Re: Loading binary data into the database

Adriaan Joubert <a(dot)joubert(at)albourne(dot)com> writes:
>> Such as? IIRC, the reason we disabled it was precisely that there was
>> no support on the client side. (What's worse, there's no support in
>> the FE/BE protocol either. I don't see how you could have made this
>> work...)

> I issued a 'copy binary <table> from stdin;' and then sent the data with
> PQputnbytes

How did you get out of COPY state? In binary mode CopyFrom will only
recognize EOF as end of data, and there's no provision in the FE/BE
protocol for making it see an EOF. You'd have had to break the
connection to get out of that --- and I'd have expected the loss of
connection to cause a transaction abort, preventing your data from
getting committed. (If it didn't abort, that's a bug that needs to be
fixed... if the line drops halfway through a copy, you don't want it
to commit do you?)

The real bottom line here is that the FE/BE protocol would need to be
changed to support binary copy properly, and no one's excited about
putting more work into the existing protocol, nor about the ensuing
compatibility problems.

> Perhaps libpq is not the answer. I've even been thinking about writing a
> SPI function that acts as a CORBA server -- but decided that that is just
> too ugly to contemplate. So what is the solution?

A CORBA-based replacement protocol has been discussed seriously, though
I haven't noticed any work getting done on it lately. Feel free to
pitch in if you think it's a good idea.

regards, tom lane