Handle synthesized delegates without refKinds

[davidwengier]

Fixes #58782

I made an incorrect assumption when adding support for synthesized delegates.

[AlekseyTs]

    internal static bool TryParseSynthesizedDelegateName(string name, out RefKindVector byRefs, out bool returnsVoid, out int generation, out int parameterCount)

I am assuming there is a method that produces the names that this method parses and that both methods should be in sync. Consider adding a doc comment with a reference to that other method and similar comment on that method pointing back.

---

Refers to: src/Compilers/CSharp/Portable/Symbols/Synthesized/GeneratedNames.cs:378 in 40ffc9e. [](commit_id = 40ffc9e, deletion_comment = False)

[AlekseyTs]

LGTM (commit 2)

[tmat]

    internal static bool TryParseSynthesizedDelegateName(string name, out RefKindVector byRefs, out bool returnsVoid, out int generation, out int parameterCount)

Consider adding unit test that directly invokes TryParseSynthesizedDelegateName and fully covers all branches of the parsing logic (including malformed strings).

Another simple test could also check roundripping with MakeSynthesizedDelegateName.

---

Refers to: src/Compilers/CSharp/Portable/Symbols/Synthesized/GeneratedNames.cs:378 in 40ffc9e. [](commit_id = 40ffc9e, deletion_comment = False)

[davidwengier]

Another simple test could also check roundripping with MakeSynthesizedDelegateName.

There is an Assert at the end of the method that does that.

Tests for the early exits is a good idea though.

[davidwengier]

Added documentation and new tests, which don't look very exciting, but they cover it all:

The part not covered is currently impossible to hit, as it would need the DelegateNamePrefixLength value to be incorrect (less than ActionDelegateNamePrefix or FuncDelegateNamePrefix) but I don't see any harm in leaving the condition in the code.

[cston]

Consider testing

• "<>F{01}#3`1"
• "<>F{0,1}`1"
• "<>A{0,}"
• "<>A{,1}"

[davidwengier]

Just to confirm, you're saying these should be successfully parsed? Currently two of them fail to because RefKindVector.TryParse doesn't like them, and the other two fail on roundtripping because RefKindVector.TryParse and RefKindVector.ToRefKindString aren't really round trippable (Parse will parse something that ToRefKindString will assert on).

I'm happy to fix all of that, just wanted to confirm, and confirm its not overstepping this PR.

[cston]

Consider testing

• "<>A{}"
• "<>A{,}"
• "<>A#"

[cston]

Duplicate blank lines above.

[tmat]

[davidwengier]

Thanks all!