https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91006
--- Comment #3 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Warray-bounds-4.C is about missed warning on array bounds. It also goes back
with -fno-ipa-sra
b.C: In function ‘int main()’:
b.C:25:27: warning: array subscript 0 is above array bounds of ‘char [0]’
[-Warray-bounds]
25 | FixedString() { contents[0] = '\0'; } // { dg-warning "above array
bounds" }
| ~~~~~~~~^
With -fno-ipa-sra we get in vrp:
main ()
{
const struct FixedString empty;
<bb 2> [local count: 1073741824]:
MEM[(struct FixedString *)&empty] ={v} {CLOBBER};
MEM[(struct FixedString *)&empty].D.2435._vptr.String = &MEM <int (*) ()[8]>
[(void *)&_ZTV11FixedStringILm0EE + 16B];
MEM[(struct FixedString *)&empty].contents[0] = 0;
print_length (&empty.D.2435);
empty ={v} {CLOBBER};
empty ={v} {CLOBBER};
return 0;
}
while with ipa-sra decides to to SRA
String::~String (struct String * const this)
{
<bb 2> :
*this_3(D) ={v} {CLOBBER};
return;
}
again it does not choose to SRA all those functions.
this makes us to get the following main at the vrp time:
main ()
{
const struct FixedString empty;
<bb 2> [local count: 1073741824]:
MEM[(int (*) () * *)&empty] = &MEM <int (*) ()[8]> [(void
*)&_ZTV11FixedStringILm0EE + 16B];
MEM[(char *)&empty + 8B] = 0;
print_length (&empty.D.2435);
empty ={v} {CLOBBER};
empty ={v} {CLOBBER};
return 0;
}
So this time it simply looks like we removed store to contents as dead
bypassing the warning. This store is taken away by IPA-SRA too:
We ISRA:
FixedString<0>::_ZN11FixedStringILm0EEC1Ev.isra.0 (int (*) () * * ISRA.10, char
* ISRA.11)
{
struct FixedString * const this;
struct FixedString * const this;
<bb 3> :
<bb 2> :
*ISRA.10_6(D) = 0B;
*ISRA.10_6(D) = &MEM <int (*) ()[8]> [(void *)&_ZTV11FixedStringILm0EE +
16B];
*ISRA.11_7(D) = 0;
return;
}
where *ISRA.11_7(D) = 0 used to be the out of bounds array store.
Now it happens by
FixedString<0>::_ZN11FixedStringILm0EEC1Ev.isra.0 (&empty, &MEM[(char *)&empty
+ 8B]);
