Thanks for pointing that out. I had forgotten the backstory on it.
It strongly implies that the I/O redirection must be done BEFORE the
fork() call on Windows, as it seems that fork() communicates stdio to
exec() inside Perl. Time for test cases...
--
Rocco Caputo - rcap...@pobox.com
On Oct 2, 2009, at 09:12, Andrew Feren wrote:
Hi all,
If it helps anyone the original discussion about adding the
Win32::Console::_SetStdHandle() and related changes can be found
here. It seems pretty clear that a lot effort went into trying to
get the original patch right.
I'd really love understand why this code is needed (on not). So far
all of my tests have performed identically (other than the leak)
with or without the _SetStdHandle() calls.
-Andrew
Rocco Caputo wrote:
First, thanks to Andrew Feren, POE::Wheel::Run::Win32 has been
merged into POE::Wheel::Run.
Andrew has also reported a memory leak POE::Wheel::Run under
Windows: https://rt.cpan.org/Ticket/Display.html?id=50068
My problem is that I don't know Win32 APIs well enough to say
whether his patch does any harm. He's bypassing
Win32::Console::_SetStdHandle() calls that set STDIN, STDOUT and
STDERR to the pipes that communicate with the parent process. It
seems like reasonable code, but I don't know any better.
Any help would be appreciated.
Thank you.
