Hi,

I've now kicked out uthash in favour of OpenBSD's native ohash.  (I've 
included espie@ in this just in case he has any ohashy wisdom.)  This 
requires that I add a compat_ohash.c file, but this is trivial.

The difference is as small as expected.  In fact, pretty much all that 
needed tweaking was making struct of's file pointer an array (since 
ohash requires that keys be within the structures--something that needs 
stating twice because I forgot) and using a dummy array declaration for 
struct str for the same reason:

   struct  str {
         ...stuff here...
         char key[1];
   };

...with allocation in the regular way (the struct size plus the string 
length).  Is this known to play nicely with all compilers?  I can't 
think of a better way without modifying the ohash routines to work with 
key pointers.

The biggest bottleneck right now is the database.  Try running mandocdb 
with -n to see the magnitude of difference.  Ugh!  The -n is checked 
directly prior to database stuff ("nodb" in the source), so there's 
little else going on there.  I don't do any fancy sqlitey stuff, so 
there's room for improvement.

Note that Marc's hash-growth is less greedy than uthash's, so I can run 
it as a normal user without running out of memory.  Also, I initially 
thought I'd have to switch from K&R's hash (ohash's default), but it 
does a pretty good job especially considering that size of the sqlite3 
bottleneck.

Best,

Kristaps