New warning (or at least I only noticed it now)

[Jean-Marc Lasgouttes]

Le 22/11/2022 à 11:25, Pavel Sanda a écrit :
> On Tue, Nov 22, 2022 at 07:49:02AM +0100, Jean-Marc Lasgouttes wrote:
>> Le 22 novembre 2022 03:04:39 GMT+01:00, Pavel Sanda <sanda at lyx.org> a écrit :
>>> Nice, so now we know the culprit. The question is whether we want
>>> to get rid of reference on the left side or fix the temps on the right.
>>
>> So what is the issue that the warning detect? I still don't get it.
> 
> My interpretation is that gcc is nervous about the temporary created via
> getFont which is then propagated to the bottom via chain of refs and it's
> not possible for it to decide whether returned value is independent of
> that temp.
> 
> Note that I'm not claiming that there is some *real* issue (I did not find any),
> it's just that we need to shut up that warning.
> 
> Another option would be to create MWE and send it to gcc bugzilla,
> but I do not have gcc 13 to deliver it.

Looking here, this feature is one month old:
https://patchwork.sourceware.org/project/gcc/patch/20221021232824.1093138-1-polacek@redhat.com/

Maybe is it indeed a false positive, but we do not want to have to cater 
for that. I would be nervous to have to separate calls to theFontMetrics 
in two for no good reason.

What I understand, is that it triggers when we store as reference the 
result of a function that uses a temporary as parameter, with the idea 
that the reference _might_ depend on the temporary. This feels wrong.

It seems as simple as:
1. theFontMetrics returns a lvalue
2. it uses a parameter that is a temporary
1+2 => the lvalue should not be stored because it may disappear when the 
temporary is destroyed.

So it might be that it is buggy and will sort itself out magically.

JMarc