gcc 4.8.2 accepts the following case silently without error or warning:
void f();
void g() __asm__(“real_g”);
void f() { g(); } // gcc emits a call to real_g() here
void real_g() __asm__(“gold”);
void real_g() { } // gcc generates a body for gold() here
gcc 4.8.2 generates a warning for the following test case:
void g() __asm__(“func1”);
void g() __asm__(“func2”); // gcc generates a warning and ignores this asm label
void g() { }
Comparing to gcc behavior, I think it is better to emit error in both cases.
Looks good to me. Thanks!
- Gao
From: Nick Lewycky [mailto:[email protected]]
Sent: Friday, December 11, 2015 12:44 AM
To: Gao, Yunzhong
Cc: Clang Commits
Subject: Re: PATCH: error on code that redeclares with an __asm__ label after
the first ODR use
On 10 December 2015 at 17:42, Gao, Yunzhong
<[email protected]<mailto:[email protected]>>
wrote:
Out of curiosity, is the following test case possible too?
void f();
void g() __asm__(“real_g”); // rename g into real_g.
void f() {
g(); // this would actually be calling real_g()
}
void real_g() __asm__("gold"); // re-declaring real_g() into gold <-- should
this be an error too?
I can't see any reason why not. Both clang and gcc will emit "real_g" here
instead of "gold", but changing the asm label in a program is highly dubious. I
added an error for this too.
Thanks!
From: cfe-commits
[mailto:[email protected]<mailto:[email protected]>]
On Behalf Of Nick Lewycky via cfe-commits
Sent: Thursday, December 10, 2015 3:15 PM
To: Clang Commits
Subject: PATCH: error on code that redeclares with an __asm__ label after the
first ODR use
PR22830 shows a case where some code was adding an __asm__ label after the
first use of a function. GCC and Clang diverged on this code, GCC emitting the
name in the last __asm__ label for all uses while clang would switch in the
middle of the TU as the redeclaration was parsed. The attached patch detects
this case and issues an error on such a redeclaration. If this breaks real
code, we can turn it down to a warning.
Please review!
Nick
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
