diff options
Diffstat (limited to 'lib/python2.7/lib2to3/fixes/fix_idioms.py')
-rw-r--r-- | lib/python2.7/lib2to3/fixes/fix_idioms.py | 152 |
1 files changed, 152 insertions, 0 deletions
diff --git a/lib/python2.7/lib2to3/fixes/fix_idioms.py b/lib/python2.7/lib2to3/fixes/fix_idioms.py new file mode 100644 index 0000000..37b6eef --- /dev/null +++ b/lib/python2.7/lib2to3/fixes/fix_idioms.py @@ -0,0 +1,152 @@ +"""Adjust some old Python 2 idioms to their modern counterparts. + +* Change some type comparisons to isinstance() calls: + type(x) == T -> isinstance(x, T) + type(x) is T -> isinstance(x, T) + type(x) != T -> not isinstance(x, T) + type(x) is not T -> not isinstance(x, T) + +* Change "while 1:" into "while True:". + +* Change both + + v = list(EXPR) + v.sort() + foo(v) + +and the more general + + v = EXPR + v.sort() + foo(v) + +into + + v = sorted(EXPR) + foo(v) +""" +# Author: Jacques Frechet, Collin Winter + +# Local imports +from .. import fixer_base +from ..fixer_util import Call, Comma, Name, Node, BlankLine, syms + +CMP = "(n='!=' | '==' | 'is' | n=comp_op< 'is' 'not' >)" +TYPE = "power< 'type' trailer< '(' x=any ')' > >" + +class FixIdioms(fixer_base.BaseFix): + explicit = True # The user must ask for this fixer + + PATTERN = r""" + isinstance=comparison< %s %s T=any > + | + isinstance=comparison< T=any %s %s > + | + while_stmt< 'while' while='1' ':' any+ > + | + sorted=any< + any* + simple_stmt< + expr_stmt< id1=any '=' + power< list='list' trailer< '(' (not arglist<any+>) any ')' > > + > + '\n' + > + sort= + simple_stmt< + power< id2=any + trailer< '.' 'sort' > trailer< '(' ')' > + > + '\n' + > + next=any* + > + | + sorted=any< + any* + simple_stmt< expr_stmt< id1=any '=' expr=any > '\n' > + sort= + simple_stmt< + power< id2=any + trailer< '.' 'sort' > trailer< '(' ')' > + > + '\n' + > + next=any* + > + """ % (TYPE, CMP, CMP, TYPE) + + def match(self, node): + r = super(FixIdioms, self).match(node) + # If we've matched one of the sort/sorted subpatterns above, we + # want to reject matches where the initial assignment and the + # subsequent .sort() call involve different identifiers. + if r and "sorted" in r: + if r["id1"] == r["id2"]: + return r + return None + return r + + def transform(self, node, results): + if "isinstance" in results: + return self.transform_isinstance(node, results) + elif "while" in results: + return self.transform_while(node, results) + elif "sorted" in results: + return self.transform_sort(node, results) + else: + raise RuntimeError("Invalid match") + + def transform_isinstance(self, node, results): + x = results["x"].clone() # The thing inside of type() + T = results["T"].clone() # The type being compared against + x.prefix = u"" + T.prefix = u" " + test = Call(Name(u"isinstance"), [x, Comma(), T]) + if "n" in results: + test.prefix = u" " + test = Node(syms.not_test, [Name(u"not"), test]) + test.prefix = node.prefix + return test + + def transform_while(self, node, results): + one = results["while"] + one.replace(Name(u"True", prefix=one.prefix)) + + def transform_sort(self, node, results): + sort_stmt = results["sort"] + next_stmt = results["next"] + list_call = results.get("list") + simple_expr = results.get("expr") + + if list_call: + list_call.replace(Name(u"sorted", prefix=list_call.prefix)) + elif simple_expr: + new = simple_expr.clone() + new.prefix = u"" + simple_expr.replace(Call(Name(u"sorted"), [new], + prefix=simple_expr.prefix)) + else: + raise RuntimeError("should not have reached here") + sort_stmt.remove() + + btwn = sort_stmt.prefix + # Keep any prefix lines between the sort_stmt and the list_call and + # shove them right after the sorted() call. + if u"\n" in btwn: + if next_stmt: + # The new prefix should be everything from the sort_stmt's + # prefix up to the last newline, then the old prefix after a new + # line. + prefix_lines = (btwn.rpartition(u"\n")[0], next_stmt[0].prefix) + next_stmt[0].prefix = u"\n".join(prefix_lines) + else: + assert list_call.parent + assert list_call.next_sibling is None + # Put a blank line after list_call and set its prefix. + end_line = BlankLine() + list_call.parent.append_child(end_line) + assert list_call.next_sibling is end_line + # The new prefix should be everything up to the first new line + # of sort_stmt's prefix. + end_line.prefix = btwn.rpartition(u"\n")[0] |