Re: sqlite3 mandocdb and apropos (etc.)

[Kristaps Dzonsons <kristaps@bsd.lv>]

On 06/06/2012 00:17, Ingo Schwarze wrote:
> Hi Kristaps,
>
> Kristaps Dzonsons wrote on Tue, Jun 05, 2012 at 04:32:17PM +0200:
>
>> Enclosed [finally] is a patch against HEAD that uses sqlite3 for
>> mandocdb and apropos (getting the others to use it is quite easy).
>> It's a work in progress, but is the simplest possible implementation
>> before extensive polishing.  I'd like to get this in to start
>> refining.
>
> I am interested in the sqlite3 integration and hope we get it
> going.  Certainly, this will need a thorough review, and i hope
> to find some time for it.  As i couldn't find the time for a full
> review right now, i'm sending comments on one aspect that became
> obvious at once.
>
>> First of all, you'll need uthash (in ports) and sqlite3 to build.  I
>> chose uthash for hashtables just because I'm familiar with it.  I'd
>> like to use espie@'s ohash stuff instead to reduce dependencies, but
>> since uthash is just a header file, I'm not that concerned.
>
> First point, the base system cannot ever use anything from ports.
> Thus, using uthash would require to import it into base and maintain
> it there.  So i had a brief look at it - and even without
> understanding what it does, i think there is no way in hell to ever
> get that imported.  The file uthash.h alone is 900 lines of abusing
> cpp(1) macro definitions instead of defining proper functions.  I
> cannot imagine that you will find anybody willing to do the required
> code audit on code of that style, let alone build consensus to import
> such code; myself, i must say it feels positively revolting on first
> sight.
>
> Of course, i might be wrong in my first impression, but if you
> suspect i am wrong and this is actually code of sufficient quality
> to be considered for inclusion in base, i'd recommend you first
> verify with some of the usual suspects for userland library hacking
> (like millert@, otto@, guenther@, deraadt@, ...) before spending
> a lot of time integrating it and then being disappointed later.
>
> Otherwise, it's probably better to choose tools available in base
> first and use those, at least for now, to avoid changing too much
> in a single step.

Ingo,

As mentioned, although important as a concept, this is a pretty 
insignificant part of the code---I used uthash simply because it was 
lightweight in terms of interfacing.  It's not a final solution, nor is 
it, as I've found, reasonable for optimistic growth---it didn't let me 
control for allocation failure, which is a must for letting the string 
table fill memory then purge when it overgrows.

So as said, I'll probably end up with espie@'s ohash, which from what I 
understand uses Knuth's AofP algorithm and lets me control the hash 
allocation.

Best,

Kristaps
--
 To unsubscribe send an email to tech+unsubscribe@mdocml.bsd.lv