RecoveryExpr incorrectly generated with MS Extension, causes "cannot compile this l-value expression yet"

[erichkeane]

Bugzilla Link	48690
Version	11.0
OS	Windows NT
CC	@DougGregor,@dyung,@zmodem,@hokein,@zygoloid,@rnk

Extended Description

See: https://godbolt.org/z/4E56c3

It seems that despite allowing this as an extension, we're still generating a RecoverExpr.

[rnk]

See also https://llvm.org/pr20249
The phase ordering between MSVC compatibility recovery and typo correction has been wrong for a while. MSVC checks should go first, but they don't right now.

[rnk]

*** Bug llvm/llvm-bugzilla-archive#48834 has been marked as a duplicate of this bug. ***

[rnk]

Hans bisected to efa9aaa in the other issue:

---

[clang] Suppress "follow-up" diagnostics on recovery call expressions.

Because of typo-correction, the AST can be transformed, and the transformed
AST is marginally useful for diagnostics purpose, the following
diagnostics usually do harm than good (easily cause confusions).

Given the following code:

```
void abcc();
void test() {
  if (abc());
  // diagnostic 1 (for the typo-correction): the typo is correct to `abcc()`, so the code is treate as `if (abcc())` in AST perspective;
  // diagnostic 2 (for mismatch type): we perform an type-analysis on `if`, discover the type is not match
}
```

The secondary diagnostic "convertable to bool" is likely bogus to users.

The idea is to use RecoveryExpr (clang's dependent mechanism) to preserve the
recovery behavior but suppress all follow-up diagnostics.

Differential Revision: https://reviews.llvm.org/D89946

---

[hokein]

Thanks for the report.

I didn't come up with a way to fix that, will look at it further. Since we're close to the release cut (next week), I have reverted it in d972d4c.

[dyung]

Hi, the test that you added in the revert is failing on targets that don't include X86. Please fix the test, or revert it.

See here for an example failure: http://lab.llvm.org:8011/#/builders/43/builds/2549

[hokein]

Hi, the test that you added in the revert is failing on targets that don't include X86. Please fix the test, or revert it.

Thanks for noticing me, (hopefully) it should be fixed by c6bd660.

[rnk]

mentioned in issue llvm/llvm-bugzilla-archive#48834

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Erich Keane (erichkeane)

Details

| | | | --- | --- | | Bugzilla Link | [48690](https://llvm.org/bz48690) | | Version | 11.0 | | OS | Windows NT | | CC | @DougGregor,@dyung,@zmodem,@hokein,@zygoloid,@rnk |

Extended Description

See: https://godbolt.org/z/4E56c3

It seems that despite allowing this as an extension, we're still generating a RecoverExpr.

[AaronBallman]

This issue no longer reproduces for me on main: https://godbolt.org/z/6M9PzqEKh

If you're still able to reproduce this, please reopen the issue.