On Thu, May 15, 2025 at 4:07 PM Jonathan Wakely <jwak...@redhat.com> wrote:
> On Thu, 15 May 2025 at 13:56, Tomasz Kaminski <tkami...@redhat.com> wrote:
> >
> >
> >
> > On Thu, May 15, 2025 at 2:53 PM Jonathan Wakely <jwak...@redhat.com>
> wrote:
> >>
> >> On Thu, 15 May 2025 at 13:43, Tomasz Kaminski <tkami...@redhat.com>
> wrote:
> >> >
> >> >
> >> >
> >> > On Thu, May 15, 2025 at 2:41 PM Jonathan Wakely <jwak...@redhat.com>
> wrote:
> >> >>
> >> >> On 15/05/25 14:35 +0200, Tomasz Kaminski wrote:
> >> >> >Please also add the message to dg-error check in format_kind_neg.cc.
> >> >> >With that LGTM.
> >> >>
> >> >> Yes, already done locally. Here's what I'm testing now.
> >> >
> >> > Any reason to not put the whole message "cannot use primary template
> of 'std::format_kind'"?
> >>
> >> It doesn't fit in 80 columns, and doesn't seem to make the test any
> >> more or less likely to PASS.
> >>
> >> e.g. if we change the static_assert message to not use single quotes,
> >> or to say 'std::format_kind<R>' or some other change, would we want
> >> the test to fail and require changes? Would that improve the value of
> >> the test? IMHO it doesn't, because if it matches some substring of the
> >> static_assert then the test is working correctly, but I'll change it
> >> if you feel strongly.
> >>
> > Not strongly, but "cannot use primary template" sounds like something
> that may be part of
> > unrelated compiler error.
>
> OK, I'll use this and push it:
>
> +// { dg-error "cannot use primary template of 'std::format_kind'" ""
> { target *-*-* } 0 }
>
Thank you.
>
> > And we do not check that this is produced from static_assert either.
>
> I'm not worried about that. I don't think we need to be too paranoid
> here, we control the code and diagnostics from GCC aren't going to
> just randomly change without us realising. And it doesn't really
> matter whether a "cannot use primary template" diagnostic for
> std::format_kind comes from a static assert, or a function using
> =delete("reason"), or a custom compiler diagnostic, or something else.
> All that matters is that it's ill-formed and the message is fairly
> user-friendly.
>
To clarify, I do not mean that we need to check if it is produced by static
assert,
just having a check "cannot use primary template" + it is from static
assert,
would also prevent my concern of accidentally matching compiler error
messages.
But checking for longer substring, including format_kind, is IMHO a better
way to do that.
