Implements the initial set of span-safety checks. #18482
Conversation
|
@OmarTawfik - please review. Thanks! |
There was a problem hiding this comment.
@VSadov My only comment is whether it makes sense to consider Span a restricted type, which they have a specific meaning in the compiler code base.
Should it be another API?
Type.IsRestrictedType()
Type.IsSpanLikeType()
In that sense, Spans are not restricted types to me, but rather their own special category. Otherwise, LGTM.
|
@OmarTawfik It is convenient that IsRestrictedType defers to IsSpanLikeType, since for nearly all purposes of this check spans are a subset of restricted types. "IsSpanLikeType" is also not a public API, yet. |
These are mostly the checks dealing with stack-only nature of the Span. With minor difference they follow the same logic as the existing checks for types such as TypedReference. ==== Not in this change: Defining and detecting span-like types is NYI. For now we just treat any type named "System.Span" and "System.ReadonlySpan" as span-like. This will change. Some of the checks result in somewhat generaic messages and happen at emit phase. That was ok when the failures were supposed to be rare. Error clarity is not the goal of this change, but we will examone what errors should say and whether they should be moved to an earlier phase.
|
@dotnet/roslyn-compiler - please take a look |
This will be allowed only for span-like containers eventually, when we can declare those. For now allow in any struct.
|
I need to merge this to test on the actual code in CoreFxLab, but will look at the PR if there are any further suggestions. |
These are mostly the checks dealing with stack-only nature of the Span. With minor differences they follow the same logic as the existing checks for types such as TypedReference.
==== Not in this change:
Defining and detecting span-like types is NYI. For now we just treat any type named "System.Span" and "System.ReadonlySpan" as span-like. This will change.
Some of the checks result in somewhat generic messages and happen at emit phase. That was ok when the errors were supposed to be rare.
Error clarity is not the goal of this change, but we will examine what errors should say and whether they should be moved to an earlier phase. (issue: #18483)