Re: [PATCH v2] c++: ICE initing lifetime-extended constexpr var [PR107079]

[Jason Merrill <jason@redhat.com>]

On 2/8/23 18:15, Marek Polacek wrote:
> On Wed, Feb 08, 2023 at 04:00:25PM -0800, Jason Merrill wrote:
>> On 2/8/23 13:01, Marek Polacek wrote:
>>> (This may not be a complete fix but I got stuck so I'm posting what
>>> I have, which at least fixes the ICE.)
>>>
>>> We ICE on the simple:
>>>
>>>     struct X { const X* x = this; };
>>>     constexpr const X& x = X{};
>>>
>>> where store_init_value initializes 'x' with
>>>
>>>     &TARGET_EXPR <D.2768, {.x=(const struct X *) &<PLACEHOLDER_EXPR struct X>}>
>>>
>>> but we must lifetime-extend via extend_ref_init_temps and we get
>>>
>>>     _ZGR1x_.x = (const struct X *) &<PLACEHOLDER_EXPR struct X> >>>;, (const struct X &) &_ZGR1x_;
>>>
>>> Since 'x' was declared constexpr, we do cxx_constant_init and we hit
>>> the preeval code added in r269003 while evaluating the INIT_EXPR:
>>>
>>>     _ZGR1x_.x = (const struct X *) &<PLACEHOLDER_EXPR struct X> >>>
>>>
>>> but we have no ctx.ctor or ctx.object here so lookup_placeholder won't
>>> find anything that could replace X and we ICE on
>>>    7861       /* A placeholder without a referent.  We can get here when
>>>    7862          checking whether NSDMIs are noexcept, or in massage_init_elt;
>>>    7863          just say it's non-constant for now.  */
>>>    7864       gcc_assert (ctx->quiet);
>>> because cxx_constant_init means !ctx->quiet.  It's not correct that
>>> there isn't a referent.  I think the following patch is a pretty
>>> straightforward fix: pass the _ZGR var down to maybe_constant_init so
>>> that it can replace the PLACEHOLDER_EXPR with _ZGR1x_.
>>>
>>> What I wasn't able to make work is the commented assert in the test.
>>> It doesn't pass because we complain that _ZGR1x_ isn't a constexpr
>>> variable,
>>
>> That sounds like we aren't (correctly) implementing
>>
>>   https://eel.is/c++draft/expr.const#4.7
> 
> Ah yes, this is DR2126 = c++/101588.  I wonder if the fix will include
> checking startswith (IDENTIFIER_POINTER (DECL_NAME (variable)), "_ZGR")
> to see if its lifetime is extended.
>   
>>> but making it so would just result in "used in its own
>>> initializer" (which is true).
>>
>> True, but not in the sense it means; its initializer doesn't depend on its
>> (uninitialized) value.
> 
> ...because we're only interested in its address.
> 
>>> diff --git a/gcc/cp/call.cc b/gcc/cp/call.cc
>>> index f7c5d9da94b..a0afab9b26a 100644
>>> --- a/gcc/cp/call.cc
>>> +++ b/gcc/cp/call.cc
>>> @@ -13583,7 +13583,7 @@ set_up_extended_ref_temp (tree decl, tree expr, vec<tree, va_gc> **cleanups,
>>>      /* If the initializer is constant, put it in DECL_INITIAL so we get
>>>         static initialization and use in constant expressions.  */
>>> -  init = maybe_constant_init (expr);
>>> +  init = maybe_constant_init (expr, var);
>>
>> We should also pass true for manifestly_const_eval as in store_init_value.
> 
> Done.
> 
> Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk?

OK.

> -- >8 --
> We ICE on the simple:
> 
>    struct X { const X* x = this; };
>    constexpr const X& x = X{};
> 
> where store_init_value initializes 'x' with
> 
>    &TARGET_EXPR <D.2768, {.x=(const struct X *) &<PLACEHOLDER_EXPR struct X>}>
> 
> but we must lifetime-extend via extend_ref_init_temps and we get
> 
>    _ZGR1x_.x = (const struct X *) &<PLACEHOLDER_EXPR struct X> >>>;, (const struct X &) &_ZGR1x_;
> 
> Since 'x' was declared constexpr, we do cxx_constant_init and we hit
> the preeval code added in r269003 while evaluating the INIT_EXPR:
> 
>    _ZGR1x_.x = (const struct X *) &<PLACEHOLDER_EXPR struct X> >>>
> 
> but we have no ctx.ctor or ctx.object here so lookup_placeholder won't
> find anything that could replace X and we ICE on
>   7861       /* A placeholder without a referent.  We can get here when
>   7862          checking whether NSDMIs are noexcept, or in massage_init_elt;
>   7863          just say it's non-constant for now.  */
>   7864       gcc_assert (ctx->quiet);
> because cxx_constant_init means !ctx->quiet.  It's not correct that
> there isn't a referent.  I think the following patch is a pretty
> straightforward fix: pass the _ZGR var down to maybe_constant_init so
> that it can replace the PLACEHOLDER_EXPR with _ZGR1x_.
> 
> The commented assert in the test doesn't pass: we complain that _ZGR1x_
> isn't a constexpr variable because we don't implement DR2126 (PR101588).
> 
> 	PR c++/107079
> 
> gcc/cp/ChangeLog:
> 
> 	* call.cc (set_up_extended_ref_temp): Pass var to maybe_constant_init.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/cpp0x/constexpr-nsdmi2.C: New test.
> ---
>   gcc/cp/call.cc                                | 2 +-
>   gcc/testsuite/g++.dg/cpp0x/constexpr-nsdmi2.C | 9 +++++++++
>   2 files changed, 10 insertions(+), 1 deletion(-)
>   create mode 100644 gcc/testsuite/g++.dg/cpp0x/constexpr-nsdmi2.C
> 
> diff --git a/gcc/cp/call.cc b/gcc/cp/call.cc
> index f7c5d9da94b..a349d8e79db 100644
> --- a/gcc/cp/call.cc
> +++ b/gcc/cp/call.cc
> @@ -13583,7 +13583,7 @@ set_up_extended_ref_temp (tree decl, tree expr, vec<tree, va_gc> **cleanups,
>   
>     /* If the initializer is constant, put it in DECL_INITIAL so we get
>        static initialization and use in constant expressions.  */
> -  init = maybe_constant_init (expr);
> +  init = maybe_constant_init (expr, var, /*manifestly_const_eval=*/true);
>     /* As in store_init_value.  */
>     init = cp_fully_fold (init);
>     if (TREE_CONSTANT (init))
> diff --git a/gcc/testsuite/g++.dg/cpp0x/constexpr-nsdmi2.C b/gcc/testsuite/g++.dg/cpp0x/constexpr-nsdmi2.C
> new file mode 100644
> index 00000000000..d711b8051c0
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp0x/constexpr-nsdmi2.C
> @@ -0,0 +1,9 @@
> +// PR c++/107079
> +// { dg-do compile { target c++11 } }
> +
> +struct X {
> +  const X* x = this;
> +};
> +constexpr const X& x = X{};
> +// TODO: The assert should pass once we implement DR2126 (c++/101588).
> +//static_assert(x.x == &x);
> 
> base-commit: f6fc79d0c90ff3a924d272eff74b32656bdf5481