tensorflow: Error on compiling from source

The grpc.patch should be applied in tensorflow/bazel-tensorflow/external/grpc/src after the bazel build started the download.

Or apply it in the .cache (P.S. not working if applied in the cache)

cd ~/.cache/bazel/*/*/external/grpc
patch -p1 < ~/grpc.patch

I’m with @IceCryptonym: clearer directions would be greatly appreciated!

Perhaps the source has changed, but I don’t see any linked patch file or even much mention of patching at the page described by @shantanu-gontia (https://gist.github.com/kmhofmann/e368a2ebba05f807fa1a90b3bf9a1e03). Nor does the Arch-Linux repo mentioned by @Mithrandir2k18 seem to make any mention of gprc, rather it’s all about mkl.

But I followed @shantanu-gontia’s instructions, and have posted a patch file here: https://gist.github.com/drscotthawley/8eb51af1b4c92c4f18432cb045698af7

It can be applied by going to the main grpc directory and running

$ git apply grpc.patch

Whats still not clear is where this should be applied, i.e. where does Bazel put grpc and how I can apply the patch after Bazel puts it there? I see tensorflow/third_party/grpc/ but it’s empty except for a zero-length file called BUILD. There’s also tensorflow/tensorflow/contrib/cmake/patches/grpc/, but that only contains the file rand.h Finally, just putting the patch file in third_party/ doesn’t seem to cause the patch to be applied.

EDIT: Seems Bazel puts it in ~/.cache/bazel/_bazel_($USER)/(big_long_random_directory_name)/external/grpc, but this is not a git directory so git apply won’t work. So…still unclear on how to apply the patch reliably and automatically within the Bazel build.

I have ran into this issue as well and I am not sure how to apply the patch. Would someone be able to give me some incite?

same here

ERROR: /home/user/.cache/bazel/_bazel_userbest/89310458958e3aedfc736fc503a62f52/external/grpc/BUILD:507:1: C++ compilation of rule '@grpc//:gpr_base' failed (Exit 1)
external/grpc/src/core/lib/gpr/log_linux.cc:43:13: error: ambiguating new declaration of 'long int gettid()'
   43 | static long gettid(void) { return syscall(__NR_gettid); }
      |             ^~~~~~
In file included from /usr/include/unistd.h:1170,
                 from external/grpc/src/core/lib/gpr/log_linux.cc:41:
/usr/include/bits/unistd_ext.h:34:16: note: old declaration '__pid_t gettid()'
   34 | extern __pid_t gettid (void) __THROW;
      |                ^~~~~~
external/grpc/src/core/lib/gpr/log_linux.cc:43:13: warning: 'long int gettid()' defined but not used [-Wunused-function]
   43 | static long gettid(void) { return syscall(__NR_gettid); }
      |             ^~~~~~
ERROR: /home/user/.cache/yay/tensorflow-git/src/tensorflow/tensorflow/python/tools/BUILD:312:1 C++ compilation of rule '@grpc//:gpr_base' failed (Exit 1)
INFO: Elapsed time: 1012.293s, Critical Path: 30.58s
INFO: 3168 processes: 3168 local.
FAILED: Build did NOT complete successfully
==> ERROR: A failure occurred in build().
    Aborting...
Error making: tensorflow-git

Arch Linux

The error seems to be due to the same symbol definitions (of gettid) introduced in GCC 9.0+ extant in grpc. https://github.com/grpc/grpc/issues/20043

The folks at grpc have patched the problem in newer releases. https://github.com/grpc/grpc/pull/20048

However, during the build process, TensorFlow downloads a cached version of an older commit of grpc from the TensorFlow mirror. At this point, I do not know which commit from the grpc repository fixes the issue in particular (I tried the latest one, it has some problem with upb_proto_library). Therefore, the best course of action, as I found from other sources (https://gist.github.com/kmhofmann/e368a2ebba05f807fa1a90b3bf9a1e03) is to patch the grpc commit downloaded using a patch file.

If the patch in https://gist.github.com/kmhofmann/e368a2ebba05f807fa1a90b3bf9a1e03 doesn’t work I would suggest cloning the grpc source from github, checkout the commit 4566c2a29ebec0835643b972eb99f4306c4234a3. Edit the following files -

• src/core/lib/gpr/log_linux.cc
• src/core/lib/gpr/log_posix.cc
• src/core/lib/iomgr/ev_epollex_linux.cc

Just change every instance of the term gettid to sys_gettid in these three files and generate a patch for yourself using git diff and use that patch.