Title: argparse assertion failure with brackets in metavars
Type: behavior Stage: patch review
Components: Library (Lib) Versions: Python 3.7, Python 3.6, Python 3.5, Python 3.4, Python 2.7
Status: open Resolution:
Dependencies: Superseder:
Assigned To: Nosy List: Zacrath, bethard, htnieman, manveru, mattpr, merwok, paul.j3, spaceone, vajrasky, wim.glenn, wolma, xuanji, ysj.ray
Priority: high Keywords: patch

Created on 2011-04-19 08:20 by htnieman, last changed 2017-09-14 20:27 by wim.glenn.

File name Uploaded Description Edit
issue_11874.diff ysj.ray, 2011-04-29 09:06
unit_test_for_inner_bracket_metavar.txt vajrasky, 2013-07-14 03:44 Unit test review
unit_test_argparse.txt vajrasky, 2013-07-14 09:42 Revamped unit test review
fix_and_unit_test_for_argparse_inner_bracket_bug.txt vajrasky, 2013-07-14 10:58 Fix and unit test for the argparse inner bracket metavar bug review
format_usage.patch paul.j3, 2014-04-24 22:12 review
Pull Requests
URL Status Linked Edit
PR 1826 open wim.glenn, 2017-05-27 00:54
Messages (18)
msg134021 - (view) Author: Hartmut Niemann (htnieman) Date: 2011-04-19 08:20
I run this script with option --help
import argparse
parser = argparse.ArgumentParser()
parser.add_argument ('--from-itm', dest="from_itm",action="store",default=None,help ="Source ITM file", metavar="ITMFILENAME")
parser.add_argument ("--from-fbm-sigtab",dest="from_sigtab",action="store",default=None,help="Source signal FB", metavar=" [LIB,]FBNAME")
parser.add_argument ("--c",dest="f",metavar="x]d")
args = parser.parse_args()
and I get an assertion failure:
D:\temp> --help
Traceback (most recent call last):
  File "D:\temp\", line 6, in <module>
    args = parser.parse_args()
  File "C:\Python27\lib\", line 1678, in parse_args
    args, argv = self.parse_known_args(args, namespace)
  File "C:\Python27\lib\", line 1710, in parse_known_args
    namespace, args = self._parse_known_args(args, namespace)
  File "C:\Python27\lib\", line 1916, in _parse_known_args
    start_index = consume_optional(start_index)
  File "C:\Python27\lib\", line 1856, in consume_optional
    take_action(action, args, option_string)
  File "C:\Python27\lib\", line 1784, in take_action
    action(self, namespace, argument_values, option_string)
  File "C:\Python27\lib\", line 993, in __call__
  File "C:\Python27\lib\", line 2303, in print_help
    self._print_message(self.format_help(), file)
  File "C:\Python27\lib\", line 2277, in format_help
    return formatter.format_help()
  File "C:\Python27\lib\", line 278, in format_help
    help = self._root_section.format_help()
  File "C:\Python27\lib\", line 208, in format_help
  File "C:\Python27\lib\", line 329, in _format_usage
    assert ' '.join(opt_parts) == opt_usage
It must be the combination of several argument lines because when I 
comment out one of the add_arguments, it works.

Is this a bug or do I do things I shouln't?
If it is not allowed to use [] in metavars (what I wanted was metavar='[optional,]mantatory'), I' like to receive a readable error message.

With best regards
Hartmut Niemann
msg134558 - (view) Author: Manveru (manveru) Date: 2011-04-27 12:41
I was a victim of the same issue, but only after my usage list does not fit in one line. Please fix!
msg134741 - (view) Author: ysj.ray (ysj.ray) Date: 2011-04-29 09:06
Seem as a problem in optparse.HelpFormatter._format_usage(): when the generated usage string is too long(longer than 78, e.g.), python tries to break the usage string into parts at some proper positions and group them to multiple lines, then join the parts with space and assert it equal with original usage string. The regular expression used to break the usage string into parts is:
So when there is '[]' in usage string, like in the example(there is '[]' in metavar), the assertion fails.

metavar="[LIB,]FBNAME" seems quite reasonable and usable, in the case that one want to define an option who has tow values and the first is optional. So I suggest '[]' should be allowed in metavar, and the _format_usage() needs fixed.

