mvdef 0.1.5

Package to move functions and their import statements between files

mvdef

Summary

Move function definitions from one file to another, moving or copying associated import statements along with them.

Installation

mvdef is available on PyPi: install it using pip install mvdef

After installing to your environment from PyPi, the mvdef command will be available on the command line. Type mvdef -h to get the following usage message.

Usage

usage: mvdef [-h] [--demo] [-m MVDEF] [--src SRC] [--dst DST] [-r] [-b] [-d]

optional arguments:
  -h, --help            show this help message and exit
  --demo
  -m MVDEF, --mvdef MVDEF
  --src SRC
  --dst DST
  -r, --report
  -b, --backup
  -d, --dry-run

Example usage

mvdef -m func1 --src path/to/source_program.py --dst path/to/destination_program.py -rb

will move the funcdef named func1 from source_program.py to destination_program.py, while reporting output (thanks to the -r flag) and making backups (thanks to the -b flag).

• Further functions can be moved by adding more -m flags each followed by a function name, e.g. mvdef -m func1 -m func2 -m func3 ...

For instance to carry out the demo from the command line, run mvdef --demo, or equivalently:

mvdef -m show_line --src mvdef/example/demo_program.py --dst mvdef/example/new_file.py -rb

If you feel like there would be a more concise or useful way of specifying command line arguments, please let me know by making a GitHub issue, however I chose not to just have anonymous parameters as over time it could get easy to forget which are which.

Motivation

My workflow typically involves a process of starting to work in one file, with one big function, and later breaking out that function into smaller functions once I've settled on the first draft of control flow organisation.

After 'breaking out' the code into multiple smaller functions in this way, it'll often be the case that some of the functions are thematically linked (e.g. they operate on the same type of variable or are connected in the workflow). In these cases, it's useful to move function definitions out of the main file, and into a module file together, then import their names back into the main file if or as needed.

• If I have two functions A and B, and my file calculates A() + B(), not only can I move A and B into some other module file mymodule, but I can put a wrapper function C in it too, and reduce the number of names I import

  def C():
      ans = A() + B()
      return ans
  

  both saving on the complexity in the main file, and giving 'more space' to focus on A, B and C separate from the complexity of what's going on in the main file (which in turn makes theme-focused tasks like documentation more straightforward).

The problem comes from then having to do the mental calculation (and often old fashioned searching for library-imported names within the file) of whether the functions I am trying to move out into another file rely on names that came from import statements, and if so, whether there are other functions which also rely on the same imported names. If I guess and get it wrong, I may then have to run the code multiple times and inspect the error message tracebacks until I figure out the full set, or else just reset it to where I was if things get particularly messy, in which case the time spent trying to move functions and import statements manually was wasted.

All of this motivates a library which can handle such operations for me, not just because it requires some thought to do manually so much as that it's a waste of development time, and what's more it interrupts the train of thought (increasingly so as the software gets more complex, with more functions and libraries to consider).

Software can scale to handle these higher levels of complexity no differently than it can handle a simple case, and I began writing this on the basis that "if I'm going to figure it out for this one instance, I may as well code it for any instance going forward".

Demo

• Above: the function show_line is moved from the source file (left) to the destination file (right), taking along import statements (or more precisely, taking individual aliases from import statements, which then form new import statements in the destination file). The top right of the image displays a report of the 'agenda' which mvdef follows, alias by alias, to carry out these changes.
• This demo can be reproduced by running python -im mvdef --demo from the main directory upon cloning this repository, and inspecting the source file (demo_program.py) and destination file (new_file.py) under mvdef/example/.
• This demo creates hidden .backup files, which can be used to 'reset' the demo by moving them back so as to overwrite the original files.

Project status and future plans

• November 2019: This library is currently working only as a proof of concept, with a demo, and not yet working for code.
• December 2019: The demo now works, and using the command line flags it works as a command line tool for any list of functions and any pair of files specified.

