Issue with _dbus_write_socket on windows

[Ralf Habacker]

Havoc Pennington schrieb:
> Hi,
>
> Looking at the code you guys have in svn, it seems like the big 
> question is how the session bus gets started. On unix, we have two 
> answers:
>
>  - normally, the session bus is started by X init scripts and then
>    located with an environment variable
>
>  - as a fallback, the bus can be autolaunched and then have its address
>    stored in a file in the home directory that is
>    per-machine-per-X-display
>
> On Windows, I think just one answer is probably fine:
>
>  - the bus is always autolaunched, and only one instance of
>    it exists at a time (using whatever mechanism tbd)
>
> My guess is that the --print-address is not useful because the 
> mechanism for ensuring uniqueness will have to have a way to get the 
> address of an existing bus that someone *didn't* spawn themselves. 
> Which means there's no point having a special alternate mechanism if 
> you did spawn the bus yourself.
>
> If you did it the way you are now, it looks like you're passing a 
> handle from _pipe() down to the child. Looking at MSDN, I get the 
> impression that this kind of handle is not interoperable with the 
> native handles e.g. 
> if you used CreatePipe() (why not use CreatePipe() btw?). 
Because _pipe fits in the current read/write implementation of dbus
> But maybe these handles are interchangeable? I don't know.
We have a working _pipe implementation, which had required very less 
effort. Using the native api is possible in some areas, but needs more 
knowledge as using the platform independent c runtime.

Currently we are concentrated to get this stuff running with as less as 
possible efforts. Nice looking will be a later aim.
> In any case, I would not name the thing in question here 
> write_stream() since this can't be just any stream; it can't be a 
> socket; it has to be a pipe. So I'd name the function write_pipe() or 
> write_file_handle() or something like that. I don't know Windows well 
> enough to make the right choice.
>
This requires to rename some more references to dbus_read/write_socket 
in main.c and bus.c
> In general all the _pipe(), _open() API is some kind of weird 
> unix-compat hack that wraps the real Windows API isn't it? For this 
> windows-specific code I don't see much point using this unix compat API.
>
There are poeple with different programming background working on this 
part. Some are more unix as windows knowledged and have to investigate 
how things runs on windows and using the c runtime as much as possible 
decrease required time for implemenation features. Contributions like 
Timothy has made will help to speed up the nice making :-)


Ralf