Many scancode dependencies are broken for Mac M1 #3205

rnjudge · 2023-01-17T20:26:19Z

Description

This is not necessarily a bug with Scancode, but a handful of the required dependencies are broken and cannot install on Mac M1 platform. This is a huge blocker for anyone using this platform as it is impossible to install scancode using pip. Perhaps the scancode docs should be updated to note that scancode is not usable on this platform?

How To Reproduce

Try to install scancode on Mac M1 computer. In my case, I am running Ubunutu VM using fusion but the same issues arise when I try to install scancode directly on my mac shell (not in a fusion VM).

(ternenv) rose@rose-vm:~/ternenv/tern$ pip install --upgrade pip setuptools wheel
(ternenv) rose@rose-vm:~/ternenv/tern$ pip install scancode-toolkit

Specific libraries with issues installing:
bitarray intbitset lxml pyahocorasick pycryptodome cffi

System configuration

What OS are you running on?
MacOS Monteray 12.2.1 (M1 chip)
What version of scancode-toolkit was used to generate the scan file?
Attempting to install scancode_toolkit-31.2.4-cp310-none-any.whl
What installation method was used to install/run scancode? (pip/source download/other)
pip

The text was updated successfully, but these errors were encountered:

rnjudge · 2023-01-17T21:16:49Z

Additionally, trying to install the [full] scancode option yields:

]ERROR: Cannot install scancode-toolkit[full]==3.1.1, scancode-toolkit[full]==3.2.0 and scancode-toolkit[full]==3.2.3 because these package versions have conflicting dependencies.

The conflict is caused by:
    scancode-toolkit[full] 3.2.3 depends on typecode-libmagic; extra == "full"
    scancode-toolkit[full] 3.2.0 depends on extractcode-7z
    scancode-toolkit[full] 3.1.1 depends on extractcode-7z

rnjudge · 2023-01-17T21:20:32Z

At least one of the issues I am seeing is being tracked here: #3169.

pombredanne · 2023-01-18T08:28:55Z

@rnjudge Thanks for the report! do you know if there is a free (as in beer) hosted CI service that runs Apple ARM silicon on macOS?

Jeeppler · 2023-01-18T09:58:22Z

Is there a big difference between the Apple ARM M1 chip and a normal ARM based chip?

Meaning, could it be enough to make everything work on a ARM chip (e. g. Raspberry PI or AWS A1 instance). And once it is working one could test it manually on a M1 chip.

pombredanne · 2023-01-18T10:14:22Z

@Jeeppler re:

Is there a big difference between the Apple ARM M1 chip and a normal ARM based chip?

I have no idea... but I would doubt that macOS can be installed (legally) on non-Apple hardware, so the issue may be part ARM and part macOS?
We could likely --with a bit a sweat--cross compile for ARM on X86, but I do not know if we can get a binary that would run on mac....

Though, I recall now that I released macOS fat binary wheels for https://github.com/inveniosoftware/intbitset/ and https://github.com/WojciechMula/pyahocorasick in the last few days.

Could you test if these pre-built binaries work on your machine with a pip install?

Jeeppler · 2023-01-18T17:31:13Z

@pombredanne I have no mac. I was just curious if the solution could be to compile the dependencies on Linux ARM and then try to run it on macOS.

@rnjudge has a macOS with M1 chip, as far as I understand. Only she can help you test.

rnjudge · 2023-01-20T22:42:44Z

Though, I recall now that I released macOS fat binary wheels for https://github.com/inveniosoftware/intbitset/ and https://github.com/WojciechMula/pyahocorasick in the last few days.

Could you test if these pre-built binaries work on your machine with a pip install?

I installed the latest pyahocorasick and intbitset libraries with pip but still can't install scancode because building the wheel for intbitset fails when I run pip install scancode-toolkit. What's strange is that when I install intbitset on its own, the wheel is created properly:

(ternenv) rose@rose-vm:~/ternenv/scancode-toolkit$ pip3 install intbitset
Collecting intbitset
  Using cached intbitset-3.0.2.tar.gz (152 kB)
  Preparing metadata (setup.py) ... done
Building wheels for collected packages: intbitset
  Building wheel for intbitset (setup.py) ... done
  Created wheel for intbitset: filename=intbitset-3.0.2-cp310-cp310-linux_aarch64.whl size=322885 sha256=b70b76050f66ed1a11940a5ce4c4ddf3592c10c476bc971c364ce587b851eb5b
  Stored in directory: /home/rose/.cache/pip/wheels/bc/f1/4c/1691139855e3477383cd895bdf4a69875d0ae83119ea20d38f
Successfully built intbitset
Installing collected packages: intbitset
Successfully installed intbitset-3.0.2

