Propose to use :> instead of :?> - fixes #1127

[forki]

This is early work for #1127 and intended for first feedback round:

[dsyme]

I believe we want to undo the results of this new "solve" operation (regardless of whether it succeeds or not, using CollectThenUndo). We don't want the results of inference to be altered by giving better errors.

[forki]

[isaacabraham]

Just as a side question - why does the compiler give a warning only to then immediately afterwards give an error?

[forki]

I think this is an artifact of using fsi. @dsyme can you say more?
On May 9, 2016 2:04 PM, "Isaac Abraham" [email protected] wrote:

Just as a side question - why does the compiler give a warning only to
then immediately afterwards give an error?

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#1149 (comment)

[isaacabraham]

Ah yes. It doesn't occur through e.g. VSCode.

[dsyme]

Nice. The TcRuntimeTypeTest path is hit on four kinds of syntax:

• :? ty in patterns
• :? ty as ident in patterns
• downcast expr in expressions
• expr :?> ty

I think the error message change is only suitable for the last case. So that means you should pass in the ContextInfo.RuntimeTypeTest from each of the four callsites of TcRuntimeTypeTest and make sure we have tests for each of the case

Cheers!
Don

[forki]

@dsyme good catch ;-)

will add corresponding tests and fix that as well.

[forki]

that's better, right?

[forki]

tests added for all 4 cases.

[forki]

Seems we always gave the error twice. Probably because it is so important to get your types right.
Anyways, I fixed this as well.

[forki]

this is a slightly different version of the hint. It#s not talking about operators

[dsyme]

OK, this looks good to go to me, assuming appveyor CI passes (I know Jenkins is having trouble right now). I've removed the "WIP" .

[forki]

I rebased this.

[forki]

I think this is ready and the appveyor error is unrelated

[forki]

it's green and ready