Re: [PATCH] Unbuffer stdout and stderr on windows

[Eli Zaretskii <eliz at gnu dot org>]

> Date: Tue, 23 Jul 2013 14:34:32 +0800
> From: Yao Qi <yao@codesourcery.com>
> CC: <gdb-patches@sourceware.org>
> 
> On 07/22/2013 11:40 PM, Eli Zaretskii wrote:
> > This is wrong, stdout should be line-buffered by default.
> >
> 
> stdout is line-buffered if it can be detected to connect to an 
> interactive device, otherwise it is fully buffered.  Cygwin emulates pty 
> with pipes, and native win32 apps blindly assume all pipes are 
> non-interactive.

Sorry, I was unclear: I meant that stdout should be line-buffered, not
unbuffered, when it is connected to a console (which is the default).
I was under the impression that you were changing stdout to be
unbuffered even in that default case, i.e. even when stdout is
connected to a console.  If that's not so, i.e. if you were changing
the buffering only when stdout is not connected to a console, I
apologize for my misunderstanding.

> > Also, this punishes batch mode with stdout redirected to a file: its
> > stdout buffering (and perhaps also that of stderr, although that's
> > less important) will now always be line-buffered, i.e. less efficient.
> 
> Yeah, this patch hurts the performance, but gets the outputs in the 
> correct order, so that testsuite can be run to get a reasonable test result.

I don't think it's right to penalize users for the benefit of the test
suite.

> > Is it possible to detect the "Cygwin ssh session", whatever that
> > means, and only do this then?  I don't think it's right to change
> 
> Unfortunately, I am unable to find a heuristics to tell "GDB is in 
> cygwin session".  There are some differences on env variables between 
> cygwin and Windows cmd.exe console, but I am afraid that they are not 
> reliable.

Then how about introducing a special option which will cause the
buffering to be what you want?  The test suite could use that option,
and the other users will not suffer any penalty.

> > behavior of a native w32 GDB just because it misbehaves when mixed
> 
> This patch is to change the buffered output to unbuffered, so the 
> behaviour of GDB is not changed, IMO.

Of course it's changed: buffering is user-visible behavior.

> In the last resort, we may add an option "--cygwin-tty"

I think this is the best alternative.

> which I don't really like.

Why not?