ref: 6408599f6398af4a1c0913c1ee8b1e63c4e4fb4a
parent: e2b1e124e7694cad72c120ae3b7eea8dbb26f4d8
author: Roberto E. Vargas Caballero <[email protected]>
date: Fri May 27 13:08:32 EDT 2016
[cc1] Remove undefined behaviour in specifier Variables local to loops are created and destroy in every iteration of the loop, and it means that they (logically) does not retain the value from the previous iteration. In the case of long long we were using the value of the previous iteration (the iteration of the first long), and it was working because moderm compilers does not create/destroy the variables in this case. It was possible to create strange results with something like: long int long because in this case p was pointing to type and not to the size. This patch fixes the problem setting the value of p to NULL in every iteration and explicitily setting the value of p in the case of long long. If the value of p is not set to the correct value we will have a segmentation fault and e will discover the error as soon as possible.
--- a/cc1/decl.c
+++ b/cc1/decl.c
@@ -371,7 +371,7 @@
spec = qlf = sign = type = cls = size = 0;
for (;;) {
- unsigned *p;
+ unsigned *p = NULL;
Type *(*dcl)(void) = NULL;
switch (yytoken) {
@@ -415,7 +415,6 @@
if (size == LONG) {
yylval.token = LLONG;
size = 0;
- break;
}
case SHORT:
p = &size;