Upgrade to Postgres 17

[msepga]

Noteworthy commits that change the build/test system:

• Add support for abstract nodes in generator scripts
  • Postgres 17 adds JsonTablePlan*, an abstract sub-type of Node*
• Add disableOnMsvc option to fingerprint test generator
  • In Add support for running on Windows #233, we disabled one of the tests for MSVC, but regenerating the fingerprint tests source removed this guard. This commit adds it as part of the generator.
• Ignore name_location in deparse tests
  • New nodes in Postgres 17 include name_location, a type that needs to be cleared in deparse tests just like location.

[lelit]

Hi, I'm quite slowly trying to upgrade my pglast tool to this branch, having some problem here and there with the new abstract nodes, but I'm confident I'll get there.

One of the first failing tests is a worrisome segmentation fault: the test simply exercises the protobuf parse/deparse entry points (ie, it just calls pg_query_parse_protobuf() and then pg_query_deparse_protobuf(), with very little massaging on the pglast side), passing a trivial query such as SELECT 1.

So I morphed your examples/simple.c into the following deparse.c:

// Welcome to the easiest way to parse an SQL query :-)
// Compile the file like this:
//
// cc -I../ -L../ deparse.c -lpg_query

#include <pg_query.h>
#include <stdio.h>
#include <stdlib.h>

size_t testCount = 6;
const char* tests[] = {
  "SELECT 1",
  "SELECT * FROM x WHERE z = 2",
  "SELECT 5.41414",
  "SELECT $1",
  "SELECT 999999999999999999999::numeric/1000000000000000000000",
  "SELECT 4790999999999999999999999999999999999999999999999999999999999999999999999999999999999999 * 9999999999999999999999999999999999999999999999999999999999999999999999999999999999999999"
};

int main() {
  PgQueryProtobufParseResult result;
  size_t i;

  for (i = 0; i < testCount; i++) {
    result = pg_query_parse_protobuf(tests[i]);

    if (result.error) {
      printf("error: %s at %d of query %s\n",
             result.error->message,
             result.error->cursorpos,
             tests[i]);
    } else {
      PgQueryDeparseResult deparse_result = pg_query_deparse_protobuf(result.parse_tree);
      if (deparse_result.error) {
        printf("\nERROR for \"%s\"\n  %s\n",
               tests[i],
               deparse_result.error->message);
        pg_query_free_deparse_result(deparse_result);
      } else {
        printf("\n%s --> %s\n",
               tests[i],
               deparse_result.query);
        pg_query_free_deparse_result(deparse_result);
      }
    }

    pg_query_free_protobuf_parse_result(result);
  }

  pg_query_exit();

  return 0;
}

And it fails in the same way:

Program received signal SIGSEGV, Segmentation fault.
0x000055555559f4ab in pg_query_protobuf_to_nodes (protobuf=...) at src/pg_query_readfuncs_protobuf.c:169
169		if (result->n_stmts > 0)
(gdb) bt
#0  0x000055555559f4ab in pg_query_protobuf_to_nodes (protobuf=...) at src/pg_query_readfuncs_protobuf.c:169
#1  0x000055555558e532 in pg_query_deparse_protobuf (parse_tree=...) at src/pg_query_deparse.c:19
#2  0x000055555558e171 in main ()

What am I missing here?

Thanks&bye.

[msepga]

Hi @lelit, thanks for the test!

I can't seem to reproduce this here. I've tried the example on macOS (M1 ARM chip) with both DEBUG=1 and regular optimized builds, as well is in an x86_64 ubuntu VM.

Can you try running make clean && make libpg_query.a just to be sure it's not some kind leftover compiled object from a previous release, interfering with the test?

[lelit]

Can you try running make clean && make libpg_query.a just to be sure it's not some kind leftover compiled object from a previous release, interfering with the test?

That fixed it indeed, thank you: didn't come to mind given that I was operating in a (evidently not enterely) fresh clone of pglast, of which libpg_query is a submodule!

[lfittl]

👍

Outstanding TODO:

• Add PL/pgSQL test for array types
• Update Makefile to copy PL/pgSQL regression tests automatically