Test failing on new darwin platforms due to warnings

[viraptor]

New macos apparently prints out warnings about xtype in unexpected places. This affects tests relying on pygraphviz not throwing warnings.

For example: https://hydra.nixos.org/build/197644084/nixlog/1

______________________ PygraphvizTest.test_binary_stream _______________________

self = <tests.test_pygraphviz.PygraphvizTest testMethod=test_binary_stream>

    def test_binary_stream(self):
        from io import BytesIO
        m = self.machine_cls(states=['A', 'B', 'C'], initial='A', auto_transitions=True,
                             title='A test', show_conditions=True, use_pygraphviz=self.use_pygraphviz)
        b1 = BytesIO()
        g = m.get_graph()
>       g.draw(b1, format='png', prog='dot')

tests/test_graphviz.py:251: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/nix/store/1k05lq8pghfp4grf3875jgf11cs19051-python3.10-pygraphviz-1.10/lib/python3.10/site-packages/pygraphviz/agraph.py:1611: in draw
    data = self._run_prog(prog, args)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <AGraph <Swig Object of type 'Agraph_t *' at 0x11017b510>>, prog = 'dot'
args = ' -Tpng'

    def _run_prog(self, prog="nop", args=""):
        """Apply graphviz program to graph and return the result as a string.
    
        >>> import pygraphviz as pgv
        >>> A = pgv.AGraph()
        >>> s = A._run_prog()  # doctest: +SKIP
        >>> s = A._run_prog(prog="acyclic")  # doctest: +SKIP
    
        Use keyword args to add additional arguments to graphviz programs.
        """
        runprog = r'"%s"' % self._get_prog(prog)
        cmd = " ".join([runprog, args])
        dotargs = shlex.split(cmd)
        p = subprocess.Popen(
            dotargs,
            shell=False,
            stdin=subprocess.PIPE,
            stdout=subprocess.PIPE,
            stderr=subprocess.PIPE,
            close_fds=False,
        )
        (child_stdin, child_stdout, child_stderr) = (p.stdin, p.stdout, p.stderr)
        # Use threading to avoid blocking
        data = []
        errors = []
        threads = [PipeReader(data, child_stdout), PipeReader(errors, child_stderr)]
        for t in threads:
            t.start()
    
        self.write(child_stdin)
        child_stdin.close()
    
        for t in threads:
            t.join()
        p.wait()
    
        if not data:
            raise OSError(b"".join(errors).decode(self.encoding))
    
        if len(errors) > 0:
>           warnings.warn(b"".join(errors).decode(self.encoding), RuntimeWarning)
E           RuntimeWarning: 2022-11-04 14:36:54.541 dot[58758:62928605] XType: com.apple.fonts is not accessible.
E           2022-11-04 14:36:54.542 dot[58758:62928605] XType: XTFontStaticRegistry is enabled.

/nix/store/1k05lq8pghfp4grf3875jgf11cs19051-python3.10-pygraphviz-1.10/lib/python3.10/site-packages/pygraphviz/agraph.py:1405: RuntimeWarning
=========================== short test summary info ============================

The same issue has been worked around in graphviz regression tests by ignoring those lines: https://gitlab.com/graphviz/graphviz/-/merge_requests/2418/diffs

[rossbar]

It looks like the warning originates from the NixOS test suite, not pygraphviz's. In this case I'd recommend adding a warnings filter to your test configuration to catch/ignore the runtime warning. Once you are able to build against the version of graphviz that has the necessary patch you should be able to remove the warnings filter.

[viraptor]

The warning originates from graphviz itself when its run on macos 12, it's not nixos-specific. Please see the issue in graphviz itself: https://gitlab.com/graphviz/graphviz/-/merge_requests/2418
They fixed it by ignoring the warnings in their test suite.

[rossbar]

The warning originates from graphviz itself when its run on macos 12, it's not nixos-specific.

Yes I understand - I'm trying to determine whether there's an action item here. We're not seeing any related errors in pygraphviz CI, likely because the macos-latest image isn't at 12 yet. If the runtime warnings are causing failures in a downstream test suite, then my recommendation from above post stands. If the proposal is to add a warnings filter to pygraphviz itself, I wouldn't necessarily be against that, but we'd have no way of testing the warnings filter until macos 12 hits the CI providers. If you're proposing that pygraphviz add a similar warning to its own test suite, I'd be +1 for that too, but again we'd likely wait until there was a way to reliably test the warnings filter.