[Monotone-devel] Solaris breakage

[William Uther]

Hi,
  Just taking a break from paper editing...  I decided to look at  
the solaris breakage.  It makes no sense to me at all.

  The test that is failing is "clone_warning_with_multiple_heads".   
The part of the test that is failing is where the checked out  
directory is removed after the failure - the directory is being left  
around.  In particular, from the buildbot we get:

mtn: setting default server to file:/zfs_space/export/home/mtbldbot/ 
slave-dir/fu
ll-sparc-solaris10-sunstudio11/build/tester_dir/ 
clone_warning_with_multiple_head
s/test.db
mtn: setting default branch include pattern to 'testbranch'
mtn: connecting to file:/zfs_space/export/home/mtbldbot/slave-dir/ 
full-sparc-sol
aris10-sunstudio11/build/tester_dir/clone_warning_with_multiple_heads/ 
test.db
mtn: finding items to synchronize:
mtn: ticks: >="bytes in"/1024, <="bytes out"/1024, c="certs in"/3,  
r="revs in"/1
mtn: ><crrcr>crcc
mtn: successful exchange with file:/zfs_space/export/home/mtbldbot/ 
slave-dir/ful
l-sparc-solaris10-sunstudio11/build/tester_dir/ 
clone_warning_with_multiple_heads
/test.db
mtn: branch testbranch has multiple heads:
mtn:   6d0f3f362b7328e94dff9fef549bbdf0095b2cd6 address@hidden  
2007-07-24T03:59
:28
mtn:   89103abeb431f83da74d9b1b99a1609c530643e1 address@hidden  
2007-07-24T03:59
:26
mtn: choose one with 'mtn checkout -r<id>'
mtn: error: could not remove '/zfs_space/export/home/mtbldbot/slave- 
dir/full-spa
rc-solaris10-sunstudio11/build/tester_dir/ 
clone_warning_with_multiple_heads/test
_dir1': Invalid argument


This is correct down to the last line.  The "could not remove" error  
comes from clone trying to remove the test_dir1 directory it  
partially checked out.  This attempted using delete_dir_recursive()  
defined around line 280 of file_io.cc.  Note that the first thing  
that function does is check that the directory to remove both exists  
and is actually a directory - i.e. the problem is not with clone  
passing in the wrong path.

I have no idea why the recursive remove is failing on Solaris.  The  
final failure prints out an os_strerror(errno) of "Invalid  
argument".  We know the file exists.  Why else could the argument be  
invalid?

I thought this patch might help illuminate things:

--- file_io.cc  358fb964dc42e7f936a136342b93bb3a4744d1b3
+++ file_io.cc  d91cbbab9f40622e4c264becb0e2067ee218738e
@@ -269,6 +269,9 @@ do_remove_recursive(any_path const & p)
        i != subdirs.end(); i++)
     do_remove_recursive(p / *i);

+  if (!directory_empty(p))
+    W(F("When attempting to remove directory '%s', we were  
unsuccessful removing all the files inside it.") % p);
+
   do_remove(p.as_external());
}

but I'm wary of committing that as it will make failures very verbose  
and slow down every recursive remove.

I'm at a loss to suggest why the recursive remove might be failing.   
I could imagine the remove of the database inside that dir failing  
(which is why the test is marked xfail on windows), but we don't get  
an error removing the db, we get an error later removing the top  
level dir.

I hope that helps people on Solaris debug.  Unfortunately, I don't  
have a solaris box easily accessible to help with.

Cheers,

Will           :-}