Issue32374 added a new subtest (test_bad_traverse) in test_importlib. The test succeeds when a script calling a bad_traverse exits with a non-zero status (e.g., crashes because of segmentation violation) This patch "fixes" Modules/_testmultiphase.c so that it exits with a non-zero status when a NULL pointer is accessed. This is what was causing other systems to 'crash' expectedly - while AIX was not. A simple demo program to demonstrate the difference behavior of AIX: /* * Program to show whether an executable crashes, or not * when a NULL pointer is accessed * Author: AIXTOOLS */ #include<stdio.h> main() { long *vpt = NULL; fprintf(stdout, "vpt = %08lx\n", (long) vpt); if (*vpt) fprintf(stdout,"True\n"); else fprintf(stdout,"False\n"); } On AIX: root@x066:[/data/prj/python/git]make nullpr; ./nullpr cc nullpr.c -o nullpr vpt = 00000000 False On a Linux server: root@x074:/data/prj/python/git# rm nullpr; make nullpr; ./nullpr make: Warning: File 'nullpr.c' has modification time 352 s in the future cc nullpr.c -o nullpr nullpr.c:8:1: warning: return type defaults to 'int' [-Wimplicit-int] main() ^ make: warning: Clock skew detected. Your build may be incomplete. vpt = 00000000 Segmentation fault
On 10/03/2019 19:31, Michael Felt wrote: > Michael Felt <aixtools@felt.demon.nl> added the comment: > > could this be backported to versions 3.7, and if applicable, to version 3.6 Only 3.7 - As 3.6 is in security mode. > > ---------- > > _______________________________________ > Python tracker <report@bugs.python.org> > <https://bugs.python.org/issue34720> > _______________________________________ >
