GitLab

CI was failing due to changes in the python environment install.

The returncode is a useful property to be able to pull out.
The iteration would terminate early if the process had already exited
by the time that the call was made. We now follow the documentation
and return the results until the buffer is empty.

Windows support work

See merge request !1

All the command declarations are now at the top of the file within
a check for the platform. Hopefully this should make it easier to
manange.

The 'netstat -i -c' command, when run as an unprivileged user, appears
to output nothing. This makes our tests fail because we rely on it
outputing something every second.

Change to using ping.

When the error number was used on Linux and BSD, the values for the
Resource Unavailable (EWOULDBLOCK) were different. For no good reason,
this was using a hard-coded number, when it should have used a symbol.
This fixes a few of the linux tests.

The failure cases when the threaded tests were not working were not great.
They would result in hanging processes, which didn't help us very much.
The processes are now tracked and we try to destroy as part of the tear
down code.

The linux and darwin systems have different implementations of the
'netstat' tool we are using for checking that the threaded version does
stream properly. So these differences are very roughly abstracted. This
should be done better in the future, when we add Windows support.

The 'netstat -i -c' command, when run as an unprivileged user, appears
to output nothing. This makes our tests fail because we rely on it
outputing something every second.

Change to using ping.

When the error number was used on Linux and BSD, the values for the
Resource Unavailable (EWOULDBLOCK) were different. For no good reason,
this was using a hard-coded number, when it should have used a symbol.
This fixes a few of the linux tests.

The failure cases when the threaded tests were not working were not great.
They would result in hanging processes, which didn't help us very much.
The processes are now tracked and we try to destroy as part of the tear
down code.

The linux and darwin systems have different implementations of the
'netstat' tool we are using for checking that the threaded version does
stream properly. So these differences are very roughly abstracted. This
should be done better in the future, when we add Windows support.

Done some time back.

The newest version cleans up better.

It seems that __globals__ isn't known to the pylint tool, and we use
that to recognise the context of the operations. So we have to make
pylint ignore that particular instance.

Some pylint fixes, including some that appear to have been actual bugs -
the inspect module was never imported, so it's not clear why the test
code worked.

Python 3 is now better supported by the tests, and should be nearly
working with these changes. There are also a few changes to try to make
it easier to work on Windows, but this is still not working properly.

When importing this module into other projects, it is important that
we can actually see the imports we've created. It's only a very
simple import line, but very important!

The output on stderr being merged with the stdout was causing problems
when errors were output on that stream. We now have the option to
not write them out there.

The bootstrap-env wasn't working properly because we didn't give a default
for $TMPDIR. We now use /tmp as the default.

Provides callback functions which are supplied with data as it is
received from a subprocess.