Re: Replace some cstring_to_text to cstring_to_text_with_len

From: Dagfinn Ilmari Mannsåker <ilmari(at)ilmari(dot)org>
To: Ranier Vilela <ranier(dot)vf(at)gmail(dot)com>
Cc: Andrew Dunstan <andrew(at)dunslane(dot)net>, John Naylor <john(dot)naylor(at)enterprisedb(dot)com>, Michael Paquier <michael(at)paquier(dot)xyz>, Pg Hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Replace some cstring_to_text to cstring_to_text_with_len
Date: 2023-08-31 15:12:14
Message-ID: 87a5u7p7ox.fsf@wibble.ilmari.org
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Ranier Vilela <ranier(dot)vf(at)gmail(dot)com> writes:

> Em qui., 31 de ago. de 2023 às 10:12, Dagfinn Ilmari Mannsåker <
> ilmari(at)ilmari(dot)org> escreveu:
>
>> Andrew Dunstan <andrew(at)dunslane(dot)net> writes:
>>
>> > On 2023-08-31 Th 07:41, John Naylor wrote:
>> >>
>> >> On Thu, Aug 31, 2023 at 6:07 PM Ranier Vilela <ranier(dot)vf(at)gmail(dot)com>
>> wrote:
>> >> >
>> >> > Em qui., 31 de ago. de 2023 às 00:22, Michael Paquier
>> >> <michael(at)paquier(dot)xyz> escreveu:
>> >> >>
>> >> >> On Wed, Aug 30, 2023 at 03:00:13PM -0300, Ranier Vilela wrote:
>> >> >> > cstring_to_text has a small overhead, because call strlen for
>> >> >> > pointer to char parameter.
>> >> >> >
>> >> >> > Is it worth the effort to avoid this, where do we know the size
>> >> of the
>> >> >> > parameter?
>> >> >>
>> >> >> Are there workloads where this matters?
>> >> >
>> >> > None, but note this change has the same spirit of 8b26769bc.
>> >>
>> >> - return cstring_to_text("");
>> >> + return cstring_to_text_with_len("", 0);
>> >>
>> >> This looks worse, so we'd better be getting something in return.
>> >
>> >
>> > I agree this is a bit ugly. I wonder if we'd be better off creating a
>> > function that returned an empty text value, so we'd just avoid
>> > converting the empty cstring altogether and say:
>> >
>> > return empty_text();
>>
>> Or we could generalise it for any string literal (of which there are
>> slightly more¹ non-empty than empty in calls to
>> cstring_to_text(_with_len)):
>>
>> #define literal_to_text(str) cstring_to_text_with_len("" str "",
>> sizeof(str)-1)
>>
> I do not agree, I think this will get worse.

How exactly will it get worse? It's exactly equivalent to
cstring_to_text_with_len("", 0), since sizeof() is a compile-time
construct, and the string concatenation makes it fail if the argument is
not a literal string.

Whether we want an even-more-optimised version for an empty text value
is another matter, but I doubt it'd be worth it. Another option would
be to make cstring_to_text(_with_len) static inline functions, which
lets the compiler eliminate the memcpy() call when len == 0.

In fact, after playing around a bit (https://godbolt.org/z/x51aYGadh),
it seems like GCC, Clang and MSVC all eliminate the strlen() and
memcpy() calls for cstring_to_text("") under -O2 if it's static inline.

- ilmari

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Robert Haas 2023-08-31 15:25:33 sandboxing untrusted code
Previous Message Hayato Kuroda (Fujitsu) 2023-08-31 14:26:21 RE: [PoC] pg_upgrade: allow to upgrade publisher node