https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83336
Bug ID: 83336
Summary: Issues with displaying inlining chain for middle-end
warnings
Product: gcc
Version: 8.0
Status: UNCONFIRMED
Keywords: diagnostic
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: dmalcolm at gcc dot gnu.org
CC: msebor at gcc dot gnu.org
Target Milestone: ---
msebor reported this issue on the gcc-patches list:
https://gcc.gnu.org/ml/gcc-patches/2017-12/msg00163.html
where the following test case...
VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV
#include <string.h>
#define FOO(d, s) strcpy (d, s)
void foo (char *d, const char *s) { FOO (d, s); }
void sink (char*);
void bar (void)
{
char a[3];
foo (a, "1234567"); // bug here
sink (a);
}
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...leads to confusing output for the middle-end warning -Wstringop-overflow:
$ ./xgcc -B. -c /tmp/test.c -Wall -O2
/tmp/test.c: In function ‘bar’:
/tmp/test.c:3:19: warning: ‘__builtin_memcpy’ writing 8 bytes into a region of
size 3 overflows the destination [-Wstringop-overflow=]
#define FOO(d, s) strcpy (d, s)
^~~~~~~~~~~~~
/tmp/test.c:5:37: note: in expansion of macro ‘FOO’
void foo (char *d, const char *s) { FOO (d, s); }
^~~
It should report something like:
./xgcc -B. -c /tmp/test.c -Wall -O2
/tmp/test.c: In function ‘bar’:
/tmp/test.c:3:19: warning: ‘__builtin_memcpy’ writing 8 bytes into a region of
size 3 overflows the destination [-Wstringop-overflow=]
#define FOO(d, s) strcpy (d, s)
^~~~~~~~~~~~~
/tmp/test.c:5:37: note: in expansion of macro ‘FOO’
inlined from ‘void bar()’ at /tmp/test.c:13:??:
foo (a, "1234567"); // bug here
^~~
void foo (char *d, const char *s) { FOO (d, s); }
^~~
i.e. displaying the chain of function inlining that led to the bogus call.
There are various issues here:
* non-display of inlining information
* interaction of chain-of-inlining with chain-of-macro-expansion
when printing diagnostics
* we don't print the source code for the inlined callsites
(diagnostic_show_locus)
* lack of test coverage for how we print the chain of inlining
* surprising API for middle-end warnings: use of a location_t for
warning_at (used for determining if we're in a system header),
*then* %K/%G and an expr, which is then used for the location
of the diagnostic.
I've been working on a set of patches for the above.
