nixpkgs/pkgs/development/python-modules/pandas/default.nix
2020-08-29 10:23:39 -07:00

142 lines
3.8 KiB
Nix

{ buildPythonPackage
, fetchPypi
, python
, stdenv
, pytest
, glibcLocales
, cython
, dateutil
, scipy
, moto
, numexpr
, pytz
, xlrd
, bottleneck
, sqlalchemy
, lxml
, html5lib
, beautifulsoup4
, hypothesis
, openpyxl
, tables
, xlwt
, runtimeShell
, isPy38
, libcxx ? null
}:
let
inherit (stdenv.lib) optional optionals optionalString;
inherit (stdenv) isDarwin;
in buildPythonPackage rec {
pname = "pandas";
version = "1.1.1";
src = fetchPypi {
inherit pname version;
sha256 = "53328284a7bb046e2e885fd1b8c078bd896d7fc4575b915d4936f54984a2ba67";
};
checkInputs = [ pytest glibcLocales moto hypothesis ];
nativeBuildInputs = [ cython ];
buildInputs = optional isDarwin libcxx;
propagatedBuildInputs = [
dateutil
scipy
numexpr
pytz
xlrd
bottleneck
sqlalchemy
lxml
html5lib
beautifulsoup4
openpyxl
tables
xlwt
];
# doesn't work with -Werror,-Wunused-command-line-argument
# https://github.com/NixOS/nixpkgs/issues/39687
hardeningDisable = optional stdenv.cc.isClang "strictoverflow";
# For OSX, we need to add a dependency on libcxx, which provides
# `complex.h` and other libraries that pandas depends on to build.
postPatch = optionalString isDarwin ''
cpp_sdk="${libcxx}/include/c++/v1";
echo "Adding $cpp_sdk to the setup.py common_include variable"
substituteInPlace setup.py \
--replace "['pandas/src/klib', 'pandas/src']" \
"['pandas/src/klib', 'pandas/src', '$cpp_sdk']"
'';
# Parallel Cythonization is broken in Python 3.8 on Darwin. Fixed in the next
# release. https://github.com/pandas-dev/pandas/pull/30862
setupPyBuildFlags = optionals (!(isPy38 && isDarwin)) [
# As suggested by
# https://pandas.pydata.org/pandas-docs/stable/development/contributing.html#creating-a-python-environment
"--parallel=$NIX_BUILD_CORES"
];
disabledTests = stdenv.lib.concatMapStringsSep " and " (s: "not " + s) ([
# since dateutil 0.6.0 the following fails: test_fallback_plural, test_ambiguous_flags, test_ambiguous_compat
# was supposed to be solved by https://github.com/dateutil/dateutil/issues/321, but is not the case
"test_fallback_plural"
"test_ambiguous_flags"
"test_ambiguous_compat"
# Locale-related
"test_names"
"test_dt_accessor_datetime_name_accessors"
"test_datetime_name_accessors"
# Can't import from test folder
"test_oo_optimizable"
# Disable IO related tests because IO data is no longer distributed
"io"
# KeyError Timestamp
"test_to_excel"
# ordering logic has changed
"numpy_ufuncs_other"
"order_without_freq"
# tries to import from pandas.tests post install
"util_in_top_level"
# Fails with 1.0.5
"test_constructor_list_frames"
"test_constructor_with_embedded_frames"
] ++ optionals isDarwin [
"test_locale"
"test_clipboard"
]);
doCheck = !stdenv.isAarch64; # upstream doesn't test this architecture
checkPhase = ''
runHook preCheck
''
# TODO: Get locale and clipboard support working on darwin.
# Until then we disable the tests.
+ optionalString isDarwin ''
# Fake the impure dependencies pbpaste and pbcopy
echo "#!${runtimeShell}" > pbcopy
echo "#!${runtimeShell}" > pbpaste
chmod a+x pbcopy pbpaste
export PATH=$(pwd):$PATH
'' + ''
LC_ALL="en_US.UTF-8" py.test $out/${python.sitePackages}/pandas --skip-slow --skip-network -k "$disabledTests"
runHook postCheck
'';
meta = {
# https://github.com/pandas-dev/pandas/issues/14866
# pandas devs are no longer testing i686 so safer to assume it's broken
broken = stdenv.isi686;
homepage = "https://pandas.pydata.org/";
description = "Python Data Analysis Library";
license = stdenv.lib.licenses.bsd3;
maintainers = with stdenv.lib.maintainers; [ raskin fridh knedlsepp ];
platforms = stdenv.lib.platforms.unix;
};
}