On Thu, 14 Jul 2011 06:16:37 -0400, David Nadlinger <s...@klickverbot.at>
wrote:
On 7/14/11 12:05 PM, Jonathan M Davis wrote:
On Thursday 14 July 2011 10:55:49 Regan Heath wrote:
On Thu, 14 Jul 2011 08:14:10 +0100, Jonathan M
Davis<jmdavisp...@gmx.com>
wrote:
I did have to create a 64-bit version for std.datetime, so it has a
private
function called endianSwap64 to do the job. So, it's not like I'm
saying
that
the situation couldn't be improved, but druntime and Phobos do
currently
give
you the exact same thing that C and C++ do.
The crypto library I have been meaning to polish up and submit for
inclusion into phobos needs these functions in several variants. I was
hoping they could/would be included somewhere more 'general' at some
stage, and any duplicates like those already mentioned removed..
Oh, it could be done, and it probably should be done. But other than
the 64-
bit version (since it's not particularly standard in C or C++), the
functions
are there if you need them.
Just for the record, I have my own version for 2/4/8 bytes as well, e.g.
at
https://github.com/klickverbot/thrift/blob/d-gsoc/lib/d/src/thrift/protocol/base.d#L436.
I would appreciate if we could add a general (but fast for the common
cases) version to Phobos.
I like your implementation. However, there are a couple things:
1. doing this: (t & 0xff00) >> 8 is subject to sign extension, it's
better to do (t >> 8) & 0x00ff. I think your current code has a bug for
the value cast(short)-10000, and in fact, any negative short less than -1.
2. in the 64-bit version, you are probably better off using bswap directly
instead of calling byteSwap recursively for the upper 32 bits, no?
Also, there should probably be a higher priority branch that will call a
member function if it exists on the type (think in terms of custom integer
types).
-Steve
