Re: Patch: Suppress MingW Build-Busting Net Code

[Mohan Embar <gnustuff at thisiscool dot com>]

Hi Michael,

>> ChangeLog
>> 2003-03-07  Mohan Embar  <gnustuff at thisiscool dot com>
>>
>> 	* gnu/java/nio/natSelectorImpl.cc (java_do_select): #ifdef
>> 	out functionality not implemented under mingw (_Jv_select())
>
>The correct solution would be to add a dummy _Jv_select to 
>libjava/win32.cc.

You're right, but how do you feel about Tom (and other) people
saying that such code should be separated?:

http://gcc.gnu.org/ml/java/2003-02/msg00175.html

>Are you sure this is the wrong exception ? What does mingw return when 
>connection is refused or not possible ?

I didn't phrase that too clearly. What I meant is that
the PortUnreachableException is correct and that commenting this
code block out, as I propose, would mean that an IOException would
be thrown instead. Lars had pointed this out:

http://gcc.gnu.org/ml/java/2003-02/msg00134.html

However, ECONNREFUSED isn't defined under Win32. Win32 returns
a WSAECONNREFUSED and puts it in WSAGetLastError() instead
of errno. All of the strerror stuff is busted in MingW right
now because of this. Ranjit had some clever defines to work
around this, but when I inquired about this, Tom and Ranjit
seemed to say that it would be better to just fork the code.
Things are kind of in limbo at this point and if you keep
going full speed with your java.nio.channels stuff
(BTW nice job!), then we'll face the same dilemma for these
classes too....

I had thought about the _Jv_select() in win32.cc, but I think
I subconsciously wanted to put a big nasty #ifdef just to force
us to confront this situation either one way or another.

-- Mohan
http://www.thisiscool.com/
http://www.animalsong.org/