You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I'm beginning to think we should back out all the "just pretend grpc works" stuff we've done (make it a pure extra), disabled by default. I also want to shoot the _testing directory in the head.
I think unit tests don't really need it, either: the bigtable tests could just use `unittest2.skipUnless' like the GAX-related tests in logging and pubsub.
At the time it was added, there was no Travis environment where grpcio installed, so that was the only way. I'm fine with your suggestion given the current state of things.
Installed c:\users\appveyor\appdata\local\temp\1\easy_install-idvc8i\grpcio-1.0.0rc1\.eggs\markupsafe-0.23-py2.7-win32.egg
warning: no files found matching '*.h' under directory 'src\python\grpcio\grpc'
warning: no files found matching '*.python' under directory 'src\python\grpcio\grpc'
warning: no previously-included files matching '*.so' found under directory 'src\python\grpcio\grpc\_cython'
warning: no previously-included files matching '*.pyd' found under directory 'src\python\grpcio\grpc\_cython'
warning: no files found matching 'src\python\grpcio\precompiled.py'
cygrpc.c
include\grpc/impl/codegen/compression_types.h(38) : fatal error C1083: Cannot open include file: 'stdbool.h': No such file or directory
a.c
I think the issue is that pip install grpcio won't grab he RC version so it's trying a pre-1.0 release. Either that or your wheel doesn't support Windows?
Note that here you'll need to use the --pre flag to get 1.0.0rc1 unless you want to be grabbing it during nosetests as part of the command's potentially problematic gathering of (constrained to egg or .tar.gz?) dependencies (AFAICT).
include\grpc/impl/codegen/compression_types.h(38) : fatal error C1083: Cannot open include file: 'stdbool.h': No such file or directory
creating Users
creating Users\appveyor
while both versions of 3.5 fail with an OpenSSL issue
IIRC grpcio can be built within a MinGW 64-bit Windows environment with vanilla Python 3.4 (NOT e.g. MSYS2's) if configured to use the GCC toolchain via e.g. -c mingw32. grpcio-tools will fail with linker errors that might be alleviated by compiling with TDM-GCC (haven't tried that yet) or finishing up C++98-ifying the plug-in code enough for MSVC 2010 to handle it.
It looks like y'all just need grpcio, not grpcio-tools, though. Is that the case?
EDIT: Turns out I recalled incorrectly and that grpcio also fails with linker errors.
@soltanmm Correct: users of gcloud-python only need grpcio: developers may need grpcio-tools for regenerating proto-based files, but we don't have any Windows-based developers, and we don't do that task via Windows CI.
Activity
tseaver commentedon Jul 20, 2016
@dhermes Looks like they need to dump the PYTHONPATH=_testing bit.
dhermes commentedon Jul 20, 2016
But
grpciosuccessfully installs in the Python 3 environment otherwise?UPDATE: Or you're saying that the GAX version wouldn't be used if the import failed?
tseaver commentedon Jul 20, 2016
I've run the Py3k system tests with
grpcioinstalled (and thegax-*libraries), but suppressing having_testingon the PYTHONPATH.dhermes commentedon Jul 20, 2016
On my machine
pip3 install grpciofails withI could put
grpc_python_pluginon my path but that won't help Travis.tseaver commentedon Jul 20, 2016
I'm beginning to think we should back out all the "just pretend grpc works" stuff we've done (make it a pure extra), disabled by default. I also want to shoot the
_testingdirectory in the head.dhermes commentedon Jul 20, 2016
The
_testingdirectory was just there for unit tests (so imports didn't break), never intended to be used in system tests. The real culprit isgrpcio.tseaver commentedon Jul 20, 2016
I think unit tests don't really need it, either: the bigtable tests could just use `unittest2.skipUnless' like the GAX-related tests in logging and pubsub.
dhermes commentedon Jul 20, 2016
At the time it was added, there was no Travis environment where
grpcioinstalled, so that was the only way. I'm fine with your suggestion given the current state of things.tseaver commentedon Jul 21, 2016
W00t! https://pypi.python.org/pypi/grpcio/1.0.0rc1/#downloads has wheels for Python 3.4 on Linux/Mac/Windows!
dhermes commentedon Jul 21, 2016
I'm surprised they're at / near a 1.0
tseaver commentedon Jul 21, 2016
I wonder if pinning the RC would unbreak Travis
system-tests3and Appveyor.nathanielmanistaatgoogle commentedon Jul 22, 2016
Apologies for the gap in 3.4-on-Windows.
daspecster commentedon Jul 22, 2016
@tseaver https://ci.appveyor.com/project/daspecster/gcloud-python-utjkc/build/job/ibd6g2k1a4gkumjo
soltanmm commentedon Jul 22, 2016
@daspecster Looks like Appveyor is trying to build from source on Windows for Python 2.7.
It should be downloading the wheels...
dhermes commentedon Jul 22, 2016
I think the issue is that
pip install grpciowon't grab he RC version so it's trying a pre-1.0 release. Either that or your wheel doesn't support Windows?soltanmm commentedon Jul 22, 2016
Note that here you'll need to use the
--preflag to get1.0.0rc1unless you want to be grabbing it duringnosetestsas part of the command's potentially problematic gathering of (constrained to egg or.tar.gz?) dependencies (AFAICT).daspecster commentedon Jul 22, 2016
Thanks @soltanmm, I actually just forced the version with
pip install grpcio==1.0.0rc1for now to try it in my branch.https://ci.appveyor.com/project/daspecster/gcloud-python-pjqjn
daspecster commentedon Jul 22, 2016
Some success and some more failures :(
https://ci.appveyor.com/project/daspecster/gcloud-python-pjqjn/build/job/1gpl7vqkrafyvjcu
dhermes commentedon Jul 25, 2016
64-bit 3.4 fails with
while both versions of 3.5 fail with an OpenSSL issue
soltanmm commentedon Jul 26, 2016
IIRC
grpciocan be built within a MinGW 64-bit Windows environment with vanilla Python 3.4 (NOT e.g. MSYS2's) if configured to use the GCC toolchain via e.g.-c mingw32.grpcio-toolswill fail with linker errors that might be alleviated by compiling with TDM-GCC (haven't tried that yet) or finishing up C++98-ifying the plug-in code enough for MSVC 2010 to handle it.It looks like y'all just need
grpcio, notgrpcio-tools, though. Is that the case?EDIT: Turns out I recalled incorrectly and that
grpcioalso fails with linker errors.tseaver commentedon Jul 26, 2016
@soltanmm Correct: users of
gcloud-pythononly needgrpcio: developers may needgrpcio-toolsfor regenerating proto-based files, but we don't have any Windows-based developers, and we don't do that task via Windows CI.tseaver commentedon Aug 25, 2016
Via #2065, #2085.
refactor: Remove duplicated comments in _function_session.py (#2001)