Opened 4 years ago

Closed 4 years ago

#4654 closed enhancement (fixed)

Analyze regress tests with valgrind

Reported by: Algunenano Owned by: Algunenano
Priority: high Milestone: PostGIS 3.1.0
Component: postgis Version: master
Keywords: Cc:

Description (last modified by Algunenano)

After https://trac.osgeo.org/postgis/ticket/4652 it seems obvious to me there is a need to run the regression tests, either once or ideally as part of the CI, under valgrind to detect issues.

I've done a test and got multiple worrisome errors

There are some around computing stats:

mar 27 20:25:24 Mordor postgres[396099]: ==396099== Conditional jump or move depends on uninitialised value(s)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    at 0x813D16B: gbox_is_valid (gbox.c:221)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x80C864B: compute_gserialized_stats_mode (gserialized_estimate.c:1469)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x3FCD5A: do_analyze_rel (analyze.c:529)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x3FBC9C: analyze_rel (analyze.c:260)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x4BF51E: vacuum (vacuum.c:413)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x4BEDA6: ExecVacuum (vacuum.c:199)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x745296: standard_ProcessUtility (utility.c:670)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x744933: ProcessUtility (utility.c:360)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x74426A: PortalRunUtility (pquery.c:1175)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x7434A1: PortalRunMulti (pquery.c:1321)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x742ADD: PortalRun (pquery.c:796)
mar 27 20:25:24 Mordor postgres[396099]: ==396099==    by 0x73DFD2: exec_simple_query (postgres.c:1215)
mar 27 20:21:18 Mordor postgres[395155]: ==395155== Conditional jump or move depends on uninitialised value(s)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    at 0x81184B1: next_float_down (lwgeom_api.c:63)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x813E720: gbox_float_round (gbox.c:784)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x8141584: gserialized2_get_gbox_p (gserialized2.c:601)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x80C861B: compute_gserialized_stats_mode (gserialized_estimate.c:1457)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x80C6351: compute_gserialized_stats (gserialized_estimate.c:1871)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x3FCD5A: do_analyze_rel (analyze.c:529)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x3FBC9C: analyze_rel (analyze.c:260)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x4BF51E: vacuum (vacuum.c:413)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x4BEDA6: ExecVacuum (vacuum.c:199)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x745296: standard_ProcessUtility (utility.c:670)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x744933: ProcessUtility (utility.c:360)
mar 27 20:21:18 Mordor postgres[395155]: ==395155==    by 0x74426A: PortalRunUtility (pquery.c:1175)

This one seem like a cast:

mar 27 20:23:26 Mordor postgres[395843]: ==395843== Conditional jump or move depends on uninitialised value(s)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    at 0x813D39B: gbox_merge (gbox.c:276)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x81470E9: lwcollection_calculate_gbox_geodetic (lwgeodetic.c:3020)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x81470E9: lwgeom_calculate_gbox_geodetic (lwgeodetic.c:3056)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x81418E0: gserialized2_from_lwgeom (gserialized2.c:1179)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x8109068: geography_serialize (lwgeom_pg.c:305)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x80CAA3E: gserialized_geography_from_lwgeom (geography_inout.c:115)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x80CAB91: geography_in (geography_inout.c:177)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x9240F5: InputFunctionCall (fmgr.c:1548)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x9244C7: OidInputFunctionCall (fmgr.c:1651)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x3E867A: stringTypeDatum (parse_type.c:659)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x3C2459: coerce_type (parse_coerce.c:304)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x3C1E25: coerce_to_target_type (parse_coerce.c:103)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x3E4770: transformAssignedExpr (parse_target.c:577)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==
mar 27 20:23:26 Mordor postgres[395843]: ==395843== Conditional jump or move depends on uninitialised value(s)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    at 0x813D3AC: gbox_merge (gbox.c:277)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x81470E9: lwcollection_calculate_gbox_geodetic (lwgeodetic.c:3020)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x81470E9: lwgeom_calculate_gbox_geodetic (lwgeodetic.c:3056)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x81418E0: gserialized2_from_lwgeom (gserialized2.c:1179)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x8109068: geography_serialize (lwgeom_pg.c:305)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x80CAA3E: gserialized_geography_from_lwgeom (geography_inout.c:115)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x80CAB91: geography_in (geography_inout.c:177)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x9240F5: InputFunctionCall (fmgr.c:1548)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x9244C7: OidInputFunctionCall (fmgr.c:1651)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x3E867A: stringTypeDatum (parse_type.c:659)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x3C2459: coerce_type (parse_coerce.c:304)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x3C1E25: coerce_to_target_type (parse_coerce.c:103)
mar 27 20:23:26 Mordor postgres[395843]: ==395843==    by 0x3E4770: transformAssignedExpr (parse_target.c:577)

Right now I get quite a bit of warnings from Postgresql syscalls (mostly the same one repeated over and over), so I first need to filter those out, and then find errors and match them with the queries that triggered them.

The tests become really slow under valgrind, so I'm not sure how to best approach them (probably test each file independently).

Change History (4)

comment:1 by Algunenano, 4 years ago

Owner: changed from pramsey to Algunenano

in reply to:  description comment:2 by Algunenano, 4 years ago

Description: modified (diff)

comment:3 by Algunenano, 4 years ago

I've spent some time running the regress tests and all I've found are a bunch of initialized variables being used (but later discarded). I'll patch them anyway to simplify looking for bugs in the future.

comment:4 by Raúl Marín <git@…>, 4 years ago

Resolution: fixed
Status: newclosed

In 2a7ebd01/git:

Address valgrind warnings in regress tests

Closes #4654

Note: See TracTickets for help on using tickets.