Use u8 for type/symbol tags

[ilevkivskyi]

While trying this I found that this is a bit fragile, in the sense that write_tag(data, 1) (with a literal 1) will not get specialized and will go through slow path (btw @JukkaL is this a bug, should type of a literal in fixed int type context be inferred as fixed int?) OTOH this is probably not a big deal since no-one will use write_tag() with literals, it will always be something like write_tag(data, FUNC_DEF).

[JukkaL]

btw @JukkaL is this a bug, should type of a literal in fixed int type context be inferred as fixed int

Yes, a literal integer should always be valid if u8 is expected. I checked that this works at least in simple cases, like def f(x: u8) -> None followed by f(2).

[JukkaL]

Try reading the error value (when there is no error).

[ilevkivskyi]

Oh, good idea, I will add it to both compiled and interpreted tests.

[ilevkivskyi]

@JukkaL

I checked that this works at least in simple cases

To be clear, it doesn't give an error, but mypyc generates slow code in case where I would expect it to generate fast code. Looking more at this, I think this works correctly for code like x // 2 because we normally use custom_op for other fixed-width-related specializations, and then they are applied "manually". I think we can make mypyc a bit "smarter", so that e.g. function_op() (like I use in this PR) will be applied in more cases, but this looks low priority now.

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