MDEV-13594 (was MDEV-18530): Implement -> and ->> JSON path operators

[kjarir]

Summary:
Adds support for MySQL-compatible JSON path operators '->' and '->>' as aliases for JSON_EXTRACT and JSON_UNQUOTE(JSON_EXTRACT).

Key Changes:

• Lexer: Modified MINUS_OR_COMMENT in sql_lex.cc to detect '->' and '->>'.
• Parser: Restructured primary_expr in sql_yacc.yy to support operator chaining without shift/reduce conflicts.
• Symbols: Added JSON_ARROW_SYM and JSON_UNQUOTED_ARROW_SYM to lex.h.
• Tests: Added comprehensive verification for literal/column extraction, chaining, and usage in VIRTUAL columns with INDEX in mysql-test/main/json_operators.test.
  Labels: gsoc26

[CLAassistant]

All committers have signed the CLA.

[kjarir]

Fixed a syntax error in the newly added test file mysql-test/main/json_operators.test where some comments were being misinterpreted by the mysqltest runner. All CI checks should now progress correctly.

[kjarir]

I've reviewed the CI logs and found that the Buildbots and AppVeyor were failing due to a mismatch in the sys_vars.ft_boolean_syntax_basic test result. Because the lexer now parses -> as a single token (JSON_ARROW_SYM), a test case that assigned an unquoted full-text boolean syntax + -><...; threw a slightly different syntax error (near '-><' instead of near '><'). I have updated the ft_boolean_syntax_basic.result file to reflect this new, expected behavior and pushed the fix. The CI should now pass successfully.

[grooverdan]

Well done on implementation with a good start to tests.

Do you know where these operators show up in terms of https://mariadb.com/docs/server/reference/sql-structure/operators/operator-precedence ? Can there be test cases around validate they stay in that order (if possible)?

Do the results of these show up the same results as the MySQL implementation? (mysql-test/suite/json/inc/json_functions.inc). If you include these into MariaDB, adapting to its form (so maybe not an include if not needed), in their own commit. Include authorship and credit to the original Oracle authors and include what changes you made also in commit message.

[grooverdan]

This doesn't use or implement any specific innodb features so this can be removed.

The MDEV header below should be echo statements.
MDEV-18530 is a MDEV closed as a duplicate therefore using the MDEV-13594 MDEV is recommented.

After adding a test case like this, mtr --record and then commit the json_operators.result file in the same commit.

[grooverdan]

as these may be ERROR conditions there will be a --error before the failing SQL statements with the error they genenrate.

[grooverdan]

Don't really need to say "in MySQL"

[grooverdan]

The final statement in a test case is"

--echo End of 13.0 tests

For the purpose of ease of merging. The end of a particular MDEV's tests becomes obvious with the beginning of the next MDEV.

[grooverdan]

There is a list memory allocation to both a local variable, that if one of the two push_backs fail, becomes a leak. The other allocations have this assigned to a lex structure of $$ at the time MYSQL_YYABORT is called.

[grooverdan]

So its technically not a ER_INVALID_JSON_PATH.

JSON_EXTRACT is documented to returning NULL and warnings on invalid paths, so its consistent.

[SORVER]

@grooverdan
Hey everyone, I thought this was a GSoC 2026 project. Should I start working on it for GSoC, or is it now done?

[grooverdan]

@grooverdan Hey everyone, I thought this was a GSoC 2026 project. Should I start working on it for GSoC, or is it now done?

It was intended to be, but its pretty much done, so there's no point starting on it now.

Let's keep discussion of GSoC, any anything not related with the review of this PR on https://mariadb.zulipchat.com.

There are other opportunities, discuss on the zulip chat to work out a mutual interest.