[Toybox] [PATCH] pidof.test: block on sleep rather than read.

[Rob Landley]

On 1/23/20 6:24 PM, enh via Toybox wrote:
> okay, i can reproduce this locally now: just run the test via adb
> (without -t to allocate a pty). my own script conspired against me
> here by adding -t to adb if `tty -s`, so locally i was always running
> with a pty where you _can_ do a blocking read on stdout. but with just
> regular adb you're talking to a socket and see these failures.
> 
> i've attached a new version of the patch with a less vague and more
> convinced checkin comment (but the same fix)...

The problem is we came to this fix because of a previous issue this reintroduces.

Maybe we should define stdin as "always there, reading from it will block", and
make the environment we run the tests from reliably provide that? I've been
meaning to make the tests run in a subshell anyway. "cat | source hello.sh" is
kind of hilarously bad but shouldn't break anything (locals and functions would
still be present in hello.sh without explicitly exporting them)...

Except cat still depends on the inherited stdin being there. Grr. I think I want
some sort of "COMMAND | true" where when true exits COMMAND does, but not
before? But I don't want to have to write a new C program (or toybox command)
for this?

I've gotta be missing something obvious... How about:

  cat >(echo hello; read i; echo $i)

That seems to have read block, but... no, it's still wrong:

  $ cat >(echo hello)
  hello

It hangs, don't get my cursor back. Grrr. (echo exiting should close the pipe!
Is this because nobody consumed the hello output? The pipe is going the WRONG
WAY, I'd halfway expect the read to fail but blocking past close I did not
expect...)

This is hard. :(

Rob