Thread: Re: pgsql: Some refactoring to export json(b) conversion functions

Re: pgsql: Some refactoring to export json(b) conversion functions

From
Alvaro Herrera
Date:
On 2023-Jul-26, Amit Langote wrote:

> Some refactoring to export json(b) conversion functions
> 
> This is to export datum_to_json(), datum_to_jsonb(), and
> jsonb_from_cstring(), though the last one is exported as
> jsonb_from_text().

After this commit, Coverity started complaining that
datum_to_jsonb_internal() leaks the JsonLexContext here

 754   │             case JSONTYPE_CAST:
 755   │             case JSONTYPE_JSON:
 756   │                 {
 757   │                     /* parse the json right into the existing result object */
 758   │                     JsonLexContext *lex;
 759   │                     JsonSemAction sem;
 760   │                     text       *json = DatumGetTextPP(val);
 761   │ 
 762   │                     lex = makeJsonLexContext(json, true);
 763   │ 
 764   │                     memset(&sem, 0, sizeof(sem));
 765   │ 
 766   │                     sem.semstate = (void *) result;
 767   │ 
 768   │                     sem.object_start = jsonb_in_object_start;
 769   │                     sem.array_start = jsonb_in_array_start;
 770   │                     sem.object_end = jsonb_in_object_end;
 771   │                     sem.array_end = jsonb_in_array_end;
 772   │                     sem.scalar = jsonb_in_scalar;
 773   │                     sem.object_field_start = jsonb_in_object_field_start;
 774   │ 
 775   │                     pg_parse_json_or_ereport(lex, &sem);
 776   │                 }
 777   │                 break;

Admittedly, our support code for this is not great, since we have no
clean way to free those resources.  Some places like json_object_keys
are freeing everything manually (even though in that particular case
it's unnecessary, since that one runs in a memcxt that's going to be
cleaned up shortly afterwards).

One idea that Tom floated was to allow the JsonLexContext to be
optionally stack-allocated.  That reduces palloc() traffic; but some
callers do need it to be palloc'ed.  Here's a patch that does it that
way, and adds a freeing routine that knows what to do in either case.
It may make sense to do some further analysis and remove useless free
calls.


It may make sense to change the structs that contain JsonLexContext *
so that they directly embed JsonLexContext instead.  That would further
reduce palloc'ing.

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/
"Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end." (2nd Commandment for C programmers)

Attachment

Re: pgsql: Some refactoring to export json(b) conversion functions

From
Alvaro Herrera
Date:
On 2023-Aug-08, Alvaro Herrera wrote:

> One idea that Tom floated was to allow the JsonLexContext to be
> optionally stack-allocated.  That reduces palloc() traffic; but some
> callers do need it to be palloc'ed.  Here's a patch that does it that
> way, and adds a freeing routine that knows what to do in either case.
> It may make sense to do some further analysis and remove useless free
> calls.

Pushed this, after some further tweaking.

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
"La verdad no siempre es bonita, pero el hambre de ella sí"