More improvements to inference in IO #36449
Conversation
| @@ -289,7 +289,7 @@ function _unique!(f, A::AbstractVector, seen::Set, current::Integer, i::Integer) | |||
| end | |||
| i += 1 | |||
| end | |||
| return resize!(A, current - firstindex(A) + 1) | |||
| return resize!(A, current - firstindex(A) + 1)::typeof(A) | |||
There was a problem hiding this comment.
I've been wondering if we should separate "call API" and "overload API." It makes much easier to imposing an invariance like this one. For example, we can have
function resize!(a, n)
__resize!__(a, n)
return a::typeof(a)
endIt asserts the return type and also makes sure that the object a is returned. This is useful for something like push! where some implementations forget to return the correct object #34274.
| readuntil(io::AbstractPipe, arg::AbstractVector; kw...) = readuntil(pipe_reader(io), arg; kw...) | ||
| readuntil_vector!(io::AbstractPipe, target::AbstractVector, keep::Bool, out) = readuntil_vector!(pipe_reader(io), target, keep, out) | ||
| readbytes!(io::AbstractPipe, target::AbstractVector{UInt8}, n=length(target)) = readbytes!(pipe_reader(io), target, n) | ||
| write(io::AbstractPipe, byte::UInt8) = write(pipe_writer(io)::IO, byte) |
There was a problem hiding this comment.
Does it make sense to define _pipe_writer(io) = pipe_writer(io)::IO?
There was a problem hiding this comment.
Not a bad idea, but annotating each call site seems reasonable too. The _pipe_writer solution would fall apart if people start defining new _pipe_writer methods
There was a problem hiding this comment.
I decided to adopt that strategy in #36453, since there were so many calls to buffer. Here, however, I could imagine there being external extensions of pipe_writer.
timholy
force-pushed
the
teh/more_io_inference
branch
from
0d11aed to
251b104
Compare
timholy
force-pushed
the
teh/more_io_inference
branch
from
251b104 to
0a6247e
Compare
|
I'll merge soon barring any further concerns. |
Also continues the work in #36280.