Known Issues

While most bugs and issues are managed using the astropy issue tracker, this document lists issues that are too difficult to fix, may require some intervention from the user to work around, or are caused by bugs in other projects or packages.

Issues listed on this page are grouped into two categories: The first is known issues and shortcomings in actual algorithms and interfaces that currently do not have fixes or workarounds, and that users should be aware of when writing code that uses astropy. Some of those issues are still platform-specific, while others are very general. The second category is of common issues that come up when configuring, building, or installing astropy. This also includes cases where the test suite can report false negatives depending on the context/ platform on which it was run.

Known Deficiencies

Quantities Lose Their Units with Some Operations

Quantities are subclassed from NumPy’s ndarray and in some NumPy operations (and in SciPy operations using NumPy internally) the subclass is ignored, which means that either a plain array is returned, or a Quantity without units. E.g., prior to astropy 4.0 and numpy 1.17:

>>> import astropy.units as u
>>> import numpy as np
>>> q = u.Quantity(np.arange(10.), u.m)
>>> np.hstack((q,q)) 
<Quantity [0., 1., 2., 3., 4., 5., 6., 7., 8., 9., 0., 1., 2., 3., 4., 5.,
           6., 7., 8., 9.] (Unit not initialised)>

And for all versions:

>>> ratio = (3600 * u.s) / (1 * u.h)
>>> ratio 
<Quantity 3600. s / h>
>>> np.array(ratio) 
>>> np.array([ratio]) 

Workarounds are available for some cases. For the above:

<Quantity 285. m2>

>>> np.array( 

>>> u.Quantity([q, q]).flatten() 
<Quantity [0., 1., 2., 3., 4., 5., 6., 7., 8., 9., 0., 1., 2., 3., 4., 5.,
           6., 7., 8., 9.] m>

An incomplete list of specific functions which are known to exhibit this behavior (prior to astropy 4.0 and numpy 1.17) follows:


  • numpy.hstack, numpy.vstack, numpy.c_, numpy.r_, numpy.append

  • numpy.where

  • numpy.choose

  • numpy.vectorize

  • pandas DataFrame(s)


Care must be taken when setting array slices using Quantities:

>>> a = np.ones(4)
>>> a[2:3] = 2*
>>> a 
array([1., 1., 2., 1.])
>>> a = np.ones(4)
>>> a[2:3] = 1*
>>> a 
array([1., 1., 1., 1.])

Either set single array entries or use lists of Quantities:

>>> a = np.ones(4)
>>> a[2] = 1*
>>> a 
array([1.  , 1.  , 0.01, 1.  ])
>>> a = np.ones(4)
>>> a[2:3] = [1*]
>>> a 
array([1.  , 1.  , 0.01, 1.  ])

Both will throw an exception if units do not cancel, e.g.:

>>> a = np.ones(4)
>>> a[2] = 1* 
Traceback (most recent call last):
TypeError: only dimensionless scalar quantities can be converted to Python scalars


Numpy array creation functions cannot be used to initialize Quantity

Trying the following example will throw an UnitConversionError on NumPy before version 1.20 and ignore the unit in later versions:

>>> my_quantity = u.Quantity(1, u.m)
>>> np.full(10, my_quantity)  
Traceback (most recent call last):
UnitConversionError: 'm' (length) and '' (dimensionless) are not convertible

A workaround for this at the moment would be to do:

>>> np.full(10, 1) << u.m
<Quantity [1., 1., 1., 1., 1., 1., 1., 1., 1., 1.] m>

As well as with full one cannot do zeros, ones, and empty.

Quantities Lose Their Units When Broadcasted

When broadcasting Quantities, it is necessary to pass subok=True to broadcast_to, or else a bare ndarray will be returned:

>>> q = u.Quantity(np.arange(10.), u.m)
>>> b = np.broadcast_to(q, (2, len(q)))
>>> b 
array([[0., 1., 2., 3., 4., 5., 6., 7., 8., 9.],
       [0., 1., 2., 3., 4., 5., 6., 7., 8., 9.]])
>>> b2 = np.broadcast_to(q, (2, len(q)), subok=True)
>>> b2 
<Quantity [[0., 1., 2., 3., 4., 5., 6., 7., 8., 9.],
           [0., 1., 2., 3., 4., 5., 6., 7., 8., 9.]] m>

This is analogous to the case of passing a Quantity to array:

>>> a = np.array(q)
>>> a 
array([0., 1., 2., 3., 4., 5., 6., 7., 8., 9.])
>>> a2 = np.array(q, subok=True)
>>> a2 
<Quantity [0., 1., 2., 3., 4., 5., 6., 7., 8., 9.] m>


Quantities Float Comparison with np.isclose Fails

Comparing Quantities floats using the NumPy function isclose fails on NumPy versions before 1.17 as the comparison between a and b is made using the formula

\[|a - b| \le (a_\textrm{tol} + r_\textrm{tol} \times |b|)\]

This will result in the following traceback when using this with Quantities:

>>> from astropy import units as u, constants as const
>>> import numpy as np
>>> np.isclose(500 *, 300 * / u.s)  
Traceback (most recent call last):
UnitConversionError: Can only apply 'add' function to dimensionless quantities when other argument is not a quantity (unless the latter is all zero/infinity/nan)

If one cannot upgrade to numpy 1.17 or later, one solution is:

>>> np.isclose(500 *, 300 * / u.s, atol=1e-8 * / u.s)

Quantities in np.linspace Failure on NumPy 1.10

linspace does not work correctly with quantities when using NumPy 1.10.0 to 1.10.5 due to a bug in NumPy. The solution is to upgrade to NumPy 1.10.6 or later, in which the bug was fixed.

mmap Support for on GNU Hurd

On Hurd and possibly other platforms, flush() on memory-mapped files are not implemented, so writing changes to a mmap’d FITS file may not be reliable and is thus disabled. Attempting to open a FITS file in writeable mode with mmap will result in a warning (and mmap will be disabled on the file automatically).


