https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88063
--- Comment #13 from Tom de Vries <vries at gcc dot gnu.org> --- Comment on attachment 45063 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=45063 combined patch >+ // We only kept the list of units to free them on failure. On >+ // success the units are retained, pointed to by the entries in >+ // addrs. >+ units.alc += units.size; >+ units.size = 0; >+ backtrace_vector_release (state, &units, error_callback, data); If I force mmap.c to alloc.c like so: ... diff --git a/libbacktrace/mmap.c b/libbacktrace/mmap.c index 32fcba62399..e7fea31b095 100644 --- a/libbacktrace/mmap.c +++ b/libbacktrace/mmap.c @@ -1,3 +1,5 @@ +#include "alloc.c" +#if 0 /* mmap.c -- Memory allocation with mmap. Copyright (C) 2012-2018 Free Software Foundation, Inc. Written by Ian Lance Taylor, Google. @@ -323,3 +325,4 @@ backtrace_vector_release (struct backtrace_state *state, vec->alc = 0; return 1; } +#endif ... we get: ... $ ./btest realloc: No such file or directory realloc: No such file or directory realloc: No such file or directory FAIL: backtrace_full noinline ... because realloc happens to return NULL when called with size == 0 in backtrace_vector_release, and then backtrace_vector_release calls the error callback: ... vec->base = realloc (vec->base, vec->size); if (vec->base == NULL) { error_callback (data, "realloc", errno); return 0; } ... Fixed by patch submitted here: https://gcc.gnu.org/ml/gcc-patches/2018-11/msg01914.html
