Bug 163188 - setting const variables inside a vararg does not change them at -O1 and -O2
Summary: setting const variables inside a vararg does not change them at -O1 and -O2
Keywords:
Status: CLOSED NOTABUG
Alias: None
Product: Fedora
Classification: Fedora
Component: gcc
Version: 4
Hardware: All
OS: Linux
medium
medium
Target Milestone: ---
Assignee: Jakub Jelinek
QA Contact:
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2005-07-13 21:11 UTC by Dan Williams
Modified: 2007-11-30 22:11 UTC (History)
1 user (show)

Fixed In Version:
Clone Of:
Environment:
Last Closed: 2005-07-14 07:52:22 UTC
Type: ---
Embargoed:


Attachments (Terms of Use)

Description Dan Williams 2005-07-13 21:11:27 UTC
See the testcase.  bar starts as -1, but should be 42 after calling foo(), even
though its const.  You could argue that bar shouldn't be const since we are
modifying it, but even so, the behavior should not be different at -O0 than -O1/2.

# gcc quux.c -O0 -o quux
# ./quux
Success: bar is 42

# gcc quux.c -O2 -o quux
# ./quux
Error: bar is not 42!  (it is -1)

Testcase:

#include <stdarg.h>
#include <stdio.h>

void foo (int n_args, ...)
{
    va_list qux;
    int *bar;

    va_start (qux, n_args);
    bar = va_arg (qux, int *);
    *bar = 42;
    va_end(qux);
}

int main (int argc, char **argv)
{
    const int bar = -1;

    foo (1, &bar);
    if (bar != 42)
        printf ("Error: bar is not 42!  (it is %d)\n", bar);
    else
        printf ("Success: bar is 42\n");
    return 0;
}

Comment 1 Dan Williams 2005-07-13 21:12:37 UTC
Testcase fails at -O1 and -O2 on both:

gcc-4.0.0-13
gcc-4.0.1-2


Comment 2 Jakub Jelinek 2005-07-14 07:52:22 UTC
The testcase is simply invalid, you must not modify a const variable.
The compiler is free to assume it is not modified (that is the purpose of
the const modifier, right?), but does not have to.  You simply trigger an
undefined behaviour and anything can happen at that point.


Note You need to log in before you can comment on or make changes to this bug.