Fedora 18. bzr pull from lp:maria trunk
bzr version-info
revision-id: svoj@mariadb.org-20130627111848-b8kc0z60ywwpqavz
date: 2013-06-27 15:18:48 +0400
build-date: 2013-06-29 15:42:46 -0400
revno: 3762
branch-nick: trunk

Just in case, I re-checked it on Fedora 18 64-bit with BUILD/compile-pentium64 (as discussed on IRC). As expected, still no failure.

According to mysqltest code, the trailing slash should not be needed. The test fails to remove files in remove_files_wildcard

191: --echo # remove tmpdir
192: --remove_files_wildcard $tmpdir *
193: --rmdir $tmpdir

But mysqltest in do_remove_files_wildcard explicitly adds a separator after a dirname:

init_dynamic_string(&ds_file_to_remove, dirname, 1024, 1024);
dir_separator[0]= FN_LIBCHAR;
dynstr_append_mem(&ds_file_to_remove, dir_separator, 1);
directory_length= ds_file_to_remove.length;

And FN_LIBCHAR is defined in my_global.h:

#ifdef _WIN32
#define FN_LIBCHAR '
'
...
#else
#define FN_LIBCHAR '/'
...
#endif

Since you are building from sources anyway, can you add a debugging print or run mysqltest under debugger to see what your FN_LIBCHAR is in mysqltest.cc:do_remove_files_wildcard ?

Could you please also run the following test to see how it goes?

--let $dirname= `SELECT uuid()`
--let $tmpdir= $MYSQLTEST_VARDIR/$dirname
--mkdir $tmpdir
--write_file $tmpdir/abcd.file
test
EOF

--list_files $MYSQLTEST_VARDIR/$dirname *

--remove_files_wildcard $tmpdir *
--rmdir $tmpdir
--echo # All good

1. End of test

Thanks.

I dug a bit further into both the rpl_binlog_index test failure and the failure you sent me. They both fail on BUILD/compile-pentium64-debug. While it was in debug mode I was able to trace the problem down to some strange code in mysys/my_lib.c:102 it does a comparison against FN_LIBCHAR + 1 which seems to fail. Below that it would add a trailing slash if it succeeds. By removing the + 1 the test you gave me and rpl_binlog_index both succeed.

The code in mysqltest.cc to add the trailing slash happens after the call to my_dir.

Thanks Nice finding.
It so happens that your uuid() end with a zero (mine don't); and the comparison with FN_LIBCHAR + 1 is basically with 0, right?

When I force a directory name to add with 0, it fails for me too. And I suppose for you the one below will pass (the only difference with the previous one is that I concatenate uuid() with '1':

--let $dirname= `SELECT CONCAT(uuid(),'1')`
--echo # Dirname: $dirname
--let $tmpdir= $MYSQLTEST_VARDIR/$dirname
--mkdir $tmpdir
--write_file $tmpdir/abcd.file
test
EOF

--list_files $MYSQLTEST_VARDIR/$dirname *

--remove_files_wildcard $tmpdir *
--rmdir $tmpdir
--echo # All good

I reworked directory_file_name to compare the ending character against '/' and it passes tests with my 0 uuid. The patch for this version is MDEV_4734_2.patch

Only 10.0 is affected, there was some refactoring in there.

Test case for completion (be careful while copying it, make sure there are no leading or trailing spaces):

--let $mydir= $MYSQLTEST_VARDIR/dir0
--mkdir $mydir
--write_file $mydir/foo
bar
EOF

--remove_files_wildcard $mydir *
--rmdir $mydir

--echo # All good

See the patch proposed by Eric.