[PATCH 2 of 3] killdaemons: use posixfile to avoid intermittent unlink errors on Windows

[Yuya Nishihara]

On Mon, 15 May 2017 00:08:31 -0400, Matt Harbison wrote:
> # HG changeset patch
> # User Matt Harbison <matt_harbison at yahoo.com>
> # Date 1494808413 14400
> #      Sun May 14 20:33:33 2017 -0400
> # Node ID 809709930080937b6b56e5cad285798f29a10280
> # Parent  024271e90987e5794dab6eb00844467065fae7e4
> killdaemons: use posixfile to avoid intermittent unlink errors on Windows
> 
> This is the aforementioned fix for the occasional cleanup error with #serve
> enabled.  There are a handful of tests that neglect to kill the daemons they
> spawned, and this code is doing a last ditch reap of them.  The test that got
> flagged was non-deterministic, and I've seen up to 3 fail in the same run.
> 
> The problem with trying to import the mercurial module is that while it is
> available for running the individual *.t files, it is not in sys.path for
> run-tests.py itself.  I couldn't think of any other way to make this work, and
> not affect sys.path for the indiviual tests.  (The main source tree _is_ in
> PYTHONPATH when this is imported from run-tests.py.)
> 
> diff --git a/tests/killdaemons.py b/tests/killdaemons.py
> --- a/tests/killdaemons.py
> +++ b/tests/killdaemons.py
> @@ -7,6 +7,16 @@
>  import sys
>  import time
>  
> +# PYTHONPATH contains the hg source tree when invoked from ./run-tests, but
> +# sys.path does not, and importing mercurial fails.  The first import works from
> +# the .t files without editing the path.
> +try:
> +    from mercurial.util import posixfile
> +except ImportError:
> +    srctree = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
> +    sys.path.insert(1, srctree)
> +    from mercurial.util import posixfile

sys.path is global. I slightly prefer moving this hack to run-tests.py if
it's okay for run-tests.py to depend on Mercurial modules.