On 4/4/20 1:56 PM, Marek Polacek wrote:
> On Fri, Apr 03, 2020 at 10:39:49PM -0400, Jason Merrill via Gcc-patches wrote:
>> On 4/3/20 9:08 PM, Marek Polacek wrote:
>>> On Fri, Apr 03, 2020 at 03:01:37PM -0400, Jason Merrill via Gcc-patches wrote:
>>>> On 3/30/20 4:28 PM, Marek Polacek wrote:
>>>>> Here we crash in the gimplifier because gimplify_init_ctor_eval doesn't
>>>>> expect null indexes for a constructor:
>>>>>
>>>>>          /* ??? Here's to hoping the front end fills in all of the indices,
>>>>>             so we don't have to figure out what's missing ourselves.  */
>>>>>          gcc_assert (purpose);
>>>>>
>>>>> The indexes weren't filled because we never called reshape_init: for
>>>>> a constructor that represents parenthesized initialization of an
>>>>> aggregate we don't allow brace elision or designated initializers.  So
>>>>> fill in the indexes manually, here we have an array, and we can simply
>>>>> assign indexes starting from 0.
>>>>>
>>>>> Bootstrapped/regtested on x86_64-linux, ok for trunk?
>>>>
>>>> Shouldn't digest_init fill in the indexes?  In
>>>> process_init_constructor_array I see
>>>>
>>>>         if (!ce->index)
>>>>           ce->index = size_int (i);
>>>
>>> Yes, that works too.  Thus:
>>>
>>> Bootstrapped/regtested on x86_64-linux, ok for trunk?
>>>
>>> -- >8 --
>>> Here we crash in the gimplifier because gimplify_init_ctor_eval doesn't
>>> expect null indexes for a constructor:
>>>
>>>         /* ??? Here's to hoping the front end fills in all of the indices,
>>>            so we don't have to figure out what's missing ourselves.  */
>>>         gcc_assert (purpose);
>>>
>>> The indexes weren't filled because we never called reshape_init: for
>>> a constructor that represents parenthesized initialization of an
>>> aggregate we don't allow brace elision or designated initializers.  So
>>> call digest_init to fill in the indexes.
>>>
>>> Bootstrapped/regtested on x86_64-linux, ok for trunk?
>>>
>>> 	PR c++/94155 - crash in gimplifier with paren init of aggregates.
>>> 	* decl.c (check_initializer): Call digest_init.
>>>
>>> 	* g++.dg/cpp2a/paren-init22.C: New test.
>>> ---
>>>    gcc/cp/decl.c                             |  5 +++++
>>>    gcc/testsuite/g++.dg/cpp2a/paren-init22.C | 15 +++++++++++++++
>>>    2 files changed, 20 insertions(+)
>>>    create mode 100644 gcc/testsuite/g++.dg/cpp2a/paren-init22.C
>>>
>>> diff --git a/gcc/cp/decl.c b/gcc/cp/decl.c
>>> index 69a238997b4..63e7bda09f5 100644
>>> --- a/gcc/cp/decl.c
>>> +++ b/gcc/cp/decl.c
>>> @@ -6754,6 +6754,11 @@ check_initializer (tree decl, tree init, int flags, vec<tree, va_gc> **cleanups)
>>>    	      init = build_constructor_from_list (init_list_type_node, init);
>>>    	      CONSTRUCTOR_IS_DIRECT_INIT (init) = true;
>>>    	      CONSTRUCTOR_IS_PAREN_INIT (init) = true;
>>> +	      /* The gimplifier expects that the front end fills in all of the
>>> +		 indices.  Normally, reshape_init_array fills these in, but we
>>> +		 don't call reshape_init because that does nothing when it gets
>>> +		 CONSTRUCTOR_IS_PAREN_INIT.  */
>>> +	      init = digest_init (type, init, tf_warning_or_error);
>>
>> But why weren't we already calling digest_init in store_init_value?  Was the
>> CONSTRUCTOR making it all the way to gimplification still having
>> init_list_type_node?
> 
> It's because we set LOOKUP_ALREADY_DIGESTED a few lines below:
>   6813               /* Don't call digest_init; it's unnecessary and will complain
>   6814                  about aggregate initialization of non-aggregate classes.  */
>   6815               flags |= LOOKUP_ALREADY_DIGESTED;
> and so store_init_value doesn't digest.  Given the comment I'd be nervous about
> not setting that flag here.

OK, then why isn't it called by build_aggr_init?  How is the CONSTRUCTOR 
getting a type without this being fixed up?

...

Ah, because build_vec_init builds up a new CONSTRUCTOR and gives it a 
type without setting the indexes like process_init_constructor_array does:

Jason