[PATCH] setup: use unbuffered I/O in py2exe

[Matt Mackall]

On Thu, 2012-08-02 at 17:30 +0400, Sergei Nosov wrote:
> # HG changeset patch
> # User Sergei Nosov <sergei.nosov at gmail.com>
> # Date 1343913078 -14400
> # Node ID 57f6129f1be240151236886fca26e5106ff01c11
> # Parent  23b247234454c945ff0b907dd6195a385c134468
> setup: use unbuffered I/O in py2exe

> If used in custom shell (e.g. GNU Emacs shell) output of any command
> was printed AFTER the command execution. Now it's printed DURING the
> command execution.

This is definitely not the right fix: I/O buffering is a good thing,
generally. 

It's standard behavior for apps to use unbuffered I/O when run on a
console/tty/pty and to automatically enable buffering when writing to
files or pipes. So either Emacs on Windows is pretending to be a console
but redirecting I/O through pipes or it's somehow foiling Python's
console detection. In either case, it's almost certainly breaking other
Mercurial assumptions too.

I'm pretty sure the problem is with the MSYS Posix emulation layer,
which apparently doesn't really support pty devices, which are key to
properly implementing things like Emacs' shell. Googling for things like
"msys pty broken" gives things like this:

http://sourceforge.net/mailarchive/message.php?msg_id=27414138

This puts this issue solidly in the realm of "not our problem".

You should instead try to arrange that PYTHONUNBUFFERED is set in your
"shell".

-- 
Mathematics is the supreme nostalgia of our time.