py_converter

Overview ¶

Package py_converter lowers Python source into gIR so the taint engine in internal/analysis can analyze it, mirroring the public shape of converters/go (NewConverter / Converter.ConvertFile).

There is no Python front end reachable from Go without a new dependency, so this converter shells out to an embedded helper script (pyast.py) that parses the file with the standard library's `ast` module and prints a compact JSON tree; convertModule/convertFunction/lowerStmt/lowerExpr turn that into gIR. ConvertFile returns a clear error if python3 is not on PATH.

Known limitations:

• A real control-flow graph IS built, via converters/ssabuild: if/while/for/ try get their own basic blocks, and PHIs are inserted on demand (Braun et al.) at the joins. A branch-free body still lowers to exactly one block, which keeps the engine's linear fast path. What remains approximate is exceptions and loop exits: lowerTry models a raise as ONE opaque edge from the try body into a single merged handler block, with no exception typing, and `break`/`continue` are dropped rather than re-routed, so a loop's exit edges are the fall-through ones only.
• Classes are only partially modeled: methods become functions named "<Class>.<method>", but other class-body statements (class attributes, nested classes, decorators) are ignored.
• Expression coverage spans calls, attribute/subscript reads, binary/unary and boolean operators, comprehensions, container literals, unpacking assignment, walrus (`:=`), `await`, f-strings, str.format, constants, and names. Lambdas and decorators are not specifically modeled; an unhandled expression becomes an OP_CODE_INTRINSIC "py.unsupported" and an unhandled statement is dropped, rather than aborting conversion.