"as" pattern with an identifier picks a literal

[IS4Code]

There has been a change in F# 6 that introduces arbitrary patterns to the right of as. However, this is also a breaking change in case an identifier is used that starts with an uppercase letter, which was an already valid syntax before F# 6 to create a new variable.

Repro steps

[<Literal>]
let Co = 2
match 1 with
| 1 as Co -> printf "1"
| _ -> printf "other"

Expected behavior

The match should succeed, bind 1 to a new variable Co, and print "1".

Actual behavior

"other" is printed, since it interprets the first pattern as 1 as 2.

Known workarounds

Not using identifiers after as that start on an uppercase letter.

Related information

I am not sure if this behaviour is by design or not, since the RFC that introduces it does not discuss this case. However, I believe it is not by design, since the feature is marked as not being a breaking change. If the breaking change stays in, warnings should be issued since the interpretation of the code differs.

Environment: https://sharplab.io/#v2:DYLgZgzgPg2gPAGQJYBcCmAnAhsAfAXQFgAoYNFAAgGEB7CgXgoCYSBbLFAYwAsKBGCgHdU3ElH4UsEanQC0uCgAcMSAHYowFAER8tYigH0K8pSvWatNFN0xagA=

Trying this in an older version of the compiler exhibits the expected behaviour.

[brianrourkeboll]

• not according to the specification.

Do you mean the F# 4.1 specification?

Arbitrary patterns on the right-hand side of as were intentionally added in F# 6:

• Language suggestion: Enable pattern matching on more specific types from type test patterns fsharp/fslang-suggestions#1025
• RFC: https://github.com/fsharp/fslang-design/blob/main/FSharp-6.0/FS-1105-Non-variable-patterns-to-the-right-of-as-patterns.md

[IS4Code]

@brianrourkeboll Oh, thank you for the reference, I suppose I got failed there by the official documentation, 4.1 specification, and discord. I really wish there was a way to find out about this stuff without stumbling upon it by coincidence.

Since the & behaviour has been confirmed to be valid since F# 6 (and I am certainly glad it is!), I will edit the issue to leave only the 1 as Co part in. I believe that part is an undocumented breaking change.

[brianrourkeboll]

I really wish there was a way to find out about this stuff without stumbling upon it by coincidence.

Yeah, there is a community effort under way to try to make it easier to keep the spec up to date: https://github.com/fsharp/fslang-spec?tab=readme-ov-file#overview

This is an initiative to create a more complete and community-maintainable F# spec.

This will be no small task, but we believe it is worthwhile and we count on community contributions.

We foresee three phases:

1. Convert the latest official spec to markdown and create the structure and tools to make it community-maintainable. This is done.
2. Add the post-4.1 features as documented in the RFCs to the spec. Our goal: a complete F# 10 spec.
3. Make spec update part of new feature development so that an up-to-date spec can be released with every new major compiler release.

Right now, I believe it is in phase 2.

I will edit the issue to leave only the 1 as Co part in. I believe that part is an undocumented breaking change.

Yes, I think you are right that this was a breaking change. Note however that the compiler would have already emitted a warning for any identifier three characters or longer (#17878) starting with an uppercase letter in this position, which probably greatly limited any effect this might have had. (Edit: well, maybe it would have; maybe not, if identifiers in that position didn't go through the regular pattern typechecking. Either way, I guess we would probably have seen more noise in the last three years if it was really a widespread problem.)

[T-Gro]

I think this RFC can be revised to reduce the scope of supported expressions on the righthand side of as.
In this specific instance, I think constant as anotherConstant could emit a new warning without introducing breaking changes.

The described problem would however still remain if Co were e.g. a union case, but that behavior was the intended one for FS-1105.