Re: BUG: jsonpath .split_part() bypasses lax-mode error suppression - Mailing list pgsql-hackers

From Dmitry Dolgov
Subject Re: BUG: jsonpath .split_part() bypasses lax-mode error suppression
Date
Msg-id 2cqv3uotlvjsjkkjganonpbghopgjwqqrwwylujmqu3qca7asn@smj4ghtqigpi
Whole thread
In response to BUG: jsonpath .split_part() bypasses lax-mode error suppression  (SATYANARAYANA NARLAPURAM <satyanarlapuram@gmail.com>)
List pgsql-hackers
> On Sat, Apr 18, 2026 at 09:31:03AM -0700, SATYANARAYANA NARLAPURAM wrote:
> Hi hackers,
> 
> The jsonpath string method  .split_part() bypasses lax-mode error
> suppression.
> This is because executeStringInternalMethod() uses
> DirectFunctionCall1(numeric_int4, ...)
> to convert the field number from numeric to int4. This throws
> ereport(ERROR) directly,
> bypassing the jspThrowErrors(cxt) / RETURN_ERROR mechanism that other
> methods
> like .double() use correctly.
> 
> Reproduction:
> -- These should return NULL in lax mode, but throw hard ERRORs:
> SELECT '"hello-world"'::jsonb @? '$.split_part("-", 99999999999)';
> SELECT '"hello-world"'::jsonb @? '$.split_part("-", 0)';
> ERROR:  integer out of range
> ERROR:  field position must not be zero

I don't have a SQL standard at hands, it might be worth checking what SQL/JSON
says about such situations. But at least from the definition of lax mode, given
in the documentation, current behavior seems to be correct:

    lax (default) — the path engine implicitly adapts the queried data to
    the specified path. Any structural errors that cannot be fixed as described
    below are suppressed, producing no match.

I.e. only structural errors are suppressed, where a structural error defined as:

    An attempt to access a non-existent member of an object or element of an
    array is defined as a structural error.

In this case what we have is an error, which happens due to an incorrect
function argument valye (the second argument of split_part, either out of
integer bound, or zero for 1 based field counting). It has nothing to do with
the jsonb document, and hence doesn't fall into "structural errors" category.



pgsql-hackers by date:

Previous
From: Daniel Gustafsson
Date:
Subject: Re: [BUG] Race in online checksums launcher_exit()
Next
From: Michael Paquier
Date:
Subject: Re: [PATCH] Add tests for Bitmapset