BUG #19822: DEFAULT boolean expressions in SQL/JSON functions are re-encoded as JSON booleans and always yield f
BUG #19822: DEFAULT boolean expressions in SQL/JSON functions are re-encoded as JSON booleans and always yield f
От:
PG Bug reporting form <noreply@postgresql.org>
Дата:
The following bug has been logged on the website:
Bug reference: 19822
Logged by: chunling qin
Email address: 303677365@qq.com
PostgreSQL version: 18.6
Operating system: Linux (TencentOS 4.4, x86_64)
Description:
Hello,
In transformJsonBehavior() (src/backend/parser/parse_expr.c), the
"coerce at runtime" branch replaces a boolean DEFAULT expression with a
JSON boolean constant using the *clause behavior type* instead of the
expression's actual value:
char *val = btype == JSON_BEHAVIOR_TRUE ? "true" : "false";
For a DEFAULT clause, btype is JSON_BEHAVIOR_DEFAULT, so the constant is
always "false", regardless of whether the expression evaluates to true,
false, or NULL. The replacement is also not guarded by !isnull, so a
NULL boolean constant becomes false as well.
The trigger is: the DEFAULT expression's type is boolean, and the
RETURNING type's base type is not int4 (the ordinary bool->int4 cast
path is unaffected).
Reproducer (self-contained, on 18.6 and on 20devel):
-- G1: DEFAULT true must yield 'true ' (padded to char(5))
SELECT JSON_VALUE(jsonb '"xyzzy!"', '$'
RETURNING char(5) DEFAULT true ON ERROR);
-- actual: false <- the boolean true was replaced by JSON false
-- expected: 'true ' (SQL boolean cast to bpchar(5))
-- G4: same with a domain over bool
CREATE DOMAIN dbool AS bool;
SELECT JSON_VALUE(jsonb '"xyzzy!"', '$'
RETURNING dbool DEFAULT true ON ERROR);
-- actual: f -- expected: t
-- H1: a NULL boolean constant is also replaced (no !isnull guard)
SELECT JSON_VALUE(jsonb '"xyzzy!"', '$'
RETURNING char(5) DEFAULT NULL::bool ON ERROR);
-- actual: false -- expected: NULL
-- Controls that show the normal paths are fine:
SELECT JSON_VALUE(jsonb '"xyzzy!"', '$'
RETURNING char(5) DEFAULT false ON ERROR); -- false (correct by
accident)
SELECT JSON_VALUE(jsonb '"xyzzy!"', '$'
RETURNING int DEFAULT true ON ERROR); -- 1 (bool->int4 path,
correct)
Expected: the value of a DEFAULT expression should be determined by the
expression itself (true -> "true", NULL -> NULL), not by the clause
behavior type; JSON boolean encoding should only apply to an explicit
FORMAT JSON.
Actual: DEFAULT true/false/NULL all come back as JSON false whenever the
runtime-coercion path is taken, silently and with no warning. A common
use of DEFAULT true is a fallback flag (feature switch, approval marker);
getting false there instead of true produces wrong business decisions
with no error anywhere.
Reproduced on:
- PostgreSQL 18.6 (REL_18_STABLE, self-built, snapshot 2026-10-26)
- PostgreSQL 20devel (snapshot of 2026-10-29)
A draft fix location would be transformJsonBehavior()
(src/backend/parser/parse_expr.c): use the actual constant value for
boolean DEFAULT expressions (and guard the replacement with !isnull).
Regards,
Chunling Qin
Re: BUG #19822: DEFAULT boolean expressions in SQL/JSON functions are re-encoded as JSON booleans and always yield f
От:
Amit Langote <amitlangote09@gmail.com>
Дата:
Hi, On Mon, Nov 2, 2026 at 8:04 PM PG Bug reporting form wrote: > > The following bug has been logged on the website: > > Bug reference: 19822 > Logged by: chunling qin > Email address: 303677365@qq.com > PostgreSQL version: 18.6 > Operating system: Linux (TencentOS 4.4, x86_64) > Description: > > Hello, > > In transformJsonBehavior() (src/backend/parser/parse_expr.c), the > "coerce at runtime" branch replaces a boolean DEFAULT expression with a > JSON boolean constant using the *clause behavior type* instead of the > expression's actual value: > > char *val = btype == JSON_BEHAVIOR_TRUE ? "true" : "false"; > > For a DEFAULT clause, btype is JSON_BEHAVIOR_DEFAULT, so the constant is > always "false", regardless of whether the expression evaluates to true, > false, or NULL. The replacement is also not guarded by !isnull, so a > NULL boolean constant becomes false as well. > > The trigger is: the DEFAULT expression's type is boolean, and the > RETURNING type's base type is not int4 (the ordinary bool->int4 cast > path is unaffected). > > Reproducer (self-contained, on 18.6 and on 20devel): > > -- G1: DEFAULT true must yield 'true ' (padded to char(5)) > SELECT JSON_VALUE(jsonb '"xyzzy!"', '$' > RETURNING char(5) DEFAULT true ON ERROR); > -- actual: false <- the boolean true was replaced by JSON false > -- expected: 'true ' (SQL boolean cast to bpchar(5)) > > -- G4: same with a domain over bool > CREATE DOMAIN dbool AS bool; > SELECT JSON_VALUE(jsonb '"xyzzy!"', '$' > RETURNING dbool DEFAULT true ON ERROR); > -- actual: f -- expected: t > > -- H1: a NULL boolean constant is also replaced (no !isnull guard) > SELECT JSON_VALUE(jsonb '"xyzzy!"', '$' > RETURNING char(5) DEFAULT NULL::bool ON ERROR); > -- actual: false -- expected: NULL > > -- Controls that show the normal paths are fine: > SELECT JSON_VALUE(jsonb '"xyzzy!"', '$' > RETURNING char(5) DEFAULT false ON ERROR); -- false (correct by > accident) > SELECT JSON_VALUE(jsonb '"xyzzy!"', '$' > RETURNING int DEFAULT true ON ERROR); -- 1 (bool->int4 path, > correct) > > Expected: the value of a DEFAULT expression should be determined by the > expression itself (true -> "true", NULL -> NULL), not by the clause > behavior type; JSON boolean encoding should only apply to an explicit > FORMAT JSON. > > Actual: DEFAULT true/false/NULL all come back as JSON false whenever the > runtime-coercion path is taken, silently and with no warning. A common > use of DEFAULT true is a fallback flag (feature switch, approval marker); > getting false there instead of true produces wrong business decisions > with no error anywhere. > > Reproduced on: > - PostgreSQL 18.6 (REL_18_STABLE, self-built, snapshot 2026-10-26) > - PostgreSQL 20devel (snapshot of 2026-10-29) > > A draft fix location would be transformJsonBehavior() > (src/backend/parser/parse_expr.c): use the actual constant value for > boolean DEFAULT expressions (and guard the replacement with !isnull). Thanks for the report. This looks to be the same as bug #19625. Please check the patch I just posted to that thread [1]. -- Thanks, Amit Langote [1] https://postgr.es/m/19625-683b498c92087bc8@postgresql.org