Horton

Re: pure parsers and reentrant scanners

Peter Eisentraut <peter@eisentraut.org>

View thread
From: Peter Eisentraut <peter@eisentraut.org>
To: pgsql-hackers <pgsql-hackers@postgresql.org>
Cc: pgbuildfarm@rjuju.net, Andreas Karlsson <andreas@proxel.se>, Tom Lane <tgl@sss.pgh.pa.us>
Date: 2025-01-09T07:55:29Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Return yyparse() result not via global variable

  2. Remove flex version checks

  3. Drop warning-free support for Flex 2.5.35

  4. plpgsql: pure parser and reentrant scanner

  5. flex code modernization: Replace YY_EXTRA_TYPE define with flex option

  6. guc: reentrant scanner

  7. jsonpath scanner: reentrant scanner

  8. syncrep parser: pure parser and reentrant scanner

  9. replication parser: pure parser and reentrant scanner

  10. bootstrap: pure parser and reentrant scanner

  11. Small whitespace improvement

  12. Prevent redeclaration of typedef yyscan_t

  13. seg: pure parser and reentrant scanner

  14. cube: pure parser and reentrant scanner

Attachments

On 20.12.24 16:23, Tom Lane wrote:
>> Ok, we can fix that, but maybe this is also a good moment to think about
>> whether that is useful.  I could not reproduce the issue with flex
>> 2.5.39.  I could find no download of flex 2.5.35.  The github site only
>> offers back to 2.5.39, the sourceforce site back to 2.5.36.  lapwing
>> says it's Debian 7.0, which went out of support in 2016 and out of
>> super-duper-extended support in 2020.  It also doesn't have a supported
>> OpenSSL version anymore, and IIRC, it has a weird old compiler that
>> occasionally gives bogus warnings.  I think it's time to stop supporting
>> this.
> 
> OK, that's fair.  I do see lapwing called out a lot in the commit log,
> though it's not clear how much of that is about 32-bitness and how
> much about old tools.  It's surely still valuable to have i386
> machines in the buildfarm, but I agree that supporting unobtainable
> tool versions is a bit much.  Could we get that animal updated to
> some newer OS version?
> 
> Presumably, we should also rip out the existing yyget_column and
> yyset_column kluges in
> 
> src/backend/parser/scan.l: extern int      core_yyget_column(yyscan_t yyscanner);
> src/bin/psql/psqlscanslash.l: extern int   slash_yyget_column(yyscan_t yyscanner);
> src/bin/pgbench/exprscan.l: extern int     expr_yyget_column(yyscan_t yyscanner);
> src/fe_utils/psqlscan.l: extern int        psql_yyget_column(yyscan_t yyscanner);

All my flex-related patches are in now.

Here is a patch that removes the workarounds for compiler warnings with 
flex 2.5.35.  This ended up being a whole lot, including the whole 
fix-old-flex-code.pl script.

The second patch contemplates raising the minimum required flex version, 
but what to?

The most recent incrementing was exactly because 2.5.35 was the oldest 
in the buildfarm.  The previous incrementings were apparently because 
certain features were required or some bugs had to be avoided.

Options:

- Leave at 2.5.35 as long as it's present in the buildfarm.

- Set to 2.5.36 because it's the oldest that compiles without warnings. 
Also, the oldest you can still download from the flex sourceforge site.

- Set to 2.5.37 because that's the next oldest in the buildfarm (for 
CentOS/RHEL 7, so it will stay around for a while).

- Set to 2.5.34 because that's the oldest we actually require as of 
commit b1ef48980dd.

- Remove version check, because these are all so old that no one cares 
anymore.