Optimizer Bug

[Martin v. Loewis]

> Turning on optimization causes a warning. I'm very worried that this
> causes INCORRECT code to be generated by the latest gcc.

Thanks for your bug report. I think you don't worry; g++ is confused
to emit this warning. It generates the following code (on
i586-pc-linux-gnu):

foo__Fv:
.LFB1:
.LEHB15:
	pushl %ebp
.LCFI0:
	movl %esp,%ebp
.LCFI1:
	subl $24,%esp
.LCFI2:
	addl $-12,%esp
	pushl $4
.LCFI3:
	call __builtin_new
	movl $3,(%eax)
	jmp .L58
.LEHE15:
	.p2align 4,,7
.L16:
.LCFI4:
	call __throw
	.p2align 4,,7
.L15:
	jmp .L16
	.p2align 4,,7
.L58:
	movl %ebp,%esp
	popl %ebp
	ret

Here, you can see that, after returning from __builtin_new, it puts a
3 into the new memory, and jumps to L58, at which point it still has
the result of the new in eax, so that gets returned from the function.

I don't fully understand why the warning is emitted; I've put it your
report into GNATS.

Martin