Improving CM core

[gfursin]

Aggregating tasks to improve CM core based on user feedback (this ticket is being gradually updated):

Functionality

• Add support to check min CM version for automations and scripts
• Move string checks to belong to [True, yes, on] to cmind.utils to avoid duplicates
• Add cm checkout repo to checkout pulled repos
• Add support to debug CM core and scripts (cmind.utils.debug_here) (tested with Visual Studio)
• Develop alias to pull mlcommons@cm4mlops instead of mlcommons@ck (when -b is not used)
• Add portable md5sum to cmind.utils from here - it is often used in various CM scripts
• Add portable tar and zip extractor to cmind.utils based on ABTF prototyping - used in many projects
• Add logging to CM core
  • Agree on CM logging names/streams
  • Convert CM prints to logging
  • Convert CM script prints to logging
• Print internal status (version, paths to repos, python used, etc): cm test core
• URGENT: setup.py will be deprecated soon and we need to provide a new way to install CM - help is appreciated!

Tests

• Unify GitHub test names
• Add more unit tests to cover most/all CM internals across different platforms
• Agree how to keep control of code quality of CM scripts and improve quality of existing CM scripts (from prototype to production)

Docs

• Agree on how to show optional flags. Should be similar to Python parser: cm run script ... [--help] [--path PATH] ...
• Generate CM API docs per version with sphynx and upload to website
• Improve basic CM framework docs/tutorials based on cmake/bazel/workflow concept
  • CLI
  • API
  • add new recipe
  • share new recipe
  • work with CM scripts
    • add new script
    • reuse new script
    • CM cache
    • workflow (env/state)
      • deps with conditions
      • customize
• Provide guide for user contributions and to report issues

Tutorials

• Basic CM concepts
• CM automations (use, extend, add)
• CM scripts

Brainstorm

• Add possibility to have strict validation of inputs and types for production environments
• Move CM index to sqlite to speed up search by tags and enable parallel execution - need to brainstorm
• Check how to parallelize cmind.access runs (if needed)
• Simplify get git, download file, extract file with custom CM cache entry ...
• Simplify docker generation via separate docker automation
• Test CM scripts with different versions of dependencies
  • Enable crowd-testing and reporting similar to cTuning