Re: [PATCH v7] c++: Add diagnostic when operator= is used as truth cond [PR25689]

[Jason Merrill <jason@redhat.com>]

On 3/13/22 19:43, Zhao Wei Liew wrote:
> On Sat, 12 Mar 2022 at 06:15, Jason Merrill <jason@redhat.com> wrote:
>> It looks good, but unfortunately regresses some other warning tests,
>> such as Wnonnull5.C.  Please remember to run the regression tests before
>> sending a patch (https://gcc.gnu.org/contribute.html#testing).
>>
>> This seems to be a complicated problem with suppress_warning, which
>> means your call to suppress_warning effectively silences all later
>> warnings, not just -Wparentheses.
>>
>> You should be able to work around this issue by only calling
>> suppress_warning in the specific case we're interested in, i.e. when
>> warn_parentheses is enabled and "call" is a MODIFY_EXPR.
> 
> My apologies. I've fixed the issue as you suggested and run the regression tests
> to ensure no test regressions. The new patch (v9) is attached.

Looks good.  One thing:

> +	/* Ideally, we'd warn for empty classes using trivial operator= (below),
> +	   but we don't do so yet as it is a non-trivial COMPOUND_EXPR. */
> +	// if (a1 = a2); 

It would be better to uncomment this line and give it an xfailed 
warning, i.e.

/* { dg-warning "suggest parentheses" "" { xfail *-*-* } } */

I've made this change and added the patch to my queue to commit when GCC 
13 stage 1 opens, probably next month.

Thanks!

Jason