Clarify diagnostic when constant pattern has non-constant conversion to input type

[ArcadeMode]

This proposes a fix for #63476
The newly introduced error message goes for example: error CS9096: Cannot implicitly convert type 'string' to type 'XName' in constant pattern using non-constant conversion error CS9098: Cannot implicitly convert type 'string' to type 'XName' using user-defined conversion in pattern expression

[dnfadmin]

All CLA requirements met.

[AlekseyTs]

It looks like there are no tests added or modified.

[AlekseyTs]

conversion.Operand

I think in some cases conversion.Operand might not correspond to the syntactic operand for the conversion, Therefore, the type could be surprising and the error message would not make sense because, for example, there is a conversion that supports constant folding between the types.

[ArcadeMode]

I have updated this to use expression instead - this should be the expression being bound and as I understand that does makes sense. Stepping through a few unit tests confirmed my beliefs. This change however, caused errors when the expression was a null literal so I have written a case to show ErrorCode.ERR_ValueCantBeNull when null literals are attempted to be converted to a non-nullable type.

For the following snipped:
static bool M1(Span<char> chars) => chars is null;
This yields the following message:
// (5,50): error CS0037: Cannot convert null to 'Span<char>' because it is a non-nullable value type
instead of
// (5,50): error CS0150: A constant value is expected

Affected test cases were adjusted to this change.

Looking forward to hear your feedback on this change.

[RikkiGibson]

What kind of conversion is actually used for Span<char> x = null?

[ArcadeMode]

That appears to be an ImplicitReference conversion.

This was wrong -- its an implicit user defined conversion. I have reverted the changes with respect to the null literal changes as that was not necessary.

[ArcadeMode]

The approach using expression did not work when explicit casts were involved, then the message would effectively turn into "cannot convert 'A' to 'A'". I have now used the SymbolDistinguisher with conversion.Operand as I have seen being used here and there when reporting 'no possible conversion' errors.

I did read up a little on constant folding and I do wonder, are there actually situations where constant folding results in type conversions?

[ArcadeMode]

Now, the error message in a pattern which attempts to convert null to Span would become CS9098: Cannot implicitly convert type 'char[]' to type 'Span<char>' using user-defined conversion in pattern expression.

Im feeling conflicted about this, one one hand its technically correct but it could be confusing. Should there perhaps be a special message for null?

[AlekseyTs]

Done with review pass (commit 6).

[RikkiGibson]

Thanks for opening the PR. Will take a look as soon as I am able.

[RikkiGibson]

It looks like some test baselines needs to be updated. https://dev.azure.com/dnceng-public/public/_build/results?buildId=214143&view=ms.vss-test-web.build-test-results-tab

For example:

xUnit.net 00:00:24.38]     Microsoft.CodeAnalysis.CSharp.UnitTests.CodeGen.SwitchTests.SwitchOnNullableWithNonConstant [FAIL]
  Failed Microsoft.CodeAnalysis.CSharp.UnitTests.CodeGen.SwitchTests.SwitchOnNullableWithNonConstant [18 ms]
  Error Message:
   
Expected:
                Diagnostic(ErrorCode.ERR_ConstantExpected, "i").WithLocation(8, 18)
Actual:
                // (8,18): error CS9135: A constant value of type 'int?' is expected
                //             case i:
                Diagnostic(ErrorCode.ERR_ConstantValueOfTypeExpected, "i").WithArguments("int?").WithLocation(8, 18)
Diff:
++>                 Diagnostic(ErrorCode.ERR_ConstantValueOfTypeExpected, "i").WithArguments("int?").WithLocation(8, 18)
-->                 Diagnostic(ErrorCode.ERR_ConstantExpected, "i").WithLocation(8, 18)
Expected: True
Actual:   False
  Stack Trace:
     at Microsoft.CodeAnalysis.DiagnosticExtensions.Verify(IEnumerable`1 actual, DiagnosticDescription[] expected, Boolean errorCodeOnly) in /_/src/Compilers/Test/Core/Diagnostics/DiagnosticExtensions.cs:line 99
   at Microsoft.CodeAnalysis.DiagnosticExtensions.Verify(IEnumerable`1 actual, DiagnosticDescription[] expected) in /_/src/Compilers/Test/Core/Diagnostics/DiagnosticExtensions.cs:line 48
   at Microsoft.CodeAnalysis.DiagnosticExtensions.Verify(ImmutableArray`1 actual, DiagnosticDescription[] expected) in /_/src/Compilers/Test/Core/Diagnostics/DiagnosticExtensions.cs:line 63
   at Microsoft.CodeAnalysis.DiagnosticExtensions.VerifyDiagnostics[TCompilation](TCompilation c, DiagnosticDescription[] expected) in /_/src/Compilers/Test/Core/Diagnostics/DiagnosticExtensions.cs:line 109
   at Microsoft.CodeAnalysis.CSharp.UnitTests.CodeGen.SwitchTests.SwitchOnNullableWithNonConstant() in /_/src/Compilers/CSharp/Test/Emit/CodeGen/SwitchTests.cs:line 2625
   at System.RuntimeMethodHandle.InvokeMethod(Object target, Void** arguments, Signature sig, Boolean isConstructor)
   at System.Reflection.MethodInvoker.Invoke(Object obj, IntPtr* args, BindingFlags invokeAttr)

[ArcadeMode]

I've updated the tests (I hope I got all of them). Oddly enough The failure reported here I couldn't reproduce locally as the baseline was already set to expect the new error code.. any clue what might be going on there?

[AlekseyTs]

Done with review pass (commit 37)

[AlekseyTs]

Done with review pass (commit 40)

[ArcadeMode]

addressed feedback

[AlekseyTs]

LGTM (commit 43)

[AlekseyTs]

@RikkiGibson For the second review

[RikkiGibson]

Thanks for the contribution @ArcadeMode!