Two fixes found while debugging with ast_backtrace():

1) If MALLOC_DEBUG is used when concurrently using ast_backtrace, the free()
used in that routine will trigger an error, because the memory was allocated
internally to libc, where we could not intercept that call to wrap it.
Therefore, it's not memory we knew about, and the free is reported as an
error.

2) Now that channels are objects, the old hack of initializing a channel
to all zeroes no longer works, since we may try to call something like
ast_channel_lock() within a function on that reference.  In that case, it's
reported as an error, because the pointer isn't an object reference.


git-svn-id: http://svn.digium.com/svn/asterisk/trunk@194101 f38db490-d61c-443f-a65b-d21fe96a405b

1 files changed, 3 insertions, 0 deletions

diff --git a/main/logger.c b/main/logger.c
index bceaf092f..91ae572f9 100644
--- a/main/logger.c
+++ b/main/logger.c
@@ -1159,6 +1159,9 @@ void ast_backtrace(void)
 		for (i = 0; i < bt->num_frames; i++) {
 			ast_log(LOG_DEBUG, "#%d: [%p] %s\n", i, bt->addresses[i], strings[i]);
 		}
+
+		/* MALLOC_DEBUG will erroneously report an error here, unless we undef the macro. */
+#undef free
 		free(strings);
 	} else {
 		ast_debug(1, "Could not allocate memory for backtrace\n");