Re: Correction: datatypes are not "faster"

On Thu, 2010-09-02 at 15:24 -0700, Josh Berkus wrote:
> >> This paragraph leaves a *lot* to be desired from an accuracy perspective
> >
> > Really?  Exactly which statements will you claim are incorrect?
>
> That the int type is definitely faster on all platforms regardless of
> circumstances.  Especially the circumstance where the user really needs
> a bigint and is doing some wonky workaround to use int instead, like the
> newbie I'm chatting with on IRC right now, who did so specifically
> because of that page of the docs.

How do you know that the newbie, as you call him, needs a bigint. He was
only trying to optimize his database according to his needs and what he
read on the documentation. He might not be a postgresql master but at
least he tries to use postgresql at its best. Being wrong does not make
him a newbie but someone who learnt something which is a pretty common
thing in computer science.

The "newbie".

>
> Besides which, datatypes are not "faster".  Specific operations with
> them may be faster.   They may require less storage and less RAM.  But
> if we call them "faster", then we're contributing to application
> developer ignorance.
>
> > I notice that your proposed rewrite omits the bit about bigint being
> > slower, which I can only conclude means you haven't tested on any
> > 32-bit platforms lately.
>
> Hmmm. Yes, but that's more of an exception now than it is a common
> circumstance.  Change this:
>
> "On very minimal operating systems the bigint type might not function
> correctly, because it relies on compiler support for eight-byte
> integers. On such machines, bigint acts the same as integer, but still
> takes up eight bytes of storage. (We are not aware of any modern
> platform where this is the case.)"
>
> To this:
>
> On 32-bit operating systems, or when PostgreSQL is complied 32-bit,
> operations using bigint can be significantly slower than those with
> integer.  On very minimal operating systems the bigint type might not
> function correctly, because it relies on compiler support for eight-byte
> integers. On such machines, bigint acts the same as integer, but still
> takes up eight bytes of storage. (We are not aware of any modern
> platform where this is the case.)
>
>
>
> --
>                                   -- Josh Berkus
>                                      PostgreSQL Experts Inc.
>                                      http://www.pgexperts.com
>