I'd like this to end up being a command line tool that assists the development workflow similar to how black has simplified linting to best practice conventions for Python code style, as a tool callable on a Python file to change it in place, and reliable enough to trust it not mess up any of your files in the process.

Approach

The idea is to run a command like mvdef src.py dst.py fn1 fn2 fn3 to do the following:

• Back up src.py and dst.py, as src.py.backup and dst.py.backup in case it doesn't work
  • Function completed in src.backup⠶backup() with dry_run parameter, called in src.demo
  • I'd also like to add the option to rename functions, using a pattern or list to rename as
    • src.rename not yet implemented
• Optional: Define some test that should pass after the refactor, when src.py imports fn1, fn2, fn3 from dst.py
  • Tests defined for all functions in example.demo_program in example.test⠶test_report, called in __main__
    • Tests are checked and raise a RuntimeError if they fail at this stage (i.e. the whole process aborts before any files are modified or created)
  • If not, it would just be a matter of testing this manually (i.e. not necessary to define test to use tool, but suggested best practice)
• Enumerate all import statements in src.py (nodes in the AST of type ast.Import)
  • src.ast_util⠶annotate_imports returns this list, which gets assigned to imports in src.ast_util⠶parse_mv_funcs
• Enumerate all function definitions in src.py (nodes in the AST of type ast.FunctionDef)
  • ast⠶parse provides this as the .body nodes which are of type ast.FunctionDef.
    • This subset of AST nodes is assigned to defs in src.ast_util⠶ast_parse.
• Find the following subsets:
  • mvdefs: subset of all function definitions which are to be moved (fn1, fn2, fn3)
    • This subset is determined by cross-referencing the names of the defs (from previous step) against the mvdefs (list of functions to move, such as ["fn1", "fn2", "fn3"]), in the dedicated function src.ast_util⠶get_def_names, then returned by src.ast_tools⠶ parse_mv_funcs as a list, assigned to mvdefs in src.ast_util⠶ast_parse.
  • nonmvdefs: subset of all function definitions not to be moved (not in mvdefs)
    • This subset is determined by negative cross-ref. to names of the defs against the mvdefs (such as ["fn4", "fn5", "fn6"]), again using src.ast_util⠶get_def_names, then returned by src.ast_util⠶parse_mv_funcs as a list, assigned to nonmvdefs in src.ast_util⠶ast_parse.
  • mv_imports: Import statements used only by the functions in mvdefs
  • nonmv_imports: Import statements used only by the functions in nonmvdefs
  • mutual_imports: Import statements used by both functions in mvdefs and nonmvdefs
  • nondef_imports: Import statements not used by any function
    • Note that these may no longer be in use, but this can only be confirmed by checking outside of function definitions too.
    • Potentially add this feature later, for now just report which imports aren't used.
• Handle the 3 types of imported names:
  • Move the import statements in mv_imports (received as "take")
  • Keep the import statements in nonmvdef_imports
  • Copy the import statements in mutual_imports (received as "echo")
• ...and also:
  • Handle moving one import name from an import statement importing multiple names (i.e. where you can't simply copy the line)
  • Handle multi-line imports (i.e. where you can't simply find the names on one line)
  • ...and remove unused import statements (neither in/outside any function definitions)
• ...and only then move the function definitions in mvdefs across
• If tests were defined in step 2, check that these tests run
  • For the demo, the tests are checked (by running test_report a 2nd time) after src.demo⠶run_demo has returned a parsed version of the source and destination files (which will only matter once the parameter nochange is set to False in run_demo, allowing it to propagate through the call to src.demo⠶parse_example into a call to src.ast_util⠶ast_parse(..., edit=True) and ultimately carry out in-place editing of the source and/or destination file/s as required).
  • If they fail, ask to restore the backup and give the altered src/dst .py files .py.mvdef_fix suffixes (i.e. always permit the user to exit gracefully with no further changes to files rather than forcing them to)