python-peps/pep-0548.rst

281 lines
8.0 KiB
ReStructuredText
Raw Normal View History

2017-09-05 13:37:52 -04:00
PEP: 548
Title: More Flexible Loop Control
Version: $Revision$
Last-Modified: $Date$
Author: R David Murray
2017-09-06 19:35:27 -04:00
Status: Rejected
Type: Standards Track
Content-Type: text/x-rst
Created: 05-Sep-2017
Python-Version: 3.7
2017-09-06 19:35:27 -04:00
Post-History: 05-Aug-2017
Rejection Note
==============
Rejection by Guido:
https://mail.python.org/pipermail/python-dev/2017-September/149232.html
Abstract
2017-09-05 14:37:37 -04:00
========
2017-09-05 14:54:46 -04:00
This PEP proposes enhancing the ``break`` and ``continue`` statements
with an optional boolean expression that controls whether or not
they execute. This allows the flow of control in loops to be
expressed more clearly and compactly.
Motivation
2017-09-05 14:37:37 -04:00
==========
2017-09-05 14:54:46 -04:00
Quoting from the rejected :pep:`315`:
2017-09-05 14:54:46 -04:00
It is often necessary for some code to be executed before each
evaluation of the while loop condition. This code is often
duplicated outside the loop, as setup code that executes once
before entering the loop::
2017-09-05 14:54:46 -04:00
<setup code>
while <condition>:
<loop body>
<setup code>
2017-09-05 14:54:46 -04:00
That PEP was rejected because no syntax was found that was superior
to the following form::
2017-09-05 14:54:46 -04:00
while True:
<setup code>
if not <condition>:
break
<loop body>
2017-09-05 14:54:46 -04:00
This PEP proposes a superior form, one that also has application to
for loops. It is superior because it makes the flow of control in
loops more explicit, while preserving Python's indentation aesthetic.
Syntax
2017-09-05 14:37:37 -04:00
======
2017-09-05 14:54:46 -04:00
The syntax of the break and continue statements are extended
as follows::
2017-09-05 14:54:46 -04:00
break_stmt : "break" ["if" expression]
continue_stmt : "continue" ["if" expression]
2017-09-05 14:54:46 -04:00
In addition, the syntax of the while statement is modified as follows::
2017-09-05 14:54:46 -04:00
while_stmt : while1_stmt|while2_stmt
while1_stmt : "while" expression ":" suite
["else" ":" suite]
while2_stmt : "while" ":" suite
Semantics
2017-09-05 14:37:37 -04:00
=========
2017-09-05 14:54:46 -04:00
A ``break if`` or ``continue if`` is executed if and only if
``expression`` evaluates to true.
2020-08-07 12:35:20 -04:00
A ``while`` statement with no expression loops until a break or return
is executed (or an error is raised), as if it were a ``while True``
2017-09-05 14:54:46 -04:00
statement. Given that the loop can never terminate except in a
way that would not cause an ``else`` suite to execute, no ``else``
suite is allowed in the expressionless form. If practical, it
should also be an error if the body of an expressionless ``while``
does not contain at least one ``break`` or ``return`` statement.
2017-09-05 14:37:37 -04:00
Justification and Examples
==========================
2017-09-05 14:54:46 -04:00
The previous "best possible" form::
2017-09-05 14:54:46 -04:00
while True:
<setup code>
if not <condition>:
break
<loop body>
2017-09-05 14:54:46 -04:00
could be formatted as::
2017-09-05 14:54:46 -04:00
while True:
<setup code>
if not <condition>: break
<loop body>
2017-09-05 14:54:46 -04:00
This is superficially almost identical to the form proposed by this
PEP::
2017-09-05 14:54:46 -04:00
while:
<setup code>
break if not <condition>
<loop body>
The significant difference here is that the loop flow control
keyword appears *first* in the line of code. This makes it easier
to comprehend the flow of control in the loop at a glance, especially
when reading colorized code.
For example, this is a common code pattern, taken in this case
from the tarfile module::
2017-09-05 14:54:46 -04:00
while True:
buf = self._read(self.bufsize)
if not buf:
break
t.append(buf)
Reading this, we either see the break and possibly need to think about
where the while is that it applies to, since the break is indented
under the if, and then track backward to read the condition that
triggers it; or, we read the condition and only afterward discover
that this condition changes the flow of the loop.
