[PATCH 02 of 11 V2] procutil: ensure that all stdio file objects are flushed at interpreter exit

[Yuya Nishihara]

On Mon, 13 Jul 2020 17:14:31 +0200, Manuel Jacob wrote:
> > Didn't we agree that the use of atexit wouldn't buy and we'll instead 
> > add
> > py3-only solution later?
> > 
> > As I said, atexit will increase the complexity of the control flow. All 
> > data
> > should be flushed inside dispatch.run().
> 
> We talked past each other, then. What I meant is that for stderr it’s 
> not worth it because 1) we have other reasons to back it out, 2) stderr 
> may be used for tracebacks etc. even after the atexit hook. For stdout 
> it’s not as important because stdout is mostly written to through the ui 
> object, but still there are some contrib scripts etc. using it directly.

I see. I meant the use of atexit isn't nice at all. Sorry for confusion.

> I don’t see any problem using atexit for this particular use case, but 
> since I didn’t invest further time trying to find a case where we need 
> it for stdout (and the stderr problem was solved otherwise), I’d be fine 
> dropping the patch. On Python 3, we don’t need to do anything. What do 
> you think?

I prefer dropping this patch.

Historically we've replaced atexit() with custom ui.atexit() function for
better control of errors and execution flow, see c13ff31818b0. I don't
remember the exact problem, but doing I/O in atexit() would be undesired.