On Mon, Jun 12, 2017 at 10:10 AM Roman Lebedev via Phabricator <
revi...@reviews.llvm.org> wrote:
> lebedev.ri added a comment.
>
> So i'm trying to analyze that stage2 warning.
>
Could you link to the buildbot failure to see the original LLVM project
code triggering this situation?
> The testcase //seems// to be: (autogenerated all the variants)
>
> void test_nop() {
> unsigned char **ptr1 = 0;
> void **ptr2 = (void **)ptr1;
> }
> void test_bad() {
> unsigned char **ptr1 = 0;
> const void **ptr2 = (const void **)ptr1; // expected-warning {{cast
> from 'unsigned char **' to 'const void **' must have all intermediate
> pointers const qualified to be safe}}
> }
> void test_good0() {
> unsigned char **ptr1 = 0;
> void *const *ptr2 = (void *const *)ptr1;
> }
> void test_good1() {
> unsigned char **ptr1 = 0;
> const void *const *ptr2 = (const void *const *)ptr1;
> }
> void test_good2() {
> unsigned char *const *ptr1 = 0;
> void *const *ptr2 = (void *const *)ptr1;
> }
> void test_good3() {
> unsigned char *const *ptr1 = 0;
> const void *const *ptr2 = (const void *const *)ptr1;
> }
> void test_good4() {
> const unsigned char **ptr1 = 0;
> const void **ptr2 = (const void **)ptr1;
> }
> void test_good5() {
> const unsigned char **ptr1 = 0;
> const void *const *ptr2 = (const void *const *)ptr1;
> }
> void test_good6() {
> const unsigned char *const *ptr1 = 0;
> const void *const *ptr2 = (const void *const *)ptr1;
> }
>
> GCC does not warn about such code at all, clang in C mode does warn about
> only one combination:
>
> $ gcc -c /tmp/test.c -Wcast-qual
> $ echo $?
> 0
> $ clang -c /tmp/test.c -Wcast-qual
> /tmp/test.c:7:38: warning: cast from 'unsigned char **' to 'const void
> **' must have all intermediate pointers const qualified to be safe
> [-Wcast-qual]
> const void **ptr2 = (const void **)ptr1; // expected-warning {{cast
> from 'unsigned char **' to 'const void **' must have all intermediate
> pointers const qualified to be safe}}
> ^
> 1 warning generated.
>
> David, you reviewed the original `-Wcast-qual` patch, does all that ^ make
> sense?
>
That seems like a reasonable warning, do you agree?
But maybe it's best to put it under another flag name so it doesn't collide
with GCC.
But really - *shrug* I'd probably leave it under the same flag, fix the
LLVM project code that causes it, and carry on.
>
> F3429854: gen.cpp <https://reviews.llvm.org/F3429854>
>
>
> Repository:
> rL LLVM
>
> https://reviews.llvm.org/D33102
>
>
>
>
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
