gcs switch to env

[vnghia]

This PR completes gcs implementation by using env from tensorflow/c

Part of #1203 #1183

[vnghia]

@yongtang Could you help me in the Bazel Windows build ? The error is

2021-03-07T16:47:10.1642526Z Execution platform: @local_config_platform//:host
2021-03-07T16:47:10.1644849Z gs.lo.lib(gcs_helper.obj) : error LNK2019: unresolved external symbol __imp_TF_GetTempFileName referenced in function "class std::basic_string<char,struct std::char_traits<char>,class std::allocator<char> > __cdecl GCSGetTempFileName(class std::basic_string<char,struct std::char_traits<char>,class std::allocator<char> > const &)" (?GCSGetTempFileName@@YA?AV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@std@@AEBV12@@Z)
2021-03-07T16:47:10.1650992Z gs.lo.lib(gcs_filesystem.obj) : error LNK2019: unresolved external symbol __imp_TF_NowSeconds referenced in function "class std::unique_ptr<class tensorflow::io::gs::tf_gcs_filesystem::ExpiringLRUCache<struct tensorflow::io::gs::tf_gcs_filesystem::GcsFileSystemStat>,struct std::default_delete<class tensorflow::io::gs::tf_gcs_filesystem::ExpiringLRUCache<struct tensorflow::io::gs::tf_gcs_filesystem::GcsFileSystemStat> > > __cdecl std::make_unique<class tensorflow::io::gs::tf_gcs_filesystem::ExpiringLRUCache<struct tensorflow::io::gs::tf_gcs_filesystem::GcsFileSystemStat>,unsigned __int64 &,unsigned __int64 &,0>(unsigned __int64 &,unsigned __int64 &)" (??$make_unique@V?$ExpiringLRUCache@UGcsFileSystemStat@tf_gcs_filesystem@gs@io@tensorflow@@@tf_gcs_filesystem@gs@io@tensorflow@@AEA_KAEA_K$0A@@std@@YA?AV?$unique_ptr@V?$ExpiringLRUCache@UGcsFileSystemStat@tf_gcs_filesystem@gs@io@tensorflow@@@tf_gcs_filesystem@gs@io@tensorflow@@U?$default_delete@V?$ExpiringLRUCache@UGcsFileSystemStat@tf_gcs_filesystem@gs@io@tensorflow@@@tf_gcs_filesystem@gs@io@tensorflow@@@std@@@0@AEA_K0@Z)
2021-03-07T16:47:10.1658059Z gs.lo.lib(gcs_filesystem.obj) : error LNK2019: unresolved external symbol __imp_TF_DefaultThreadOptions referenced in function "public: __cdecl tensorflow::io::gs::tf_gcs_filesystem::RamFileBlockCache::RamFileBlockCache(unsigned __int64,unsigned __int64,unsigned __int64,class std::function<__int64 __cdecl(class std::basic_string<char,struct std::char_traits<char>,class std::allocator<char> > const &,unsigned __int64,unsigned __int64,char *,struct TF_Status *)>,class std::function<unsigned __int64 __cdecl(void)>)" (??0RamFileBlockCache@tf_gcs_filesystem@gs@io@tensorflow@@QEAA@_K00V?$function@$$A6A_JAEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@std@@_K1PEADPEAUTF_Status@@@Z@std@@V?$function@$$A6A_KXZ@6@@Z)
2021-03-07T16:47:10.1663481Z gs.lo.lib(gcs_filesystem.obj) : error LNK2019: unresolved external symbol __imp_TF_StartThread referenced in function "public: __cdecl tensorflow::io::gs::tf_gcs_filesystem::RamFileBlockCache::RamFileBlockCache(unsigned __int64,unsigned __int64,unsigned __int64,class std::function<__int64 __cdecl(class std::basic_string<char,struct std::char_traits<char>,class std::allocator<char> > const &,unsigned __int64,unsigned __int64,char *,struct TF_Status *)>,class std::function<unsigned __int64 __cdecl(void)>)" (??0RamFileBlockCache@tf_gcs_filesystem@gs@io@tensorflow@@QEAA@_K00V?$function@$$A6A_JAEBV?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@std@@_K1PEADPEAUTF_Status@@@Z@std@@V?$function@$$A6A_KXZ@6@@Z)
2021-03-07T16:47:10.1667309Z gs.lo.lib(gcs_filesystem.obj) : error LNK2019: unresolved external symbol __imp_TF_JoinThread referenced in function "public: void __cdecl <lambda_cec3005b8c03726e011f3e503838b21d>::operator()(struct TF_Thread *)const " (??R<lambda_cec3005b8c03726e011f3e503838b21d>@@QEBAXPEAUTF_Thread@@@Z)
2021-03-07T16:47:10.1669291Z bazel-out\x64_windows-fastbuild\bin\tensorflow_io\core\python\ops\libtensorflow_io_plugins.so : fatal error LNK1120: 5 unresolved externals

It seems that the functions in env are not exported properly on Windows. But I don't have access to a Windows machine right now so I could not do anything. Thank you !

[kvignesh1420]

@vnvo2409 the windows tests are failing due to:

tensorflow.python.framework.errors_impl.AlreadyExistsError: File system for gs already registered

This might be because the TensorFlow team has a separate process of releasing nightly builds for windows.

[vnghia]

@kvignesh1420 Thank you! I haven't known about it before, that might be the reason why this PR failed on the first place. Do you know when the nightly build for Windows will be released ?

[yongtang]

@vnvo2409 @kvignesh1420 from the latest pypi files it looks like latest windows tf-nightly was released just 40 mins ago:
https://pypi.org/project/tf-nightly/2.5.0.dev20210324/#files

I think re-trigger of the PR may allow the test to pass?

[kvignesh1420]

Thanks for the update @yongtang. I have restarted the jobs.
cc: @vnvo2409

[vnghia]

@kvignesh1420 It seems that we did the same thing in exactly the same time 😄

[vnghia]

My commit failed but thanks to @kvignesh1420 , we still have one CI passed https://github.com/tensorflow/io/actions/runs/682804240

I think we could merge this PR @yongtang @kvignesh1420.

[kvignesh1420]

LGTM.

[yongtang]

LGTM, Thanks!