libmozilla_voice_stt.so linked against macOS 10.15 instead of 10.10

[Arpitrf]

Hello,

I am getting the following error when running the below command.

`(deepspeechenv) 192:deepspeech apple$ deepspeech --model deepspeech-0.7.0-models.pbmm --scorer deepspeech-0.7.0-models.scorer --audio audio/2830-3980-0043.wav

Traceback (most recent call last):
File "/Users/apple/Desktop/Citi/FRM/deepspeech/deepspeechenv/bin/deepspeech", line 5, in
from deepspeech.client import main
File "/Users/apple/Desktop/Citi/FRM/deepspeech/deepspeechenv/lib/python3.6/site-packages/deepspeech/init.py", line 23, in
from deepspeech.impl import Version as version
File "/Users/apple/Desktop/Citi/FRM/deepspeech/deepspeechenv/lib/python3.6/site-packages/deepspeech/impl.py", line 13, in
from . import _impl
ImportError: dlopen(/Users/apple/Desktop/Citi/FRM/deepspeech/deepspeechenv/lib/python3.6/site-packages/deepspeech/_impl.cpython-36m-darwin.so, 2): Symbol not found: ____chkstk_darwin
Referenced from: /Users/apple/Desktop/Citi/FRM/deepspeech/deepspeechenv/lib/python3.6/site-packages/deepspeech/lib/libdeepspeech.so (which was built for Mac OS X 10.15)
Expected in: /usr/lib/libSystem.B.dylib
in /Users/apple/Desktop/Citi/FRM/deepspeech/deepspeechenv/lib/python3.6/site-packages/deepspeech/lib/libdeepspeech.so`

[lissyx]

Please document what version that is, because we force targetting à lower version...

[lissyx]

Here, 10.10: https://github.com/mozilla/STT/blob/master/taskcluster/darwin-opt-base.tyml#L52

So I don't know why it would require 10.15 ...

[Arpitrf]

deepspeech version: 0.8.1
Python version: Python 3.6.5
pip version: pip 20.2.2

[Arpitrf]

Ok I found a work-around.

pip uninstall deepspeech
pip install deepspeech==0.7.0

Version 0.7.0 works.

Is deepseech version 0.8.1 incompatible with macOS?

[lissyx]

Is deepseech version 0.8.1 incompatible with macOS?

Did you read my replies? It should not.

[lissyx]

@Arpitrf Please, can you check with native_client.tar.xz, to check if it is libdeepspeech.so build or just the Python wrapper.

[Arpitrf]

I am sorry I did not understand the above.

[lissyx]

https://github.com/mozilla/STT/releases/download/v0.8.1/native_client.amd64.cpu.osx.tar.xz

[Arpitrf]

This installed this - native_client.amd64.cpu.osx.tar.xz. What to do next?

[lissyx]

This installed this - native_client.amd64.cpu.osx.tar.xz. What to do next?

Extract and run ./deepspeech --version to see if you have the same problem?

[Arpitrf]

Extract and run ./deepspeech --version to see if you have the same problem?

Output: DeepSpeech 0.8.1

[lissyx]

Extract and run ./deepspeech --version to see if you have the same problem?

Output: DeepSpeech 0.8.1

Can you try to run a model then?

./deepspeech --model deepspeech-0.7.0-models.pbmm --scorer deepspeech-0.7.0-models.scorer --audio audio/2830-3980-0043.wav

[lissyx]

@Arpitrf Please?

[Arpitrf]

Still does not work.

**(deepspeechenv) 192:deepspeech apple$** ./deepspeech --model deepspeech-0.7.0-models.pbmm --scorer deepspeech-0.7.0-models.scorer --audio audio/2830-3980-0043.wav

TensorFlow: v2.2.0-24-g1c1b2b9dd9
DeepSpeech: v0.8.1-0-gfa883eb8
2020-08-18 17:48:29.840025: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
dyld: lazy symbol binding failed: Symbol not found: ____chkstk_darwin
  Referenced from: /Users/apple/Desktop/Citi/FRM/deepspeech/libdeepspeech.so (which was built for Mac OS X 10.15)
  Expected in: /usr/lib/libSystem.B.dylib

dyld: Symbol not found: ____chkstk_darwin
  Referenced from: /Users/apple/Desktop/Citi/FRM/deepspeech/libdeepspeech.so (which was built for Mac OS X 10.15)
  Expected in: /usr/lib/libSystem.B.dylib

Abort trap: 6

[lissyx]

dyld: lazy symbol binding failed: Symbol not found: ____chkstk_darwin
Referenced from: /Users/apple/Desktop/Citi/FRM/deepspeech/libdeepspeech.so (which was built for Mac OS X 10.15)

Thanks, so I fear this is a regression from using XCode for iOS support :/ cc @reuben

There's an upstream Bazel bug matching that, where Bazel does not respect minimum sdk we force, but it was only fixed in 3.4 and TensorFlow r2.2 used for r0.8 uses Bazel 2.0.0, and I don't see any workaround so far.

We just upgraded master to r2.3, but:

• it is using Bazel 3.1, not 3.4
• if it fixes, it requires complete rebuild of TensorFlow and our macOS CI is not really in good shape.

I'm unsure if we can fix that soonish.

[lissyx]

Also, changing Bazel version is kind of invasive, I'm not sure we want to do that on 0.8 ...

[Arpitrf]

Thanks for your help!

[lissyx]

Turns out @reuben might have found a proper fix that is not invasive as we first tought, but it will require some full rebuild so it may take a few days. Hopefully we might do 0.8.2 to ship it.

[lissyx]

Fixed in master: #3259

[lissyx]

And r0.8: #3260