Re: hissqlite (was Re: ovsqlite)

Bo Lindbergh <[email protected]> writes:
> Quoth Russ Allbery <[email protected]>:

>> (I have been dubious for a while whether that level of indirection in the
>> build system is worth it.  The theory is that it makes it easy to drop in
>> a new backend, and makes it easy to deactivate backends, but honestly, how
>> often do we do that?  One option might be to get rid of that indirection
>> entirely.)

> And it's apparently too hard to get right.
> From storage/buildconfig.in before I ever touched it:
>                                            vvvvvvv
>>             if (defined $$config{number}{$number}) {
>>                 die "$dir/$file: method number $number was already "
>>                     . "allocated in $$config{number}{$number}\n";
>>             }
>>             $$config{number}{$dir} = $number;
>                                ^^^^
> Checking $$config{number}{$number} for collisions doesn't work
> when the actual assignment is to $$config{number}{$dir}.

Heh.  Yeah, all the pieces of that code that were never tested because we
never dynamically add new methods.

-- 
Russ Allbery ([email protected])             <https://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <https://www.eyrie.org/~eagle/faqs/questions.html> explains why.