Bug with Unicode Endianness in io.fits for Big Endian Processors

On big endian processors (e.g. SPARC, PowerPC, MIPS), string columns in FITS files may not be correctly read when using the interface. This will be fixed in a subsequent bug fix release of astropy (see bug report here).

Color Printing on Windows

Colored printing of log messages and other colored text does work in Windows, but only when running in the IPython console. Colors are not currently supported in the basic Python command-line interpreter on Windows.

numpy.int64 does not decompose input Quantity objects

Python’s int() goes through __index__ while numpy.int64 or numpy.int_ do not go through __index__. This means that an upstream fix in numpy` is required in order for ``astropy.units to control decomposing the input in these functions:

>>> np.int64((15 * / (15 * u.imperial.foot))
>>> np.int_((15 * / (15 * u.imperial.foot))
>>> int((15 * / (15 * u.imperial.foot))

To convert a dimensionless Quantity to an integer, it is therefore recommended to use int(...).

Inconsistent behavior when converting complex numbers to floats

Attempting to use float or NumPy’s numpy.float on a standard complex number (e.g., 5 + 6j) results in a TypeError. In contrast, using float or numpy.float on a complex number from NumPy (e.g., numpy.complex128) drops the imaginary component and issues a numpy.ComplexWarning. This inconsistency persists between Quantity instances based on standard and NumPy complex numbers. To get the real part of a complex number, it is recommended to use numpy.real.

Build/Installation/Test Issues

Anaconda Users Should Upgrade with conda, Not pip

Upgrading astropy in the Anaconda Python distribution using pip can result in a corrupted install with a mix of files from the old version and the new version. Anaconda users should update with conda update astropy. There may be a brief delay between the release of astropy on PyPI and its release via the conda package manager; users can check the availability of new versions with conda search astropy.

Locale Errors in MacOS X and Linux

On MacOS X, you may see the following error when running pip:

ValueError: unknown locale: UTF-8

This is due to the LC_CTYPE environment variable being incorrectly set to UTF-8 by default, which is not a valid locale setting.

On MacOS X or Linux (or other platforms) you may also encounter the following error:

  stderr = stderr.decode(stdio_encoding)
TypeError: decode() argument 1 must be str, not None

This also indicates that your locale is not set correctly.

To fix either of these issues, set this environment variable, as well as the LANG and LC_ALL environment variables to e.g. en_US.UTF-8 using, in the case of bash:

export LANG="en_US.UTF-8"
export LC_ALL="en_US.UTF-8"
export LC_CTYPE="en_US.UTF-8"

To avoid any issues in future, you should add this line to your e.g. ~/.bash_profile or .bashrc file.

To test these changes, open a new terminal and type locale, and you should see something like:

$ locale

If so, you can go ahead and try running pip again (in the new terminal).

Failing Logging Tests When Running the Tests in IPython

When running the Astropy tests using astropy.test() in an IPython interpreter, some of the tests in the astropy/tests/ might fail depending on the version of IPython or other factors. This is due to mutually incompatible behaviors in IPython and pytest, and is not due to a problem with the test itself or the feature being tested.


Some Docstrings Can Not Be Displayed in IPython < 0.13.2

Displaying long docstrings that contain Unicode characters may fail on some platforms in the IPython console (prior to IPython version 0.13.2):

In [1]: import astropy.units as u

In [2]: u.Angstrom?
Out[2]: ERROR: UnicodeEncodeError: 'ascii' codec can't encode character u'\xe5' in
position 184: ordinal not in range(128) []

This can be worked around by changing the default encoding to utf-8 by adding the following to your file:

import sys

Note that in general, this is not recommended, because it can hide other Unicode encoding bugs in your application. However, if your application does not deal with text processing and you just want docstrings to work, this may be acceptable.

The IPython issue:

Compatibility Issues with pytest 3.7 and later

Due to a bug in pytest related to test collection, the tests for the core astropy package for version 2.0.x (LTS), and for packages using the core package’s test infrastructure and being tested against 2.0.x (LTS), will not be executed correctly with pytest 3.7, 3.8, or 3.9. The symptom of this bug is that no tests or only tests in RST files are collected. In addition, astropy 2.0.x (LTS) is not compatible with pytest 4.0 and above, as in this case deprecation errors from pytest can cause tests to fail. Therefore, when testing against astropy v2.0.x (LTS), pytest 3.6 or earlier versions should be used. These issues do not occur in version 3.0.x and above of the core package.

There is an unrelated issue that also affects more recent versions of astropy when testing with pytest 4.0 and later, which can cause issues when collecting tests — in this case, the symptom is that the test collection hangs and/or appears to run the tests recursively. If you are maintaining a package that was created using the Astropy package template, then this can be fixed by updating to the latest version of the file. The root cause of this issue is that pytest now tries to pick up the top-level test() function as a test, so we need to make sure that we set a test.__test__ attribute on the function to False.