JSString layout is broken

[Vizepi]

JSString::is_wide_char and JSString::atom_type are using uint8_t type instead of uint32_t.

This prevents bitfields to function as intended, and struct ends up being 48 bytes instead of 40 bytes.

I did not investigate further as changing the size of this struct may have side effects

[bnoordhuis]

With what compiler and compile flags? It's 40 bytes on my linux system, with gcc and tcc.

Let me guess: you're seeing 48 byte allocations? That's malloc rounding up to the next bucket size.

[Vizepi]

I'm using MSVC v143, and it's not the allocation size. It breaks compilation if I add a static assert on sizeof == 40. And static assert then pass if I replace uint8 by uint32

[chqrlie]

This looks like a boolean size use 32-bit instead of 8. We should not use boolean types at all.

[Vizepi]

The exact issue can be reduced to this example:
https://godbolt.org/z/PfMT98Yrr
It is indeed only an issue on MSVC

[bnoordhuis]

This looks like a boolean size use 32-bit instead of 8. We should not use boolean types at all.

JSString doesn't have bool fields.

It is indeed only an issue on MSVC

Last time I tried to change the way we use bitfields, it caused tcc to miscompile programs, but that may have been fixed upstream. I'll give it a shot.

[Vizepi]

Should we maybe add static asserts on the expected size of data so we can catch this kind of issue in the future?

[bnoordhuis]

I don't think so, that's brittle.