Обсуждение: replace_text optimization (StringInfo to varlena)

  Hi,

replace_text() in varlena.c builds the result in a StringInfo buffer,
and finishes by copying it into a freshly allocated varlena structure
with cstring_to_text_with_len(), in the same memory context.

It looks like that copy step could be avoided by preprending the
varlena header to the StringInfo to begin with, and return the buffer
as a text*, as in the attached patch.

On large strings, the time saved can be significant. For instance
I'm seeing a ~20% decrease in total execution time on a test with
lengths in the 2-3 MB range, like this:

  select sum(length(
    replace(repeat('abcdefghijklmnopqrstuvwxyz', i*10), 'abc', 'ABC')
   ))
  from generate_series(10000,12000) as i;

Also, at a glance, there are a few other functions with similar
StringInfo-to-varlena copies that seem avoidable:
concat_internal(), text_format(), replace_text_regexp().

Are there reasons not to do this? Otherwise, should it be considered
in in a more principled way, such as adding to the StringInfo API
functions like void InitStringInfoForVarlena() and
text *StringInfoAsVarlena()?


Best regards,
--
Daniel Vérité
PostgreSQL-powered mailer: http://www.manitou-mail.org
Twitter: @DanielVerite

At Wed, 13 Feb 2019 16:38:50 +0100, "Daniel Verite" <daniel@manitou-mail.org> wrote in
<bc319ec6-60d0-4878-a800-bcc12a190c02@manitou-mail.org>
>   Hi,
> 
> replace_text() in varlena.c builds the result in a StringInfo buffer,
> and finishes by copying it into a freshly allocated varlena structure
> with cstring_to_text_with_len(), in the same memory context.
> 
> It looks like that copy step could be avoided by preprending the
> varlena header to the StringInfo to begin with, and return the buffer
> as a text*, as in the attached patch.
> 
> On large strings, the time saved can be significant. For instance
> I'm seeing a ~20% decrease in total execution time on a test with
> lengths in the 2-3 MB range, like this:
> 
>   select sum(length(
>     replace(repeat('abcdefghijklmnopqrstuvwxyz', i*10), 'abc', 'ABC')
>    ))
>   from generate_series(10000,12000) as i;
> 
> Also, at a glance, there are a few other functions with similar
> StringInfo-to-varlena copies that seem avoidable:
> concat_internal(), text_format(), replace_text_regexp().
> 
> Are there reasons not to do this? Otherwise, should it be considered
> in in a more principled way, such as adding to the StringInfo API
> functions like void InitStringInfoForVarlena() and
> text *StringInfoAsVarlena()?

First, I agree that the waste of cycles should be eliminated.

Grepping with 'cstring_to_text_with_len\(.*[\.>]data,.*\)' shows
many instances of the use. Though StringInfo seems very
open-minded, the number of instances would be a good reason to
have new API functions.

That is, I vote for providing a set of API for the use in
StringInfo. But it seems to be difficult to name the latter
function. The name convention for the object is basically
<verb>StringInfo. getVarlenaStringInfo/getTextStringInfo
apparently fits the convention but seems to me a bit strange.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center