Re: New function proposal

[Michael Goffioul]

David Bateman a écrit :
> Doesn't the code here effectively do something almost equivalent to
> popen2? In that case why can't this just be used as a MWVC/MINGW
> implementation of the popen2 function rather than introduce a new
> function? If not what is different? Do we need both popen2 and
> exec_with_pipes?
>   

Indeed, exec_with_pipes does the same as popen2. And the logical next 
step was
to use it in popen2, when the current pipe/fork implementation cannot be 
used.
I didn't know what as the best way to handle this, as popen2 is m-code 
and exec_with_pipe
is C++ code. If I wanted to keep popen2 in m-code, I needed at least 
another C++
function that I could call from popen2; this is exec_with_pipes. Another 
possibility
was to move the complete popen2 implementation to C++; then you can 
merge both.
I implemented the first approach.

A 3rd approach could be to provide octave interface to the required 
building blocks
I used in exec_with_pipes: this means CreatePipe (~= pipe), 
DuplicateHandle (~= dup2,
although I'm not sure how handle inheritance is handle by Win32 dup2) and
CreateProcess (this one would probably need a new function). Then you 
can rewrite
popen2 in m-code, as it is done with pipe/fork. But this is more work.

Michael.