On Thu, Jan 12, 2012 at 9:23 AM, Doug Evans wrote: > The symfile.c change is ok with me, modulo can you add comments for > each of the functions? Done. I've also refactored the code a bit more to avoid searching the same directory again. > [caveat: I think it doesn't properly handle paths with dos drives, but > the code already has that problem, so no requirement to fix that in > this patch] I don't think ".gnu_debuglink" is supported on non-ELF platforms, so not handling DOS drives is likely not a problem. > I see sepdebug.exp uses remote_exec so best use that instead of exec. Done. > Also, I'm not sure what the portability requirements are w.r.t. symlinks. > Probably best to watch for errors in the "ln -s" and handle appropriately. The test exits if gdb_gnu_strip_debug fails. I don't believe it will currently succeed on any platform without symlinks. Testing for errors will just add noise. > [Even better, is there a utility routine that will create a symlink? > All the portability concerns could be tucked away in there.] I didn't find one. > Also, can you use clean_restart instead of the gdb_exit/gdb_start/... sequence? Done. Thanks, -- Paul Pluzhnikov 2012-01-11 Paul Pluzhnikov PR gdb/9538 * symfile.c (find_separate_debug_file): New function. (terminate_after_last_dir_separator): Likewise. (find_separate_debug_file_by_debuglink): Also try realpath. testsuite/ChangeLog: PR gdb/9538 * gdb.base/sepdebug.exp: New test.