      running build_ext
      building 'intbitset' extension
      creating build/temp.linux-aarch64-cpython-310
      creating build/temp.linux-aarch64-cpython-310/intbitset
      aarch64-linux-gnu-gcc -Wno-unused-result -Wsign-compare -DNDEBUG -g -fwrapv -O2 -Wall -g -fstack-protector-strong -Wformat -Werror=format-security -g -fwrapv -O2 -fPIC -I/home/rose/ternenv/include -I/usr/include/python3.10 -c intbitset/intbitset.c -o build/temp.linux-aarch64-cpython-310/intbitset/intbitset.o -O3 -march=core2 -mtune=native
      intbitset/intbitset.c: In function ‘__pyx_pf_9intbitset_9intbitset___cinit__’:
      intbitset/intbitset.c:2267:13: warning: ‘PyObject_AsReadBuffer’ is deprecated [-Wdeprecated-declarations]
       2267 |             __pyx_t_4 = ((PyObject_AsReadBuffer(__pyx_v_tmp, (&__pyx_v_buf), (&__pyx_v_size)) < 0) != 0);
            |             ^~~~~~~~~
      In file included from /usr/include/python3.10/genobject.h:12,
                       from /usr/include/python3.10/Python.h:110,
                       from intbitset/intbitset.c:4:
      /usr/include/python3.10/abstract.h:343:17: note: declared here
        343 | PyAPI_FUNC(int) PyObject_AsReadBuffer(PyObject *obj,
            |                 ^~~~~~~~~~~~~~~~~~~~~
      intbitset/intbitset.c: In function ‘__pyx_f_9intbitset_9intbitset_fastload’:
      intbitset/intbitset.c:9543:7: warning: ‘PyObject_AsReadBuffer’ is deprecated [-Wdeprecated-declarations]
       9543 |       __pyx_t_9 = ((PyObject_AsReadBuffer(__pyx_v_tmp, (&__pyx_v_buf), (&__pyx_v_size)) < 0) != 0);
            |       ^~~~~~~~~
      /usr/include/python3.10/abstract.h:343:17: note: declared here
        343 | PyAPI_FUNC(int) PyObject_AsReadBuffer(PyObject *obj,
            |                 ^~~~~~~~~~~~~~~~~~~~~
      Assembler messages:
      Error: unknown architecture `core2'
      
      Error: unrecognized option -march=core2
      error: command '/usr/bin/aarch64-linux-gnu-gcc' failed with exit code 1
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
  Rolling back uninstall of intbitset
  Moving to /home/rose/ternenv/lib/python3.10/site-packages/__pycache__/intbitset_helper.cpython-310.pyc
   from /tmp/pip-uninstall-u9t5ci_4/intbitset_helper.cpython-310.pyc
  Moving to /home/rose/ternenv/lib/python3.10/site-packages/__pycache__/intbitset_version.cpython-310.pyc
   from /tmp/pip-uninstall-u9t5ci_4/intbitset_version.cpython-310.pyc
  Moving to /home/rose/ternenv/lib/python3.10/site-packages/intbitset-3.0.2.dist-info/
   from /home/rose/ternenv/lib/python3.10/site-packages/~ntbitset-3.0.2.dist-info
  Moving to /home/rose/ternenv/lib/python3.10/site-packages/intbitset.cpython-310-aarch64-linux-gnu.so
   from /tmp/pip-uninstall-l9czslo5/intbitset.cpython-310-aarch64-linux-gnu.so
  Moving to /home/rose/ternenv/lib/python3.10/site-packages/intbitset_helper.py
   from /tmp/pip-uninstall-l9czslo5/intbitset_helper.py
  Moving to /home/rose/ternenv/lib/python3.10/site-packages/intbitset_version.py
   from /tmp/pip-uninstall-l9czslo5/intbitset_version.py
error: legacy-install-failure

× Encountered error while trying to install package.
╰─> intbitset

pombredanne · 2023-01-21T08:38:18Z

@rnjudge I had NOT paid attention to all the details you brought... you are not running a macOS with an ARM chip ;) .... you are running Linux on ARM, right?

In this case this totally untested and I have no pre-built wheels for pyahocorasick or intbitset (and for also several other more common libs) but I see lxml as aarch builds.

