mitogen: Fix non-blocking IO errors in first stage of bootstrap

[moreati]

When /etc/sudoers has log_output (or similar) enabled the process spawned by ctx.sudo() via mitogen.parent.Connection.start_child() receives a stdin that is in non-blocking mode. The immediate symptom is that os.openfd(0, ...).read(n) sometimes returns None, causing the first stage to raise an unhandled TypeError.

The fix (for now) is to use select.select() in as while loop to read stdin. This increases the command size slightly, but I think it's a reasonable tradeoff until/unless the cause is more fully understood.

All CI tests are now run with sudoers log_output enabled, in order to catch regressions. first_stage_test.CommandLineTest has been amended, because it relied on implementation details of the bootstrap process that are no longer true.

Fixes #1306, thanks to @Forgetyk for their detailed error reporting, diagnosis, and initial PR #1299

Preamble size

Before

SSH command size: 755
Preamble (mitogen.core + econtext) size: 18227 (17.80KiB)

                        Original           Minimized           Compressed
mitogen.core         152218 148.7KiB  68437 66.8KiB 45.0%  18124 17.7KiB 11.9%
mitogen.parent        98853  96.5KiB  51103 49.9KiB 51.7%  12881 12.6KiB 13.0%
mitogen.fork           8445   8.2KiB   4139  4.0KiB 49.0%   1652  1.6KiB 19.6%
mitogen.ssh           10827  10.6KiB   6893  6.7KiB 63.7%   2099  2.0KiB 19.4%
mitogen.sudo          12089  11.8KiB   5924  5.8KiB 49.0%   2249  2.2KiB 18.6%
mitogen.select        12325  12.0KiB   2929  2.9KiB 23.8%    964  0.9KiB  7.8%
mitogen.service       41581  40.6KiB  22398 21.9KiB 53.9%   5847  5.7KiB 14.1%
mitogen.fakessh       15767  15.4KiB   8149  8.0KiB 51.7%   2676  2.6KiB 17.0%
mitogen.master        55317  54.0KiB  28846 28.2KiB 52.1%   7528  7.4KiB 13.6%

After

SSH command size: 798
Preamble (mitogen.core + econtext) size: 18227 (17.80KiB)

                        Original           Minimized           Compressed     
mitogen.core         152218 148.7KiB  68437 66.8KiB 45.0%  18124 17.7KiB 11.9%
mitogen.parent        98944  96.6KiB  51180 50.0KiB 51.7%  12910 12.6KiB 13.0%
mitogen.fork           8445   8.2KiB   4139  4.0KiB 49.0%   1652  1.6KiB 19.6%
mitogen.ssh           10827  10.6KiB   6893  6.7KiB 63.7%   2099  2.0KiB 19.4%
mitogen.sudo          12089  11.8KiB   5924  5.8KiB 49.0%   2249  2.2KiB 18.6%
mitogen.select        12325  12.0KiB   2929  2.9KiB 23.8%    964  0.9KiB  7.8%
mitogen.service       41581  40.6KiB  22398 21.9KiB 53.9%   5847  5.7KiB 14.1%
mitogen.fakessh       15767  15.4KiB   8149  8.0KiB 51.7%   2676  2.6KiB 17.0%
mitogen.master        55317  54.0KiB  28846 28.2KiB 52.1%   7528  7.4KiB 13.6%

[moreati]

Nothing in the test suite is currently making the first stage stdin non-blocking, based on 0a247a3 passing in https://github.com/mitogen-hq/mitogen/actions/runs/16418222551.

[moreati]

Adding "Default log_output" to /etc/sudoers didn't reproduce it (based on 39b78df and https://github.com/mitogen-hq/mitogen/actions/runs/16419018956

[moreati]

Now have a failing test in 20e4be0 to form the basis of a regression test

======================================================================
ERROR: test_local (sudo_test.SudoLogOutputTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/runner/work/mitogen/mitogen/tests/sudo_test.py", line 73, in test_local
    context = self.router.sudo()
  File "/home/runner/work/mitogen/mitogen/mitogen/parent.py", line 2507, in sudo
    return self.connect(u'sudo', **kwargs)
  File "/home/runner/work/mitogen/mitogen/mitogen/parent.py", line 2452, in connect
    return self._connect(klass, **mitogen.core.Kwargs(kwargs))
  File "/home/runner/work/mitogen/mitogen/mitogen/parent.py", line 2432, in _connect
    conn.connect(context=context)
  File "/home/runner/work/mitogen/mitogen/mitogen/parent.py", line 1714, in connect
    raise self.exception
mitogen.parent.EofError: EOF on stream; last 100 lines received:
MITO000
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "<string>", line 18, in <module>
AssertionError

[moreati]

Using select.select() in the first stage is currently causing the tests to hang indefinitely. I believe the culprit is test_first_stage.Commandline, it runs subprocesss.Popen(<bootstrap args>, stdin=open('/dev/null, 'r'), ...). That worked previously because reading from /dev/null returns a zero length string, which is interepreted as EOF by file.read(). It now fails because /dev/null is always ready to read, even though it will return b''. E.g.

>>> import select, time
>>> f = open('/dev/null', 'r')
>>> f.read(1000)
''
>>> import time; time.time(); select.select([f], [], []); time.time()
1755249587.856334
([<_io.TextIOWrapper name='/dev/null' mode='r' encoding='UTF-8'>], [], [])
1755249587.856464

[moreati]

The first attempt passing all tests is 4ef8c9c. I switched the failing test from /dev/null to /dev/zero.

[moreati]

Rebased on 0.3.28dev