On Wed, Mar 27, 2019 at 05:37:40PM +0300, Alexander Korotkov wrote:
On Wed, Mar 27, 2019 at 4:49 PM Tom Lane <t...@sss.pgh.pa.us> wrote:
Alexander Korotkov <a.korot...@postgrespro.ru> writes:
> Still no reproduction.
Annoying, but it's probably not worth expending more effort on
right now. I wonder whether that buildfarm animal can be upgraded
to capture core dump stack traces --- if so, then if it happens
again we'd have more info.
Hopefully, Andrew will manage to get a backtrace [1].
BTW, while searching for this bug, I've collected this backtrace using valgrind.
==00:00:00:14.596 10866== Conditional jump or move depends on
uninitialised value(s)
==00:00:00:14.596 10866== at 0x579F8C4: ____strtod_l_internal (in
/usr/lib64/libc-2.17.so)
==00:00:00:14.596 10866== by 0x771561: float8in_internal_opt_error
(float.c:394)
==00:00:00:14.596 10866== by 0x7718B9: float8in_internal (float.c:515)
==00:00:00:14.596 10866== by 0x7718B9: float8in (float.c:336)
==00:00:00:14.596 10866== by 0x842D43: DirectFunctionCall1Coll (fmgr.c:803)
==00:00:00:14.596 10866== by 0x7C9649: numeric_float8 (numeric.c:3417)
==00:00:00:14.596 10866== by 0x842D43: DirectFunctionCall1Coll (fmgr.c:803)
==00:00:00:14.596 10866== by 0x7A1D8D: jsonb_float8 (jsonb.c:2058)
==00:00:00:14.596 10866== by 0x5F8F54: ExecInterpExpr (execExprInterp.c:649)
==00:00:00:14.596 10866== by 0x6A2E19: ExecEvalExprSwitchContext
(executor.h:307)
==00:00:00:14.596 10866== by 0x6A2E19: evaluate_expr (clauses.c:4827)
==00:00:00:14.596 10866== by 0x6A45FF: evaluate_function (clauses.c:4369)
==00:00:00:14.596 10866== by 0x6A45FF: simplify_function (clauses.c:3999)
==00:00:00:14.596 10866== by 0x6A31C1:
eval_const_expressions_mutator (clauses.c:2474)
==00:00:00:14.596 10866== by 0x644466: expression_tree_mutator
(nodeFuncs.c:3072)
==00:00:00:14.596 10866==
{
<insert_a_suppression_name_here>
Memcheck:Cond
fun:____strtod_l_internal
fun:float8in_internal_opt_error
fun:float8in_internal
fun:float8in
fun:DirectFunctionCall1Coll
fun:numeric_float8
fun:DirectFunctionCall1Coll
fun:jsonb_float8
fun:ExecInterpExpr
fun:ExecEvalExprSwitchContext
fun:evaluate_expr
fun:evaluate_function
fun:simplify_function
fun:eval_const_expressions_mutator
fun:expression_tree_mutator
}
Not sure whether it's related to 16d489b0fe. Will investigate more.
This might be another case of false positive due to SSE (which I think is
used by strtod in some cases). But I'd expect strncasecmp in the stack in
that case, so maybe that's not it.
cheers
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
