compiler.parse("d[1] = 1") should have a single tuple as subs


>>> compiler.parse("d[1] = 1")
Module(None, Stmt([Assign([Subscript(Name('d'), 'OP_ASSIGN',
[Const(1)])], Const(1))]))
>>> compiler.parse("d[1,] = 2")
Module(None, Stmt([Assign([Subscript(Name('d'), 'OP_ASSIGN',
[Const(1)])], Const(2))]))
>>> compiler.parse("d[1,2] = 3")
Module(None, Stmt([Assign([Subscript(Name('d'), 'OP_ASSIGN', [Const(1),
Const(2)])], Const(3))]))
>>> compiler.parse("d[(1,)] = 2")
Module(None, Stmt([Assign([Subscript(Name('d'), 'OP_ASSIGN',
[Tuple([Const(1)])])], Const(2))]))

I just see that my patch is not correct, since the following is
supported by the language:

O[1:2:3, 4:5:6]

Where O[1:2:3, 4:5:6] == O[slice(1,2,3), slice(4,5,6)] ==
O.__getitem__((slice(1,2,3), slice(4,5,6)))

patch which honors O[1:2:3, 4:5:6]

your patch is no more a diff file (the 'previous' file is empty)

Sorry. Renamed .bak to .orig ...
Here's the patch compiler/transformer.py (against python 2.5)

PythonTidy encounters this problem.

o http://lacusveris.com/PythonTidy/PythonTidy.python

It is unable correctly to render line 694 of test_grammar.py in the
Python Test Suite:

    d[1,] = 2

becomes:

    d[1] = 2

because the *compiler* module does not return an abstract syntax tree
containing a tuple for the subscript.

The patch should have a test.

I think this should be closed out, since the compiler package was deprecated in 2.6.

Here are four ways to generate things called in the documentation Abstract Syntax Trees (ASTs).  Obviously they are not all the same kind of object:

#!/usr/local/bin/python2.6

import sys
import compiler
import parser
import ast

STATEMENT = 'd[1,] = 2'

print 'level %s' % sys.version
print compiler.parse(STATEMENT)
print parser.suite(STATEMENT).tolist()
print ast.dump(
    compile(STATEMENT, '<string>', 'exec', ast.PyCF_ONLY_AST),
    annotate_fields=False,
    include_attributes=False,
    )
print ast.dump(
    ast.parse(STATEMENT),
    annotate_fields=False,
    include_attributes=False,
    )

# Fin

Here are the results:

>>> level 2.6.2 (r262:71600, Jun 29 2009, 08:08:18) 
[GCC 4.3.2]
>>> Module(None, Stmt([Assign([Subscript(Name('d'), 'OP_ASSIGN', [Const(1)])], Const(2))]))
>>> [257, [267, [268, [269, [270, [327, [304, [305, [306, [307, [308, [310, [311, [312, [313, [314, [315, [316, [317, [318, [1, 'd']], [322, [9, '['], [323, [324, [304, [305, [306, [307, [308, [310, [311, [312, [313, [314, [315, [316, [317, [318, [2, '1']]]]]]]]]]]]]]]], [12, ',']], [10, ']']]]]]]]]]]]]]]]], [22, '='], [327, [304, [305, [306, [307, [308, [310, [311, [312, [313, [314, [315, [316, [317, [318, [2, '2']]]]]]]]]]]]]]]]]], [4, '']]], [0, '']]
>>> Module([Assign([Subscript(Name('d', Load()), Index(Tuple([Num(1)], Load())), Store())], Num(2))])
>>> Module([Assign([Subscript(Name('d', Load()), Index(Tuple([Num(1)], Load())), Store())], Num(2))])

To me the *compiler* module has vestigial utility.  It would be nice if it returned correct results.

I don't see why the compiler module is any better than any of the other ways that produce reasonable AST.

> I don't see why the compiler module is any better than any of the other 
> ways that produce reasonable AST.

It is available on Python releases older than 2.6?

It's available at least since 2.4.

We're using it in Pyjamas (pyjs.org) to generate javascript code from python code.

If there are better ways to produce python asts, I'd be happy to know that.

As I said before, the patch needs a test and should remove those ugly \ from the list comprehension.

This can only be fixed in 2.7, where compiler is deprecated in favor of ast. Compiler is gone in 3.x. Kees, are you planning to update the patch (with no guarantee anyone will apply) or should we close this?

Added fix for python 2.7, which includes a test (testDictWithTupleKey) for the compiler test (Lib/test/test_compiler.py).

Can somebody set this to "patch review" and do the honours please.  FWIW I don't like "tulplesub" in the patch.

I am closing this because a) Meador is correct that we do not normally patch deprecated features and b) the current emphasis on 2.7-only patches is security and keeping 2.7 working on current systems.  The fix would only benefit 2.7.9+ code or 2.7.9+ and 3.x code, neither of which should be using compiler module.

Benjamin, if you disagree, review and apply, as no one else will.