Horton

Re: Another regexp performance improvement: skip useless paren-captures

Mark Dilger <mark.dilger@enterprisedb.com>

View thread View on postgr.es
From: Mark Dilger <mark.dilger@enterprisedb.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Andrew Dunstan <andrew@dunslane.net>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Joel Jacobson <joel@compiler.org>
Date: 2021-08-09T23:37:18Z
Lists: pgsql-hackers 

> On Aug 9, 2021, at 12:14 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> 
> Pushed, but while re-reading it before commit I noticed that there's
> some more fairly low-hanging fruit in regexp_replace().  As I had it
> in that patch, it never used REG_NOSUB because of the possibility
> that the replacement string uses "\N".  However, we're already
> pre-checking the replacement string to see if it has backslashes
> at all, so while we're at it we can check for \N to discover if we
> actually need any subexpression match data or not.  We do need to
> refactor a little to postpone calling pg_regcomp until after we
> know that, but I think that makes replace_text_regexp's API less
> ugly not more so.
> 
> While I was at it, I changed the search-for-backslash loops to
> use memchr rather than handwritten looping.  Their use of
> pg_mblen was pretty unnecessary given we only need to find
> backslashes, and we can assume the backend encoding is ASCII-safe.
> 
> Using a bunch of random cases generated by your little perl
> script, I see maybe 10-15% speedup on test cases that don't
> use \N in the replacement string, while it's about a wash
> on cases that do.  (If I'd been using a multibyte encoding,
> maybe the memchr change would have made a difference, but
> I didn't try that.)

I've been reviewing and testing this (let-regexp_replace-use-NOSUB.patch) since you sent it 4 hours ago, and I can't seem to break it.  There are pre-existing problems in the regex code, but this doesn't seem to add any new breakage.

—
Mark Dilger
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Commits

  1. Fix regexp misbehavior with capturing parens inside "{0}".

  2. Let regexp_replace() make use of REG_NOSUB when feasible.

  3. Fix bogus assertion in BootstrapModeMain().

  4. Avoid determining regexp subexpression matches, when possible.

  5. Check the size in COPY_POINTER_FIELD

  6. Make regexp engine's backref-related compilation state more bulletproof.