Add linux aarch64 wheel build support

[odidev]

Added linux aarch64 wheel build support. Related to https://github.com/jnwatson/py-lmdb/issues/298. @jnwatson Could you please review this PR?

[odidev]

Please remove cp35-cp35m on line 153. This wasn't related to your change, but CI won't pass until it is removed.

Done.

[jnwatson]

One test failed. `___ PreloadTest.test_preload ___

self =

@unittest.skipIf(not sys.platform.startswith('linux'), "test only works on Linux")
def test_preload(self):
    """
    Test that reading just the key doesn't prefault the value contents, but
    reading the data does.
    """

    import resource
    self.c.put(B('a'), B('a') * (256 * 1024 * 1024))
    self.txn.commit()
    self.env.close()
    # Just reading the data is obviously going to fault the value in.  The
    # point is to fault it in while the GIL is unlocked.  We use the buffers
    # API so that we're not actually copying the data in.  This doesn't
    # actually show that we're prefaulting with the GIL unlocked, but it
    # does prove we prefault at all, and in 2 correct places.
    self.path, self.env = testlib.temp_env(path=self.path, writemap=True)
    self.txn = self.env.begin(write=True, buffers=True)
    self.c = self.txn.cursor()
    minflts_before = resource.getrusage(resource.RUSAGE_SELF)[6]
    self.c.set_key(B('a'))
    assert bytes(self.c.key()) == B('a')
    minflts_after_key = resource.getrusage(resource.RUSAGE_SELF)[6]

    self.c.value()
    minflts_after_value = resource.getrusage(resource.RUSAGE_SELF)[6]

    epsilon = 20

    # Setting the position doesn't prefault the data

  assert minflts_after_key - minflts_before < epsilon

E AssertionError: assert (88293 - 88273) < 20`

Bumping epsilon up to 30 should do it.

[odidev]

One test failed. `___ PreloadTest.test_preload ___

self =

@unittest.skipIf(not sys.platform.startswith('linux'), "test only works on Linux")
def test_preload(self):
    """
    Test that reading just the key doesn't prefault the value contents, but
    reading the data does.
    """

    import resource
    self.c.put(B('a'), B('a') * (256 * 1024 * 1024))
    self.txn.commit()
    self.env.close()
    # Just reading the data is obviously going to fault the value in.  The
    # point is to fault it in while the GIL is unlocked.  We use the buffers
    # API so that we're not actually copying the data in.  This doesn't
    # actually show that we're prefaulting with the GIL unlocked, but it
    # does prove we prefault at all, and in 2 correct places.
    self.path, self.env = testlib.temp_env(path=self.path, writemap=True)
    self.txn = self.env.begin(write=True, buffers=True)
    self.c = self.txn.cursor()
    minflts_before = resource.getrusage(resource.RUSAGE_SELF)[6]
    self.c.set_key(B('a'))
    assert bytes(self.c.key()) == B('a')
    minflts_after_key = resource.getrusage(resource.RUSAGE_SELF)[6]

    self.c.value()
    minflts_after_value = resource.getrusage(resource.RUSAGE_SELF)[6]

    epsilon = 20

    # Setting the position doesn't prefault the data

  assert minflts_after_key - minflts_before < epsilon

E AssertionError: assert (88293 - 88273) < 20`

Bumping epsilon up to 30 should do it.

Done.

[jnwatson]

Sorry, I couldn't get it working. github actions is having trouble running the qemu container. See https://github.com/jnwatson/py-lmdb/runs/4588510079?check_suite_focus=true

[jnwatson]

Please disregard my last comment. Docker was down yesterday.