Support emitting function pointers in signatures

[333fred]

@AlekseyTs @dotnet/roslyn-compiler for review.

[AlekseyTs]

@333fred It looks like tests are failing #Closed

[AlekseyTs]

internal struct TupleTypeDecoder

It feels like other decoders should also be adjusted: nullable, dynamic, etc. #Closed

---

Refers to: src/Compilers/CSharp/Portable/Symbols/Metadata/PE/TupleTypeDecoder.cs:63 in c83dd64. [](commit_id = c83dd642d5f722a90583a2058438a16a8fc562ea, deletion_comment = False)

[AlekseyTs]

return CreateITypeSymbol [](start = 12, length = 24)

I think the code reuse doesn't give us a lot of benefits, but disallows us to assert that we properly use internal cash for public symbols. #Closed

[AlekseyTs]

return [](start = 12, length = 6)

Debug.Assert(nullableAnnotation != DefaultNullableAnnotation);? #Closed

[AlekseyTs]

ParamInfo retInfo = retAndParamTypes[0] [](start = 12, length = 51)

Why can't we do all this return type related work within the constructor? #Pending

[333fred]

We could. But given that Chuck just asked me to move what was in the constructor into a helper method, I'm going to leave this where it is.

---

In reply to: 374405675 [](ancestors = 374405675)

[AlekseyTs]

We could. But given that Chuck just asked me to move what was in the constructor into a helper method, I'm going to leave this where it is.

I am not sure how one is related to the other. I think, from the API design perspective, it would be better if constructor were taking all data in raw form, or all data in "preprocessed" form.

---

In reply to: 374952548 [](ancestors = 374952548,374405675)

[AlekseyTs]

CodeGenFunctionPointersTests [](start = 17, length = 28)

It might be useful to have targeted tests for metadata import from IL to ensure that things end up in the right places. #Closed

[AlekseyTs]

Done with review pass (iteration 2) #Closed

[AlekseyTs]

SignatureCallingConvention [](start = 27, length = 26)

It feels like translating between two enums is just another hoop to jump over. Would it be possible to remove the Cci version of the enum and switch to using SignatureCallingConvention, which, I assume, is a public type and could be used in our public APIs as well? #Closed

[333fred]

internal struct TupleTypeDecoder

I'll put a prototype. I was specifically avoiding them in this pr.

---

In reply to: 581664963 [](ancestors = 581664963)

---

Refers to: src/Compilers/CSharp/Portable/Symbols/Metadata/PE/TupleTypeDecoder.cs:63 in c83dd64. [](commit_id = c83dd642d5f722a90583a2058438a16a8fc562ea, deletion_comment = False)

[333fred]

@AlekseyTs @cston addressed feedback.

---

In reply to: 581682378 [](ancestors = 581682378)

[cston]

convention & SignatureMask [](start = 35, length = 26)

Are we silently ignoring bits outside SignatureMask? Should we throw UnsupportedSignatureContent() when decoding those cases instead? #Pending

[cston]

Cci.CallingConventionUtils.FromSignatureConvention(signatureHeader.CallingConvention) [](start = 63, length = 85)

signatureHeader.CallingConvention.FromSignatureConvention() #WontFix

[333fred]

I specifically didn't do that here because we don't import Cci today, and I'm not willing to go through and remove the Cci. qualifiers from everything else in this file.

---

In reply to: 375063146 [](ancestors = 375063146)

[333fred]

Actually, there aren't that many here, I'll just do it.

---

In reply to: 375578610 [](ancestors = 375578610,375063146)

[333fred]

there aren't that many here

But there are conflicting names between Cci. and SRM., so I'll leave it.

---

In reply to: 375582556 [](ancestors = 375582556,375578610,375063146)

[cston]

Consider extracting a helper method for use here and in CreateFromMetadata() above. #Pending

[AlekseyTs]

MakeParametersFromMetadata [](start = 70, length = 26)

It looks like there is only one call site, consider making this a local function #Pending

[AlekseyTs]

LGTM (iteration 3), with refactoring suggestions.

[ghost]

Auto-approval