For what it is worth, that explanation more or less mirrors my recollection of why that wasn't a zero to poll, or infinite, timeout. Working around broken something; I would have said more likely MacOS-X of the period, which I think still has a broken implementation of the `poll` syscall, but I don't fully recall.

(That was still in the era I payed significant attention to TRAMP development)

On Wed, Jan 23, 2019 at 1:38 PM Michael Albinus <michael.albinus@gmx.de> wrote:

Eli Zaretskii <eliz@gnu.org> writes:

Hi Eli,

>> At least the w32 version of accept-process-output was broken in the
>> past, see the comment in tramp-accept-process-output. But this was Emacs
>> 22 time, so hopefully it is fixed now.
>
> That comment doesn't tell much about the problem, so I couldn't verify
> that the problem it mentioned is no longer relevant.

Archeology has uncovered
<http://lists.gnu.org/archive/html/tramp-devel/2006-12/msg00007.html>,
which relates to that comment. Sorry, I haven't anything else.

The respective commit in the Tramp repository was CVS 2.516 from
10-Dec-06, if that matters :-)

In that message thread, the OP has mentioned another problem after I've
added the timeout and the comment; this seems to be unrelated.

Best regards, Michael.