math.tan has poor accuracy near pi/2 on OS X Tiger

[mdickinson]

BPO	27953
Nosy	@ronaldoussoren, @mdickinson, @ned-deily

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/mdickinson'
closed_at = <Date 2016-09-04.10:24:35.287>
created_at = <Date 2016-09-04.08:35:39.552>
labels = ['extension-modules', 'OS-mac', 'type-bug']
title = 'math.tan has poor accuracy near pi/2 on OS X Tiger'
updated_at = <Date 2016-09-05.18:01:13.201>
user = 'https://github.com/mdickinson'

bugs.python.org fields:

activity = <Date 2016-09-05.18:01:13.201>
actor = 'mark.dickinson'
assignee = 'mark.dickinson'
closed = True
closed_date = <Date 2016-09-04.10:24:35.287>
closer = 'mark.dickinson'
components = ['Extension Modules', 'macOS']
creation = <Date 2016-09-04.08:35:39.552>
creator = 'mark.dickinson'
dependencies = []
files = []
hgrepos = []
issue_num = 27953
keywords = []
message_count = 7.0
messages = ['274350', '274352', '274353', '274395', '274406', '274409', '274413']
nosy_count = 4.0
nosy_names = ['ronaldoussoren', 'mark.dickinson', 'ned.deily', 'python-dev']
pr_nums = []
priority = 'normal'
resolution = 'wont fix'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue27953'
versions = ['Python 2.7', 'Python 3.5', 'Python 3.6']

[mdickinson]

Opening this for the record; I intend to close as "wont fix".

The tan function has poor accuracy for inputs near pi/2 on OS X 10.4. This is a direct consequence of a poor libm implementation, so there's little we can do about it short of re-implenting tan ourselves. 10.4 is ancient enough by now that it's difficult to care too much.

Example failures:

======================================================================
FAIL: test_testfile (test.test_math.MathTests)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/Users/db3l/buildarea/3.x.bolen-tiger/build/Lib/test/test_math.py", line 1190, in test_testfile
    '\n  '.join(failures))
AssertionError: Failures in test_testfile:
  tan0064: tan(1.5707963267948961): expected 1978937966095219.0, got 1978945885716843.0 (error = 7.92e+09 (31678486496 ulps); permitted error = 0 or 5 ulps)

and a corresponding failure in cmath:

======================================================================
FAIL: test_specific_values (test.test_cmath.CMathTests)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/Users/db3l/buildarea/3.x.bolen-tiger/build/Lib/test/test_cmath.py", line 398, in test_specific_values
    msg=error_message)
  File "/Users/db3l/buildarea/3.x.bolen-tiger/build/Lib/test/test_cmath.py", line 147, in rAssertAlmostEqual
    '{!r} and {!r} are not sufficiently close'.format(a, b))
AssertionError: tan0064: tan(complex(1.5707963267948961, 0.0))
Expected: complex(1978937966095219.0, 0.0)
Received: complex(1978945885716843.0, 0.0)
Received value insufficiently close to expected value.

[python-dev]

New changeset b4d52df5595e by Mark Dickinson in branch 'default':
Issue bpo-27953: skip failing math and cmath tests for tan on OS X 10.4.
https://hg.python.org/cpython/rev/b4d52df5595e

[mdickinson]

The relevant tests are now skipped on OS X < 10.5, and test_math and test_cmath are passing on the OS X Tiger buildbot. Closing (as "wont fix", since we haven't actually fixed the underlying issue).

[ned-deily]

It would be nice to use the existing @requires_mac_ver() decorator to skip the tests (see Lib/test/support/init.py).

[mdickinson]

Ned: that's for skipping an entire test method, right? Here a single test method tests a few hundred different special cases; we want to exclude only one of those cases.

[ned-deily]

Mark, ah, yes, that's true. So the tests would have to be separated out into a separate test case method. So, not worth the effort.

[mdickinson]

Given enough time, it would make a lot of sense to adapt these tests to use the subtests machinery (though it's not immediately clear to me how to make @requires_mac_ver work with the subtests machinery). I'm afraid this issue was a bit of a quick fix to get the buildbots back to working state.