/mbsnlen-tests: New file.

        mbsnlen: Fix bug (regression 2023-09-26).
        * lib/mbsnlen.c (mbsnlen): Fix bug in GNULIB_MCEL_PREFER implementation.

0001-mbsnlen-Fix-bug-regression-2023-09-26.patch
Description: Text Data

0002-mbsnlen-Add-tests.patch
Description: Text Data