My specific use case came from embedding the IPython QtConsole (http://ipython.org/ipython-doc/stable/interactive/qtconsole.html) into a Qt app.  Specifically, when a user types "print('foo')" into the QtConsole, I expect 'foo' to be printed in the QtConsole; however when "print('foo')" is called from the application code, I expect 'foo' to be printed in __stdout__ (the original stdout, i.e. the terminal).  An in fact, IPython achieves this through temporary stdout redirection while exec'ing user code (https://github.com/ipython/ipython/blob/master/IPython/kernel/inprocess/ipkernel.py#L113).
Obviously, this breaks down if application code tries to call print while redirection is taking place; more subtly, if user code (that is, typed by the user at the QtConsole) starts a new thread, then it would be nice if that new thread could inherit the current settings (that is, stdout being redirected to print to the QtConsole), which is not the case right now (delayed prints go to the terminal as the global stdout gets restored to __stdout__, not to the QtConsole).
Perhaps a bit specific, but that was the original motivation for this request.

Antony

2014-03-31 11:22 GMT-07:00 Eric Snow <ericsnowcurrently@gmail.com>:
On Thu, Feb 6, 2014 at 2:37 PM, Antony Lee <antony.lee@berkeley.edu> wrote:
> The recent discussion about lexical vs. dynamic scoping reminded me of the
> following issue, that plagues e.g. the new contextlib.redirect_stdout
> context manager: the standard file-like objects sys.std* are global instead
> of thread-local, making their manipulation unsafe in multithreaded programs
> (this issue is related to lexical vs dynamic scoping because in a language
> with (optional) dynamic scoping, these objects can be made dynamically
> scoped, thus solving the issue).

How much of a problem is this in practice?  What's your specific use
case?  I can't say that's I've ever had such a need, but then I
haven't used contextlib.redirect_stdout() in a multi-threaded context.
 At that point I'd probably opt for using Python's logging system
rather than sys.std*.

> Of course, changing sys.std* to being thread-local would be backwards
> incompatible, but perhaps some new objects, e.g. sys.thread_local_std*,
> could be added to sys, with print() and related functions using
> sys.thread_local_std* if it is set and sys.std* otherwise.

Regardless of the validity of the proposal, coming up with an API to
support thread-local sys.std* should be doable in a
backward-compatible way, as you imply.  It's just a question of how
much complexity we can justify in the underlying implementation and in
changes to sys.std* (if any).

FYI, for 3.5 I plan on working on two things related to your proposal:

* restructuring the sys module (or something like it);
* adding an API for managing thread-local contexts (a la decimal).

That's not to say that either will find enough support to actually
land, but I will be working on them. :)

-eric