Avoid a hot loop in retry(), there's no need to retry more than 50/s.
Also use luv.sleep() to implement sleep() instead of spinning the
event-loop, so events are not silently discarded.
1. Don't check elapsed time in children_kill_cb(), it's already implied
by the start-time of the timer itself.
2. Restart timer from children_kill_cb() for PTY jobs, to send SIGKILL
after SIGTERM. There is an edge case where SIGKILL might follow
SIGTERM too quickly, if jobstop() is called near the 2-second timer
window. But this edge case is not worth code complication.
* Reading from stdin on Windows is fixed in the same way as it was in
#8267.
* The file_read function was returning without filling the
destination buffer when it was called with a non-blocking file
descriptor.
Problems:
- In two places in shada_read_when_writing() memory just was not freed. Both
places were verified to cause test failures.
- Numbered marks got assigned incorrect (off-by-one compared to position in the
array) numbers in replace_numbered_mark.
- It was possible to have non-continuously populated array of numbered marks
which messed up code for merging them.
(Note about tests: marks with additional data are always compared different when
merging, that caused some confusion regarding why test did not work the way
I expected.)
MSBuild still returns a non-zero exit code because it detects the word "error" in the stdout which is caused by some of the test names such as api/buf {get,set,del}_line get_line : out-of-bounds is an error.
CMake mailing list thread:
https://cmake.org/pipermail/cmake-developers/2015-October/026775.html
There isn't any good solution for it, so I modified the build script to detect the error message printed by RunTests.cmake.
cmd.exe (shell) is faster and more reliable than powershell (.NET frontend).
It's best for short and basic tests that don't require non-trivial scripting.
cmd.exe doesn't support sleep so use powershell's Start-Sleep as substitute.
