Re: [PATCH v2 1/3] test_terminal: redirect child process' stdin to a pty

[Paul Tan <pyokagan@gmail.com>]

On Fri, Aug 7, 2015 at 6:15 AM, Eric Sunshine <sunshine@sunshineco.com> wrote:
> An alternative would be to have git-am detect that it is being tested
> and pretend that isatty() returns true.

I would vastly prefer a solution that would work for everything, for
all the C code and scripts, instead of implementing a workaround in
git-am :(

In this case, I implemented a generic solution in test-terminal.perl
that works for POSIX systems, so if there are no problems with its
implementation, I do think it's better. Other than the fact that it
does not work on non-Unix platforms, of course.

The other approach I would consider is to implement a xisatty()
function that returns true for xisatty(0) if TEST_TTY=0 or something.

However, I do wonder if this would lead us to have to hack around
other functions of terminals as well (e.g. if xisatty(0),
tcgetattr()), which would be a big can of worms I think...

> There is some precedent for
> having core functionality recognize that it is being tested. See, for
> instance, environment variable TEST_DATE_NOW,

(Hmm, I took a look, and it seems that TEST_DATE_NOW is only checked
in test-date.c...)

> and rev-list --test-bitmap.
> Doing so would allow the tests work on non-Unix
> platforms, as well.

Ehh, if the non-Unix platforms do not implement terminals, it means
that the git-am logic to detect if we are attempting to feed it a
patch by checking if stdin is a TTY is invalid anyway, so implementing
a "yeah-it-is-a-tty" workaround for the sake of tests would be hiding
the problem, I think.

Thanks,
Paul