Re: Crash with empty dictionary

Bruce Momjian wrote:
> Tom Lane wrote:
>> "Hamid Quddus Akhtar" <hamid(dot)akhtar(at)enterprisedb(dot)com> writes:
>>>> Offhand I'd say that an empty file is a legitimate corner case,
>>>> so we should just take it silently.
>>> Shouldn't we be warning about an empty file rather than just swallowing
>>> up the error?
>> You are jumping to a conclusion, namely that it is an error. If it's
>> a legitimate corner case, throwing a warning every time the file is
>> read would be incredibly annoying.
>>
>> If it's not a legitimate case, then we should throw a real error.
>> A warning just strikes me as the worst of both worlds.
>
> A zero-length file seems fine to me in this case.

It also seems to have problems with an affix-file with a single entry.

Looking closer at the tmpCtx hack, it looks like it can't just be
replaced by setting CurrentMemoryContext to a temporary context. Some
stuff needs to be allocated in the ts cache entry's dictCtx, while other
stuff is temporary. I'll try to at least comment it.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com