fix IO::ARGF#read should always return i32

[stakach]

seeing errors like this on nightly

[stakach]

This pull is probably not actually useful as I just made a stab at what needed to be changed

[asterite]

Probably some IO type outside of the std library is returning an Int64. I don't think that's totally wrong, so maybe we should revert this return type annotation

[stakach]

related to this pull #10580
it might be related to the definition of SizeT
https://github.com/crystal-lang/crystal/blob/master/src/lib_c/aarch64-linux-musl/c/stddef.cr#L2

[straight-shoota]

As long as Slice#size is fixed to Int32, there's no point in returning Int64 on IO#read. It's impossible to read more than Int32::MAX. So IMO the type restriction is good.

The real problem however is that the error message points to the abstract def, not the implementation which breaks the contract by returning Int64. That makes it hard to identify the culprit. I thought there was already an issue about that, but I can't find it.

[asterite]

But this is breaking someone's code. I don't think we are allowed to do that.

[straight-shoota]

Returning anything else but Int32 is a bug. Adding that type restriction in the abstract def just makes it visible.
If we want to avoid breaking any code that might be working even though it's incorrect, we can't do any fixes at all because it may break something.

[straight-shoota]

Ooops, sry I incorrectly identified this error as being caused by adding a return type to an abstract def. But this is really not the case in #10580. It adds a type restriction to the implementation. As @stakach suggested, this might be caused by a missing conversion from ˋSizeTˋ. So maybe it's our own code that's broken. We should have noticed that, though because we run CI against aarch64-musl.

@stakach Can you provide information on the target platform and possibly a reduced reproduction? (ˋ--error-traceˋ would already help as well)

[stakach]

This is where I'm seeing the error
https://github.com/PlaceOS/driver/runs/2844680637?check_suite_focus=true

Which looks like the fix is to call .to_i32 on a SizeT?

[asterite]

We can probably call to_i32 on read_count right before we return it, from ARGF.

[asterite]

Could you try changing that in this PR and see if all specs pass here, and also pass in that ssh library?

[straight-shoota]

This fix is just a workaround, though.

ARGF#read doesn't introduce a Int64 type. It just receives the result from IO#read on the wrapped IO. And IO#read should return Int32. Unfortunately, there is currently no return type restriction on that abstract def. But adding that would be the correct fix.
We probably can't do that until 2.0, though.

But I think we should add a TODO comment and open an issue to track this.