[kris]

Dsc is now checked-in to subversion, and is browsable via the "projects" tab above. A download is available via the project download section.

If you look in the repository root you'll find a readme.txt and a quickref.doc. The former contains revision history, whilst the latter provides a general overview of Dsc functionality. There's also an dsc_io_overview.doc which discusses the design of the IO package. File unittest.d provides a variety of code examples.

A few things will not operate correctly until dmd v0.83 is released, due to Interface runtime issues. Those classes include TextWriter, ColumnWriter, Printf, and SerializableWriter. On dmd v0.82 they will fail under particular circumstances.

Please note that a couple of things are not implemented on linux as yet. These items include FileProxy.toList() and FileConduit.truncate(), and dependents such as FileBucket will not operate without them. Is there anyone with a linux box who is willing to make these operational?

There are undoubtedly some bugs lurking in there ... please post any issues you run into, and I'll get them resolved ASAP.

Lastly: I do hope you'll find these packages to be useful either as part of your own projects, or perhaps as part of learning the D language; not much point in my writing them otherwise <g>

- Kris

[kris]

Dsc.io now supports memory-mapped files (up to 2GB) on Win32 platforms. Buffers, Readers and Writers operate on such files in the same manner as any other file. Use FileConduit.createMappedBuffer() to take advantage of this.

64bit systems will extend the 2GB limit.

[kris]

Now up to 4GB (actually 4GB - 1) of memory-mapped file IO, on 32bit platform.

[kris]

please note that the latest download is not actually on the Project "home page" -- instead it is over here: http://svn.dsource.org/svn/projects/dsc/ in file dsc.zip

We'll do something about keeping the home-page-download "up to date" during next week.

[JJR]

Walter has just released dmd 0.86 with a number of language updates. I assume the non-working portions of dsc will now compile and run properly? (haven't tried yet).

[kris]

[tyfon]

I'll have a swing at those linux functions. Have to burn to cd and take home, so dont expect it to happen within a few hrs
_________________
--
Øyvind Jægtnes

[tyfon]

I've been looking a bit through the code, and since im not very used to OO programming, i've only seen a couple things i'd put in so far.

Socket.d:
in closure()...

[JJR]

Shared libraries cannot be made for Linux yet using dmd.

see http://www.digitalmars.com/d/dcompiler.html#general

Check under Linux bugs.

[kris]

Hi Tyfon,

John and I did play around with both suggestions, but to no avail. I would agree that it's best to shutdown a socket before closing it, though that's traditionally been done on the client side (not that tradition is necessarily good!). Part of the problem here lies in the fact that there's really no clear rules involved regarding 'soft' termination (and perhaps the client/browser has already cancelled the request...). Google will happily unearth hundreds of folks with differing opinions <g>. And then Win32 does things differently than BSD, and apparently some of the BSD implementation are slightly different in this regard ... tricky, but perhaps best to use shutdown() as you say.

ReuseAddress() should definately be in there also, but not for the reason that you appear to state. I could easily be misunderstanding, but you note that the server will run out of sockets if it doesn't apply reuseAddress(). That's not quite accurate, since (a) there's only one listening socket in use and (b) reuseAddress() is simply about ensuring you can bind multiple listeners on the same port. This latter aspect is typically applied such that a halted server can be restarted again immediately, rather that waiting for the OS to scavange the previous listener. Regardless, it should definately be there for this purpose (the walkthrough is for the benefit of others, who may now decide to correct me!)

Both good points.

- Kris

[kris]