a pyrex-inspired for-i-from-1-to-n construct backported to Python.Good idea? Bad Idea? PEP?

Skip Montanaro skip at pobox.com
Wed Apr 2 17:26:45 CEST 2003


    Peter> I just scanned a few tens of thousands of lines of our code, and
    Peter> as near as I can tell only 13 cases out of over 160 instances of
    Peter> "for" loops are using range() or xrange().  

Similar numbers here: 146 out of 898.

It appears the builtins namespace (or at least foreign modules' namespaces)
will get somewhat less dynamic -- just enough less that the byte compiler
can tell if an instance of x?range() is calling the usual builtin instead of
a shadow copy -- so tools that care will be able to optimize "for i in
range(x)" just as well as the pyrex syntax.  There have been discussions
recently about this topic in python-dev.  Also, one of the sprints at PyCon
was aimed at a new version of the bytecode compiler, probably one which
generates code from an abstract syntax tree, allowing easier analysis (and
thus optimization) before generation of bytecode.  So instead of

    LOAD_GLOBAL range

the bytecode might be

    LOAD_BUILTIN range

which tools like Psyco can detect can use to their advantage.

Skip





More information about the Python-list mailing list