faster isascii(::Char)

[stevengj]

The previous implementation of isascii(c::Char), by @StefanKarpinski in #24999, was bswap(reinterpret(UInt32, c)) < 0x80. However, I noticed that base/io.jl is using a nicer trick: c ≤ '\x7f', which oddly enough is also by @StefanKarpinski in #24999. The latter seems to compile to one fewer instruction because it omits the bswap.

@StefanKarpinski, any comment here?

The PR also cleans up io.jl to use isascii(c) rather than c ≤ '\x7f', since they should now be equivalent.

[matthias314]

The new version looks cleaner, but are you sure that it's faster? On master I've tried

s = String(' ':'~')
@btime all(isascii, $s)

and the run time is the same as before.

[stevengj]

The new version looks cleaner, but are you sure that it's faster?

Not sure; it's really hard to benchmark such a simple function. However, I'm operating on the principle that 1 cmp instruction is better than 1 cmp + 1 bswap.

Also, it's weird to have two different versions of the isascii check in our codebase — seems like we should settle on one, and we might as well settle on the one with fewer instructions.

[stevengj]

Actually, it looks like the c ≤ '\x7f' is simpler because it is simply wrong for malformed Char

julia> c = reinterpret(Char, 0x7f000000 - 0x01)
'\x7e\xff\xff\xff': Malformed UTF-8 (category Ma: Malformed, bad data)

julia> c ≤ '\x7f'
true

julia> isascii(c)
false

So, this actually seems like a bug in io.jl.

I'll close this PR then, change io.jl to use isascii, and add a test.