Pattern matching for chars are broken

[amiralies]

let isA = c =>
  switch c {
  | 'a' => true
  }

produces this

// Generated by ReScript, PLEASE EDIT WITH CARE
'use strict';


function isA(c) {
  return true;
}

exports.isA = isA;
/* No side effect */

without warnigs.

[cristianoc]

Indeed thanks for reporting. Moving to the syntax repo.

[IwanKaramazow]

What makes this a syntax problem? Is this also unicode related?

[cristianoc]

No quotes are added in the generated code.

[cristianoc]

Which means it's probably a compiler problem.
And might need to transfer this back.

[cristianoc]

Not sure what the issue is. The generated code looks file.
Please reopen with an example that illustrates the issue.

[amiralies]

The above isA function is certainly includes a partial matching.
It should return true when the input is 'a' and throw Match_failure on other inputs.
The generated code for that block is something like this in 9.1.4

function isA(c) {
  if (c === 97) {
    return true;
  }
  throw {
        RE_EXN_ID: "Match_failure",
        _1: [
          "Main.res",
          2,
          2
        ],
        Error: new Error()
      };
}

[cristianoc]

Got it.
Have you observed anything except char? Is that the only type leading to this issue?

[cristianoc]

So the No side effect seems like the clue. Perhaps the code is transformed before it even reached the type checker, which would definitely complain.

[cristianoc]

It seems to be only for chars, e.g. not strings or integers. That's peculiar.

[amiralies]

Got it. Have you observed anything except char? Is that the only type leading to this issue?

Didn't see this happening with other types. just char and char interval (which is the same from typechecker pov)

[IwanKaramazow]

It seems to be only for chars, e.g. not strings or integers. That's peculiar.

Did something change with Unicode support? I remember chars now accepting codepoints

[cristianoc]

I've chased down pattern matching to receiving an integer constant, and presumably expecting a char constant.

I think there's no way to change the type of char without adapting the pattern matching also. As the notion of exhaustive matching would change.

@IwanKaramazow do you have any thoughts?

[cristianoc]

@amiralies is it true that the only issue you have observed is the detection of exhaustiveness?

[cristianoc]

Perhaps one possibility would be to make char be an alias to int, and try to get int exhaustive matching to kick in. But is that even right?

[amiralies]

@amiralies is it true that the only issue you have observed is the detection of exhaustiveness?

Yes. I'm not sure if there are other problems or no.
But it seems unused match detection is working somehow

ie

 switch c {
| 'a' => true
| 'a' => false
}

shows a warning

[IwanKaramazow]

Seems like it's a lot more involved cf. the original doc: https://docs.google.com/document/d/1V99iVjIncN7seNe-oECa_PU7Sx4ajz8xIH-qDg1vS6Y/edit

[cristianoc]

I've moved it to milestone v10.1 so there's a little time to figure it out.

[bobzhang]

Perhaps one possibility would be to make char be an alias to int, and try to get int exhaustive matching to kick in. But is that even right?

This is correct and seems cleaner. Maybe type char = private int ?

[cristianoc]

Perhaps one possibility would be to make char be an alias to int, and try to get int exhaustive matching to kick in. But is that even right?

This is correct and seems cleaner. Maybe type char = private int ?

That would be nice, though not sure how one would do that in practice.
char is used early on in the compiler, as a predefined type, so not sure where type char = private int would go to make it globally visible.

[cristianoc]

Specifically the constant 'x' would need to have type char, which is not the same as int, but the compiler internally somewhere would need to consider it as int (for pattern matching mechanics) but also not as int (for unification).