Can you try a pip install --upgrade lxml and tell me if the wheel is fetched as-is or if it built from sources?

rnjudge · 2023-01-23T23:31:25Z

Apologies for not being more clear, @pombredanne! As I noted in my original issue, I do see the same install issues in both environments (M1 shell vs Ubunutu shell on M1 via Fusion) but I am most interested in getting this working for my Linux environment given that is where I do all my development for Tern.

I ran the pip upgrade command for lxml but I already had the latest version installed so it didn't try to fetch or build from source. I uninstalled lxml and re-installed it to see what it would pull and it fetched and installed the wheel as is:

(ternenv) rose@rose-vm:~/ternenv$ pip install lxml --no-cache-dir
Collecting lxml
  Downloading lxml-4.9.2-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_24_aarch64.whl (6.8 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 6.8/6.8 MB 25.4 MB/s eta 0:00:00
Installing collected packages: lxml
Successfully installed lxml-4.9.2

stevespringett · 2023-02-23T23:47:39Z

@pombredanne, my experience on an M1 (both native macOS and Ubuntu w/Parallels) is exactly the same as what @rnjudge experiences. Even have the same issue with intbitset.

rnjudge · 2023-05-19T20:52:28Z

@pombredanne any update on this? This is a large blocker from my POV and one that is making me rethink Tern's integration with Scancode given that there is an increasing developer community (myself included) that still cannot install Scancode :/

pombredanne · 2023-05-21T21:58:42Z

@rnjudge We support running ScanCode toolkit on macOS M1 using X86 roseta emulation alright as documented in the installation instructions.

If you have a properly installed toolchain you should be able otherwise to compile from sources intbitset and pyahocorasick and there are ways to create plugins and configure typecode and extractcode to use OS-provided binaries for these natives. This is rather involved and requires some time to spend.

You can also try to use scancode-toolkit-mini for a pip installation that has fewer dependencies (and also fewer features).

So the net is that this is not a trivial thing to do AND things are otherwise working fine using the documented roseta/X86 emulation installation when using an application archive.

To "support natively macOS on M1 and Linux on ARM" -- we would need all these covered:

- We need someone to step up to sponsor and financially support a paid Ci/CD that provides ARM on Linux and ARM on macOS. Of note, CircleCI announced support for Apple ARM a few weeks ago: https://circleci.com/blog/m1-mac-resource-class/ in their paid plans. They seem to also have ARM support for Linux.
- We need to pre-build and release all the native deps, including possibly a few other than are not under our control, forking these as needed. And this each time there are updates.
- We need to adjust the Ci/CD and release scripts to also build and test on ARM for macOS and Linux and test all this to ensure this works correctly and amend the tests if needed.
- Whoever steps in to help -- by either doing or funding -- needs to be committing for the long term as this is not just a one time effort.

A rough estimate is that this is going to take 15 to 20 days of initial setup work and about 1 day per month afterwards to maintain and support this. And the Ci/CD should to be in the $50/month range.

Since this requires quite a bit of work and demands using extra paid resources for a CI/CD, are you willing to help there in cash or in kind? I cannot see how this can happen without funding and help.

@heliocastro I know you are interested too and started some work there, @stevespringett you too? so may be there is a way to pool efforts to fund this?

pombredanne · 2023-05-21T22:02:59Z

@soimkim ping since you also raised an issue in (Installation) Mac M1. python3.8 > Error : Building wheel for intbitset (setup.py) ... error #3169
@rohitcoder ping since you raised scancode installed via pip command not working on Mac M1 #3166
@JoeSzymanski ping for Unable to configure scancode on a macOS jenkins server #3108
@bwjohnson-ss ping for Unable to configure scancode on a macOS jenkins server #3108
@enricozb ping for Unable to configure scancode on a macOS jenkins server #3108

pombredanne · 2023-05-22T17:11:34Z

@rnjudge note also that we have a GSoC student selected to help in this area with a project to provide a few pure python, slower alternative for three core libraries: intbitset, pyahocorasick and lxml. This will run over the summer and is only a partial solution.

enricozb · 2023-05-22T17:21:27Z

responding to ping. I can't contribute. I was just trying to help solve that specific wheel issue, I'm not doing anything on M1 or ARM.

rnjudge · 2023-06-01T18:51:52Z

@pombredanne I was able to install scancode-toolkit-mini and will suggest this to Tern's M1 users going forward. I unfortunately can't pledge other resources right now.

pombredanne · 2023-06-01T19:00:53Z

@rnjudge re:

I was able to install scancode-toolkit-mini and will suggest this to Tern's M1 users going forward.

This is awesome and this was designed for this use case too! Glad you got it to work.