With the new syntax this becomes::
while:
buf = self._read(self.bufsize)
break if not buf
t.append(buf)
Reading this we first see the ``break``, which obviously applies to
2017-09-05 14:54:46 -04:00
the while since it is at the same level of indentation as the loop
body, and then we read the condition that causes the flow of control
to change.
Further, consider a more complex example from sre_parse::
while True:
c = self.next
self.__next()
if c is None:
if not result:
raise self.error("missing group name")
raise self.error("missing %s, unterminated name" % terminator,
len(result))
2017-09-05 14:54:46 -04:00
if c == terminator:
if not result:
raise self.error("missing group name", 1)
break
result += c
return result
This is the natural way to write this code given current Python
loop control syntax. However, given ``break if``, it would be more
natural to write this as follows::
while:
c = self.next
self.__next()
break if c is None or c == terminator
result += c
if not result:
raise self.error("missing group name")
elif c is None:
raise self.error("missing %s, unterminated name" % terminator,
len(result))
return result
This form moves the error handling out of the loop body, leaving the
loop logic much more understandable. While it would certainly be
possible to write the code this way using the current syntax, the
proposed syntax makes it more natural to write it in the clearer form.
The proposed syntax also provides a natural, Pythonic spelling of
the classic ``repeat ... until <expression>`` construct found in
other languages, and for which no good syntax has previously been
found for Python::
while:
...
break if <expression>
The tarfile module, for example, has a couple of "read until" loops like
the following::
while True:
s = self.__read(1)
if not s or s == NUL:
break
2017-09-05 14:54:46 -04:00
With the new syntax this would read more clearly::
2017-09-05 14:54:46 -04:00
while:
s = self.__read(1)
break if not s or s == NUL
2017-09-05 14:54:46 -04:00
The case for extending this syntax to ``continue`` is less strong,
but buttressed by the value of consistency.
It is much more common for a ``continue`` statement to be at the
end of a multiline if suite, such as this example from zipfile ::
while True:
try:
self.fp = io.open(file, filemode)
except OSError:
if filemode in modeDict:
filemode = modeDict[filemode]
continue
2017-09-05 14:54:46 -04:00
raise
break
The only opportunity for improvement the new syntax would offer for
this loop would be the omission of the ``True`` token.
On the other hand, consider this example from uuid.py::
for i in range(adapters.length):
ncb.Reset()
ncb.Command = netbios.NCBRESET
ncb.Lana_num = ord(adapters.lana[i])
if win32wnet.Netbios(ncb) != 0:
continue
ncb.Reset()
ncb.Command = netbios.NCBASTAT
ncb.Lana_num = ord(adapters.lana[i])
ncb.Callname = '*'.ljust(16)
ncb.Buffer = status = netbios.ADAPTER_STATUS()
if win32wnet.Netbios(ncb) != 0:
continue
status._unpack()
bytes = status.adapter_address[:6]
if len(bytes) != 6:
continue
return int.from_bytes(bytes, 'big')
This becomes::
for i in range(adapters.length):
ncb.Reset()
ncb.Command = netbios.NCBRESET
ncb.Lana_num = ord(adapters.lana[i])
continue if win32wnet.Netbios(ncb) != 0
ncb.Reset()
ncb.Command = netbios.NCBASTAT
ncb.Lana_num = ord(adapters.lana[i])
ncb.Callname = '*'.ljust(16)
ncb.Buffer = status = netbios.ADAPTER_STATUS()
continue if win32wnet.Netbios(ncb) != 0
status._unpack()
bytes = status.adapter_address[:6]
continue if len(bytes) != 6
return int.from_bytes(bytes, 'big')
This example indicates that there are non-trivial use cases where
``continue if`` also improves the readability of the loop code.
It is probably significant to note that all of the examples selected
for this PEP were found by grepping the standard library for ``while
True`` and ``continue``, and the relevant examples were found in
the first four modules inspected.
Copyright
2017-09-05 14:37:37 -04:00
=========
2017-09-05 14:54:46 -04:00
This document is placed in the public domain.