decode non-UTF-8 strings as Latin-1 instead of panicking

[LinusU]

This fixes Moto Racer which has a copyright symbol in the window title 😁

I think that using a Cow was quite smart, and it is a quite small patch. But potentially we'd want to always read the strings as Windows-1252 maybe? 🤔

From Claude Fable 5.1 (high):

Programs of this era use Windows-1252, and a window title with a copyright sign in it took read_str's from_utf8 unwrap down. read_str now returns a Cow: borrowed for valid UTF-8, an owned Latin-1 decode otherwise.

The first commit adds Memory::read_cstr and moves the byte-oriented callers onto it (lstrlen/lstrcpy/lstrcat, wsprintfA, the NLS length computations), so decoding never changes a length or re-encodes a copy.

[evmar]

I think having read_cstr return &[u8] is right. The read_str thing returning a UTF-8 String was always a hack, and the .unwrap() in there was intended exactly to help notice when this case came up.

My understanding is that the fully-correct thing to do is obey the current locale. I doubt we'd ever care to implement such a thing though. So maybe the main concern is making it easy to find this if anyone ever does care? With that framing, I think the .read_str() is sort of a "this is a convenient but technically wrong" method, which means having it return a String is probably better, since any caller that really cares about correctness will use some locale-aware method.

This is all an incredibly tiny detail and what you have is probably fine too. If it's easy to switch to non-Cow I'd prefer it, what do you think?

(Also, somehow I woke up this morning and thought "I haven't heard from Linus in a while, wonder what's going on", then opened GitHub and saw I'd somehow missed all of these, sorry for the delay!)

[LinusU]

Yeah, makes sense! I pushed an update.

Let me know if you want me to log something in the Err-path, e.g. when a program has a copyright-symbol in the titlebar 😁

From Fable 5.1:

Switched read_str to return a String and dropped the into_owned calls, with a doc comment saying it's the convenient reading and pointing at read_cstr for anyone who cares about the encoding. One thing the fallback loses is the unwrap's "notice when this happens" signal; happy to add a log::debug on the Latin-1 path if you'd like that kept.

(No worries on the delay! Glad you thought to check in 🙌)