30-Aug-2013 21:20, monarch_dodra пишет:
On Friday, 30 August 2013 at 15:16:15 UTC, Dmitry Olshansky wrote:
I think the problem is "put" overstepping its boundaries. If
"r.put(someSlice)" compiles, "put" has no reason to think that R
actually owns the elements in the slice.
It should and this is where we differ I guess. I can't think of a
useful output range that stores away aliases to slices it takes.
Not "aliases to slices", but slices themselves. For example, a
dictionary, which is a container of "words" (strings) could define a
sink that accepts strings to feed it word.
The interesting observation is that one can safely alias string or for
that matter any immutable slices. However any such activity with mutable
slices is prone to some funky issues, continuing my point that int[][]
is bad output range :)
Or, well anything that
defines the *element* itself as the object. For example, something that
accumulates *lists* of ints.
Correcting myself - anyway it should deal with potential _mutable_
aliasing if it is to keep them around for some reason. If the output
range can just use it in place - fine.
In any case, I get your point about functionality. I can rework my pull
to make it work as before, while still keeping the trans-coding
functionality :/
But I'm not a huge fan.
Me neither.
--
Dmitry Olshansky
