On Tuesday, 8 September 2015 at 17:16:45 UTC, Jonathan M Davis
wrote:
On Saturday, 5 September 2015 at 19:59:03 UTC, bitwise wrote:
[...]
I would have to study the problem to give a good answer. It's
been too long since I looked at the issue for me to remember
the finer details at this point. IIRC, a big part of the
problem had something to do with the fact that you essentially
want a range of bytes, but the underlying C APIs really give
something more like a range of a range of bytes, because you
read the data into a buffer rather than reading it a byte at a
time (which gets even further complicated by the fact that you
generally want to reuse that buffer and not allocate extra
memory if you can avoid it). On some level at least, that's the
problem that byLine and byChunk in std.stdio run into. I really
don't remember the details of what needed to be done
differently for solid stream support though.
But all of the stuff relating to getting objects out of a byte
stream can almost certainly be done by building it on top of a
range of bytes rather than having any of that talk to the disk
or generally care where the data comes from, and I very much
doubt that std.stream does anything like that. So, at minimum,
what std.stream does would have to be shifted so that all of
the transformations are done on top of ranges of ubyte rather
than talking to the disk or network stack or whatever. But I'd
have to go digging through the code to even know what
std.stream does right now, and I don't have the time right now
to figure out how it should work. It's been discussed before,
and it was decided that what's there was not what we wanted,
though how far off it is, I really don't know other than the
fact that it's not range-based. Any stream solution should be
consuming and producing ranges and should be agnostic to
wherever the data comes from unless it has a _really_ good
reason not to be.
Steven Schveighoffer has been working off and on on a
replacement for std.stdio which supports streams and cleans up
some of the uglier things with std.stdio.File, so we may get
something out of that at some point, but he has yet to get far
enough with it to actually present it for review and possible
inclusion in Phobos. However, I haven't heard of any other work
being done on streams, and it seems like everyone's simply
making do without, for better or worse.
- Jonathan M Davis
Ok, I guess I'll try rummaging through the archives and see if I
can find the rest of the story... Or maybe just ask Steven where
he's at if this becomes important.
2.070 seems quite a ways away though. I'm not sure if DMD is on a
steady release schedule, but I imagine there is quite a lot of
work to do in terms of optimization and refactoring before even
2.069 comes out.
Anyways, Thanks for the info,
Bit
