Bruce Momjian <pgman@candle.pha.pa.us> writes:
> I found I didn't need to clear the date cache.
>>
>> Hmm, are you sure about that? I'm not.
> I checked and it caches a pointer to the struct, not the values
> themselves, and we don't change the structure, just the secondary values
> and not the key used by the bsearch.
Now I'm going to object LOUDLY. You cannot convince me that the above
is a good implementation --- it's a complete crock, and will break the
instant someone looks at it sidewise.
My inclination would actually be to rip out the cache entirely. bsearch
in a table this size is not so expensive that we need to bypass it, nor
is it apparent that we are going to see lots of successive lookups for
the same keyword anyway. How long has that cache been in there, and
what was the motivation for adding it to begin with?
regards, tom lane