README.md

Notes

This project is now under inactive development.

Want to take over? Email me.

Moshmosh

An advanced syntax extension system implemented in pure python.

pip install -U moshmosh-base --no-compile

Note that --no-compile is required.

Preview

Working with IPython

You should copy moshmosh_ipy.py to $USER/.ipython/profile_default/startup/.

If this directory does not exist, use command ipython profile create to instantiate.

Some examples about pattern matching, pipelines and quick lambdas:

Some examples about the scoped operators:

Working with regular Python files

Import moshmosh in your main module:

Then, in mypackage.py, start coding with a pragma comment # moshmosh?, then you can use moshmosh extension system.

Case Study : Pattern Matching

The matching protocol which stems from Python-ideas mailing list is introduced in, which means you can define your own patterns conveniently. The link is here.

# moshmosh?
# +pattern-matching

class GreaterThan:
    def __init__(self, v):
        self.v = v

    def __match__(self, cnt: int, to_match):
        if isinstance(to_match, int) and cnt is 0 and to_match > self.v:
            return () # matched
        # 'return None' indicates 'unmatched'

with match(114, 514):
    if (GreaterThan(42)() and a, b):
        print(b, a)
# 514 114

Note that the matching clauses should be exhaustive, otherwise, a moshmosh.extensions.pattern_matching.runtime.NotExhaustive might get raised.

The supported Patterns are listed here, which is of course much more powerful than most programming languages.

• And pattern: pat1 and pat2 and pat3 ...
• Or pattern: pat1 or pat2 or pat3...
• Pin pattern: pin(value), this is quite useful. See Elixir Pin Operator
• Literal pattern: 1, "str", 1+2j, (1, 2)
• As pattern: a, var
• Wildcard: _
• Guard: when(cond1, cond2, cond3)
• Nested patterns:
  • Tuple: (pat1, pat2, pat3), (pat1, *pat2, pat3)
  • List: [pat1, pat2, pat3], [pat1, *pat2, pat3]
  • Recogniser: Cons(pat1, pat2, pat3), note that, the function Cons.__match__(<n arg>, value_to_match) is exact the protocol.

The pattern matching should be more efficient than those hand-written codes without ugly optimizations.

Besides, Moshmosh's pattern matching is orders of magnitude faster than any other alternatives.

Case Study : Template-Python

This is relatively a simple quasiquote implementation, inspired by MetaOCaml. It does not support manual splices or nested quotations, but the function arguments are automatically spliced.

# moshmosh?
# +template-python

@quote
def f(x):
    x + 1
    x = y + 1

from moshmosh.ast_compat import ast
from astpretty import pprint

stmts = f(ast.Name("a"))
pprint(ast.fix_missing_locations(stmts[0]))
pprint(ast.fix_missing_locations(stmts[1]))

# =>
Expr(
    lineno=7,
    col_offset=4,
    value=BinOp(
        lineno=7,
        col_offset=4,
        left=Name(lineno=7, col_offset=4, id='a', ctx=Load()),
        op=Add(),
        right=Num(lineno=7, col_offset=8, n=1),
    ),
)
Assign(
    lineno=8,
    col_offset=4,
    targets=[Name(lineno=8, col_offset=4, id='a', ctx=Store())],
    value=BinOp(
        lineno=8,
        col_offset=8,
        left=Name(lineno=8, col_offset=8, id='y', ctx=Load()),
        op=Add(),
        right=Num(lineno=8, col_offset=12, n=1),
    ),
)

Case Study: Lazy Import

# moshmosh?
# +lazy-import
import numpy as np
# -lazy-import

# in fact numpy is not imported here,
# and once you use it, it gets imported.

def arr10():
    # The first time call
    # arr10 will enforce the import of numpy.
    return np.zeros(10)

After the lazy modules are actually imported, there's no overhead to access their members.

However, please only import modules when using lazy-import.

The use case is about the necessary cross-import when you want to organise your codebase in a more fine-grained way.

Acknowledgements

• future-fstrings
• Pattern matching in Python
  • older implementations
  • search "pattern matching" at Python-ideas.