[PATCH] D110216: [clang] retain type sugar in auto / template argument deduction

[Matheus Izvekov via Phabricator via cfe-commits]

mizvekov added a comment.

In D110216#3135461 <https://reviews.llvm.org/D110216#3135461>, @hans wrote:

> We're seeing a diagnostic change in Chromium which looks funny. For the following code (span<> is our own class):
> ....
> the diagnostic changes from:
>
>   fatal error: no viable conversion from 'span<const int, [...]>' to 'span<int, [...]>
>
> to
>
>   fatal error: no viable conversion from 'span<T, [...]>' to 'span<int, [...]>
>
> This looks very strange.

Hi! thanks for the report.

I am not sure how to reproduce this.

I tried mocking this situation:

  template <class T> struct vector {};
  template <class T, class=void> struct span {};
  
  template <class T> auto make_span(const vector<T> &) { return span<const T>(); }
  
  void WontCompile() {
    const vector<int> v;
  
    span<int> s1 = make_span(v);
  }

Which I get:

  error: no viable conversion from 'span<const int>' to 'span<int>'

Looking at the `dump` of that type:

  AutoType 0x246b83676c0 'span<const int>' sugar
  `-ElaboratedType 0x246b8366d60 'span<const int>' sugar
    `-TemplateSpecializationType 0x246b8366d20 'span<const int>' sugar span
      |-TemplateArgument type 'const int':'const int'
      | `-QualType 0x246b8366751 'const int' const
      |   `-SubstTemplateTypeParmType 0x246b8366750 'int' sugar
      |     |-TemplateTypeParmType 0x246b835c790 'T' dependent depth 0 index 0
      |     | `-TemplateTypeParm 0x246b835c740 'T'
      |     `-BuiltinType 0x246b82e81e0 'int'
      `-RecordType 0x246b8366d00 'struct span<const int>'
        `-ClassTemplateSpecialization 0x246b8366be8 'span'

In particular, the `SubstTemplateTypeParmType` desugars to int, not `T`, and I am not sure how you managed to get that ellipsis (from the type diff) on the template specialization arguments.

I am looking though at a clang built from `D112374`, which is main with what I believe are other patches that have no effect here besides the addition of the `ElaboratedType` node on that dump.

To be sure, it's going to take me a while to get a main built here, so hopefully this is not a big problem for you right now?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D110216/new/

https://reviews.llvm.org/D110216