I unfortunately can't pledge other resources right now.

that's cool too.

jayanth-kumar-morem · 2024-01-27T10:26:38Z

Facing the same issue,

Im on Mac M1 and I am using the ubuntu docker shell with volume mounted

docker run -it -v <path>:<path> ubuntu:22.04

Inside the docker shell, when I install the pre-requisites and run ./configure --clean && ./configure --dev, the dependecies seems to have conflicts.

Collecting regipy>=3.1.0 (from scancode-toolkit==32.0.8)
  Using cached regipy-3.1.0-py3-none-any.whl (65 kB)
INFO: pip is looking at multiple versions of scancode-toolkit[packages,testing] to determine which version is compatible with other requirements. This could take a while.
ERROR: Cannot install scancode-toolkit[packages,testing]==32.0.8 because these package versions have conflicting dependencies.

The conflict is caused by:
    scancode-toolkit[packages,testing] 32.0.8 depends on packagedcode-msitools>=0.101.210706; platform_system == "Linux" and extra == "packages"
    The user requested (constraint) packagedcode-msitools==0.101.210706

To fix this you could try to:
1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflict

ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/topics/dependency-resolution/#dealing-with-dependency-conflicts

pombredanne · 2024-01-29T13:58:25Z

@jayanth-kumar-morem the workaround is to enable the X86 "rosetta" emulation on your shell (which is also done automatically when you use the app archive's "scancode" root script https://github.com/nexB/scancode-toolkit/blob/f70bbb7d9d9bab40a9d504e664bc945b6a1630e8/scancode#L124) or use the arch command to force a one-off x86 mode like in this script.
And of course getting a better way to automate the build of all our native deps on ARM in general would be awesome!

rnjudge added the bug label Jan 17, 2023

rnjudge mentioned this issue Jan 17, 2023

Error if Tern calls Scancode-Toolkit tern-tools/tern#1202

Closed

pombredanne mentioned this issue Jan 18, 2023

scancode installed via pip command not working on Mac M1 #3166

Open

rnjudge mentioned this issue Mar 7, 2023

Tern fails in license_expression tern-tools/tern#1203

Closed

keshav-space mentioned this issue Feb 1, 2024

Add GitHub Actions CI for Apple M1 #3649

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Many scancode dependencies are broken for Mac M1 #3205

Many scancode dependencies are broken for Mac M1 #3205

rnjudge commented Jan 17, 2023 •

edited

Loading

rnjudge commented Jan 17, 2023

rnjudge commented Jan 17, 2023

pombredanne commented Jan 18, 2023

Jeeppler commented Jan 18, 2023

pombredanne commented Jan 18, 2023

Jeeppler commented Jan 18, 2023

rnjudge commented Jan 20, 2023

pombredanne commented Jan 21, 2023 •

edited

Loading

rnjudge commented Jan 23, 2023

stevespringett commented Feb 23, 2023

rnjudge commented May 19, 2023

pombredanne commented May 21, 2023 •

edited

Loading

pombredanne commented May 21, 2023 •

edited

Loading

pombredanne commented May 22, 2023

enricozb commented May 22, 2023

rnjudge commented Jun 1, 2023

pombredanne commented Jun 1, 2023

jayanth-kumar-morem commented Jan 27, 2024

pombredanne commented Jan 29, 2024

Many scancode dependencies are broken for Mac M1 #3205

Many scancode dependencies are broken for Mac M1 #3205

Comments

rnjudge commented Jan 17, 2023 • edited Loading

Description

How To Reproduce

System configuration

rnjudge commented Jan 17, 2023

rnjudge commented Jan 17, 2023

pombredanne commented Jan 18, 2023

Jeeppler commented Jan 18, 2023

pombredanne commented Jan 18, 2023

Jeeppler commented Jan 18, 2023

rnjudge commented Jan 20, 2023

pombredanne commented Jan 21, 2023 • edited Loading

rnjudge commented Jan 23, 2023

stevespringett commented Feb 23, 2023

rnjudge commented May 19, 2023

pombredanne commented May 21, 2023 • edited Loading

pombredanne commented May 21, 2023 • edited Loading

pombredanne commented May 22, 2023

enricozb commented May 22, 2023

rnjudge commented Jun 1, 2023

pombredanne commented Jun 1, 2023

jayanth-kumar-morem commented Jan 27, 2024

pombredanne commented Jan 29, 2024

rnjudge commented Jan 17, 2023 •

edited

Loading

pombredanne commented Jan 21, 2023 •

edited

Loading

pombredanne commented May 21, 2023 •

edited

Loading

pombredanne commented May 21, 2023 •

edited

Loading