search

symforce-org / symforce

Fast symbolic computation, code generation, and nonlinear optimization for robotics
https://symforce.org
Apache License 2.0
1.41k stars 145 forks source link

`generate_function` should not generate files on storage automatically #171

Open KOLANICH opened 2 years ago

KOLANICH commented 2 years ago

Is your feature request related to a problem? Please describe. Could generate_function instead of creating files just return their source code? It feels strange that the files on storage are generated, since if a user needs a file, he can easily save it himself.

Describe the solution you'd like

  1. a property for accessing sources. Made of objects of GeneratedSource class, each one contains the properties
    • path of PurePath with a path of the file rel to the output dir
    • warnings - an array of warjings
    • source - source code
  2. generated_files lazily take source of corresponding fikes qnd write them to disk. Usage of generated_files causes a deprecation warning to be emitted.

Describe alternatives you've considered No.

aaron-skydio commented 2 years ago

Yeah I think this is a good idea - probably wouldn't deprecate the option of having symforce write to disk but would be good to have the option not to. One question would be whether all other tools we call out to are ok or can be made to work with this (e.g. autoformatters, skymarshal).

What are some examples of warnings to include in point 1 bullet 2? Not sure what you mean there (or if there are some warnings we currently emit in other forms that this would be replacing)

KOLANICH commented 2 years ago

What are some examples of warnings to include in point 1 bullet 2? Not sure what you mean there (or if there are some warnings we currently emit in other forms that this would be replacing)

I guess no for now, but in future it may make sense to wind up an own DSL (probably DAG-of-dicts-based (can be serialized/parsed into/from many formats (s.a. CBOR, terraform tf2, json, NEON and so on), but YAML is usually used), as lot of projects do). Some projects doing that have a model that 1 input file corresponds to 1 output file. Then for 1 input compiler generates 1 output + bunch of warnings, if any, or, in the case of an error, just null output + bunch of errors. Then it can be either written to disk using the attached path, or just used disklessly, for example by showing to a user, or just disklessly compiling and executing.