Use consistent abseil version in TensorFlow

[kpedro88]

The change to the abseil version in #8565 introduced an inconsistency in TensorFlow, leading to errors like this:

/cvmfs/cms.cern.ch/slc7_amd64_gcc11/external/gcc/11.4.1-30ebdc301ebd200f2ae0e3d880258e65/bin/../lib/gcc/x86_64-unknown-linux-gnu/11.4.1/../../../../x86_64-unknown-linux-gnu/bin/ld.bfd: tmp/slc7_amd64_gcc11/src/RecoMET/METPUSubtraction/plugins/RecoMETMETPUSubtraction_plugins/ccjAPFCF.ltrans0.ltrans.o: in function `DeepMETProducer::DeepMETProducer(edm::ParameterSet const&, tensorflow::SessionCache const*)':
<artificial>:(.text+0x6d6c): undefined reference to `tensorflow::TensorShapeBase<tensorflow::TensorShape>::TensorShapeBase(absl::lts_20230125::Span<long const>)'
/cvmfs/cms.cern.ch/slc7_amd64_gcc11/external/gcc/11.4.1-30ebdc301ebd200f2ae0e3d880258e65/bin/../lib/gcc/x86_64-unknown-linux-gnu/11.4.1/../../../../x86_64-unknown-linux-gnu/bin/ld.bfd: <artificial>:(.text+0x6da7): undefined reference to `tensorflow::TensorShapeBase<tensorflow::TensorShape>::TensorShapeBase(absl::lts_20230125::Span<long const>)'
collect2: error: ld returned 1 exit status
gmake: *** [config/SCRAM/GMake/Makefile.rules:1793: tmp/slc7_amd64_gcc11/src/RecoMET/METPUSubtraction/plugins/RecoMETMETPUSubtraction_plugins/libRecoMETMETPUSubtraction_plugins.so] Error 1

The problem as I understand it:

1. The TensorFlow shared libraries are linked to the CMSSW abseil library and have the correct symbols, but the TensorFlow headers have a different abseil namespace (lts_20220623 rather than lts_20230125).
2. This appears to be because TensorFlow still uses its local Bazel workspace information for abseil, even with our bazel-absl.patch file that updates the central Bazel info on this package.

I patched all the places I could find where the Bazel version was hardcoded in the TensorFlow source (based on the upstream abseil version update in TensorFlow 2.13.0, one minor version after ours: tensorflow/tensorflow@ad938db) and was able to remove the DeepMETProducer workaround from cms-sw/cmssw#42228 (see cms-sw/cmssw#42682). I'm happy to reimplement the patches in whatever way is preferred (but it's going to be ugly no matter what).

attn: @iarspider @yongbinfeng

[cmsbuild]

A new Pull Request was created by @kpedro88 (Kevin Pedro) for branch IB/CMSSW_13_3_X/master.

@cmsbuild, @smuzaffar, @aandvalenzuela, @iarspider can you please review it and eventually sign? Thanks.
@perrotta, @dpiparo, @antoniovilela, @rappoccio you are the release manager for this.
cms-bot commands are listed here

[kpedro88]

please test

[iarspider]

@kpedro88 thanks for the fix. Can you open a PR in https://github.com/cms-externals/tensorflow/ targeting cms/v2.12.0 with your changes?

[kpedro88]

Done, see cms-externals/tensorflow#12

[cmsbuild]

+1

Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-6ea053/34534/summary.html
COMMIT: ed0e178
CMSSW: CMSSW_13_3_X_2023-08-29-1100/el8_amd64_gcc11
User test area: For local testing, you can use /cvmfs/cms-ci.cern.ch/week0/cms-sw/cmsdist/8672/34534/install.sh to create a dev area with all the needed externals and cmssw changes.

Comparison Summary

Summary:

• No significant changes to the logs found
• Reco comparison results: 30 differences found in the comparisons
• DQMHistoTests: Total files compared: 48
• DQMHistoTests: Total histograms compared: 3153095
• DQMHistoTests: Total failures: 1493
• DQMHistoTests: Total nulls: 0
• DQMHistoTests: Total successes: 3151580
• DQMHistoTests: Total skipped: 22
• DQMHistoTests: Total Missing objects: 0
• DQMHistoSizes: Histogram memory added: 0.0 KiB( 47 files compared)
• Checked 207 log files, 159 edm output root files, 48 DQM output files
• TriggerResults: no differences found

[smuzaffar]

closing it in favor of #8675