我正在尝试使用 kivy 创建一个使用 astropy 的 android 应用程序。困难在于 astropy 在安装过程中使用了 numpy,而我无法让它成功加载 numpy 库。我认为问题在于它正在寻找为 ARM 架构编译的库 - 我怎样才能让它找到正确的库以在构建时使用?
我使用的是kivy提供的linux虚拟机。 numpy 配方本身工作正常。对于 astropy,我创建了一个 recipe.sh 脚本,除了标准设置外,还包含以下构建函数:
function build_astropy() {
cd $BUILD_astropy
export BUILDLIB_PATH="$BUILD_hostpython/build/lib.linux-`uname -m`-2.7/"
export PYTHONPATH=$SITEPACKAGES_PATH:$BUILDLIB_PATH
push_arm
try $HOSTPYTHON setup.py install
pop_arm
}
BUILDLIB_PATH 行允许它成功找到并导入 _io.so 模块 - 我从另一个处理类似架构问题的配方中复制了它。但是当我尝试运行分发脚本时:
./distribute.sh -f -m "astropy kivy" -d astropy
它尝试导入 numpy,但出现此错误:
ImportError: /home/kivy/android/python-for-android/build/python-install/lib/python2.7/site-packages/numpy/core/multiarray.so: cannot open shared object file: No such file or directory
该文件确实存在,但可能是为 ARM 编译的。我尝试添加本地 numpy 安装:
/usr/lib/python2.7/dist-packages/numpy/core/
到 BUILDLIB_PATH 但它没有任何区别。有没有人有任何提示可以让 kivy 在构建时找到合适的库?
请您参考如下方法:
失败导入的回溯包含 get_numpy_include_path(),它来自 astropy_helpers/astropy_helpers/setup_helpers.py。查看 numpy 的 get_include() 的源代码,它在交叉编译时无法在系统上运行,因此即使您设法为主机 Python 构建 numpy,它也会得到不正确的值。
修补 astropy 以允许通过环境提供 numpy 包含路径(在哪里可以找到 numpyconfig.h 等文件)将解决此问题,因此创建 recipes/patches/add_numpy_include。包含以下内容的补丁:
--- astropy-0.4.4.orig/astropy_helpers/astropy_helpers/setup_helpers.py 2015-02-15 18:16:02.135642283 -0600
+++ astropy-0.4.4/astropy_helpers/astropy_helpers/setup_helpers.py 2015-02-15 18:15:10.985674250 -0600
@@ -1307,6 +1307,11 @@
# install, since Numpy may still think it's in "setup mode", when
# in fact we're ready to use it to build astropy now.
+ # Allow the environment to override the include path, to allow
+ # cross-compilation
+ if 'NUMPY_INCLUDE_PATH' in os.environ:
+ return os.environ['NUMPY_INCLUDE_PATH']
+
if sys.version_info[0] >= 3:
import builtins
if hasattr(builtins, '__NUMPY_SETUP__'):
如果您尝试仅使用此补丁进行构建,则会导致有关 struct lconv 的构建错误,因此请创建第二个补丁文件,recipes/patches/lconv_fix.patch包含以下内容:
--- astropy-0.4.4.orig/cextern/wcslib/C/wcsutil.c 2015-02-15 18:07:46.549935299 -0600
+++ astropy-0.4.4/cextern/wcslib/C/wcsutil.c 2015-02-15 18:36:07.062452345 -0600
@@ -247,7 +247,7 @@
{
struct lconv *locale_data = localeconv();
- const char *decimal_point = locale_data->decimal_point;
+ const char *decimal_point = ".";
if (decimal_point[0] != '.' || decimal_point[1] != 0) {
size_t decimal_point_len = strlen(decimal_point);
@@ -311,7 +311,7 @@
{
struct lconv *locale_data = localeconv();
- const char *decimal_point = locale_data->decimal_point;
+ const char *decimal_point = ".";
if (decimal_point[0] != '.' || decimal_point[1] != 0) {
char *out = outbuf;
--- astropy-0.4.4.orig/cextern/cfitsio/fitscore.c 2015-02-15 18:07:46.553935299 -0600
+++ astropy-0.4.4/cextern/cfitsio/fitscore.c 2015-02-15 18:41:16.626440539 -0600
@@ -9226,7 +9226,7 @@
if (!decimalpt) { /* only do this once for efficiency */
lcc = localeconv(); /* set structure containing local decimal point symbol */
- decimalpt = *(lcc->decimal_point);
+ decimalpt = '.';
}
errno = 0;
@@ -9296,7 +9296,7 @@
if (!decimalpt) { /* only do this once for efficiency */
lcc = localeconv(); /* set structure containing local decimal point symbol */
- decimalpt = *(lcc->decimal_point);
+ decimalpt = '.';
}
errno = 0;
lconv 修复并不是一个很好的修复,因为它假定 Android 设备上的区域设置有“.”。作为小数点分隔符,但是 patch for numpy在官方存储库中这样做,因此 astropy 将在与 numpy 相同的情况下失败。
根据 official repository 使用 numpy 1.7.1 和配方,以下 prebuild_astropy() 和 build_astropy() 函数适用于 astropy 0.4.4,调用 ./distribute.sh -m "numpy astropy kivy"-d天文:
function prebuild_astropy() {
cd $BUILD_astropy
if [ -f .patched ]; then
return
fi
try patch -p1 < $RECIPE_astropy/patches/add_numpy_include.patch
try patch -p1 < $RECIPE_astropy/patches/lconv_fix.patch
touch .patched
true
}
function build_astropy() {
cd $BUILD_astropy
export BUILDLIB_PATH="$BUILD_hostpython/build/lib.linux-`uname -m`-2.7/"
export PYTHONPATH=$SITEPACKAGES_PATH:$BUILDLIB_PATH
export NUMPY_INCLUDE_PATH="$BUILD_PATH/python-install/lib/python2.7/site-packages/numpy/core/include"
push_arm
try $BUILD_PATH/python-install/bin/python.host setup.py install
pop_arm
unset BUILDLIB_PATH
unset PYTHONPATH
unset NUMPY_INCLUDE_PATH
}