Here is a patch that fixes by changing the way breaking the usage string to parts and remove joining the parts and assert it equal to original usage string. Now the usage string is broken into intact each action of group usage strings. I think this breaking granularity is enough.
msg149547 - (view) Author: Steven Bethard (bethard) * (Python committer) Date: 2011-12-15 12:42
I agree this is a bug. The patch needs to add unit tests that make sure metavars with [] work as expected.
msg193021 - (view) Author: Vajrasky Kok (vajrasky) * Date: 2013-07-14 03:37
Attached the unit test.

Here is the simpler script to product the bug:

import argparse
parser = argparse.ArgumentParser(prog='PROG')
parser.add_argument ('--a', metavar='a'*76)
parser.add_argument ('--b', metavar="[innerpart]outerpart")
parser.add_argument ('c', metavar='c'*76)
parser.add_argument ('d', metavar="[innerpart2]outerpart2")
args = parser.parse_args()

python --help
msg193022 - (view) Author: Vajrasky Kok (vajrasky) * Date: 2013-07-14 03:44
Sorry, got typo in last unit test.
msg193042 - (view) Author: Vajrasky Kok (vajrasky) * Date: 2013-07-14 09:42
Tidied up the unit test. Mixed with arguments without metavar.
msg193046 - (view) Author: Vajrasky Kok (vajrasky) * Date: 2013-07-14 10:58
Attached the preliminary fix and the unit test for this ticket.
msg193068 - (view) Author: paul j3 (paul.j3) * Date: 2013-07-14 21:41
If the arg_parts are passed through the same cleanup as the 'text' (and empty strings removed), then 

    text = ' '.join(arg_parts)

In that case there would be no need to return both (text, arg_parts).

Parenthesis in the metavar could also create the problem addressed in this thread, except as noted in that 'text' cleanup removes them.

nargs='*' or '+' or integer is another way in which [] could be introduced into the metavar.
msg193072 - (view) Author: paul j3 (paul.j3) * Date: 2013-07-15 03:58
I just filed a patch with (argparse add_mutually_exclusive_group should accept existing arguments to register conflicts) that includes the latest patch from this issue.  I tweaked it so _format_actions_usage only returns arg_parts.  The block of _re. statements (# clean up separators for mutually exclusive groups) are in a nested function so it can be applied to each of the parts.

In that issue I wrote a custom formatter that handles groups even if they share actions, or the action order does not match definition order.  For that it is easier to work with the arg_parts as generated here rather than the whole usage line.
msg193112 - (view) Author: Vajrasky Kok (vajrasky) * Date: 2013-07-15 15:45
paul j3, thanks for reviewing my patch and giving me credit in your patch for another ticket.

Yeah, as you could see, the reason I return arg_parts and text is because the text still needs to undergo the cleanup process. You solved it by putting cleaning up in inner function.

I am thinking whether it is best to do "assert ' '.join(opt_parts) == opt_usage" inside _format_actions_usage helper function.

In that way, we can avoid returning the text. We can return only the arg_parts.

Anyway, my patch still got some unused variables, notably part_regexp and inner. My bad.

Let me check the code more deeply. See whether I can architect my patch in a better way. Maybe we can avoid building separate list inside _format_actions_usage.

Beside of that, this bug is not introduced solely by bracket character. It needs another non-space character on the right side of it.

This line is fine:
parser.add_argument ('--b', metavar="[innerpart] outerpart")

This line will fail the assertion:
parser.add_argument ('--b', metavar="[innerpart]outerpart")
msg193187 - (view) Author: paul j3 (paul.j3) * Date: 2013-07-16 20:26
Here's a patch that takes a different approach - rewrite _format_actions_usage() so it formats groups (and unattached actions) directly.  

It uses the same logic to determine when to format a group, but then it calls _format_group_usage() to format the group, without the error prone insert and cleanup business.  And by keeping a list of the parts, it avoids the complications of parsing the text.  The only cleanup regex removes () from single element groups.

In the nested get_lines function, I added 'line and' to 

   if line and line_len + 1 + len(part) > text_width:

so it does not create a blank line before an extra long entry (e.g. 'c'*76).

I'm using the _format_group_usage() and _format_just_actions_usage() in the Issue 10984 MultiGroupFormatter, just rearranging the logic in its _format_actions_usage() method.  That patch could be rewritten to depend on this one.

This patch also fixes, since the metavar is never parsed.
msg216812 - (view) Author: paul j3 (paul.j3) * Date: 2014-04-18 22:09
Another example of code hitting this AssertionError.  Here the problem was a space in the option argument, '--out '.

'parser.add_argument('-o', '--out ', help='b', required = True)'

That space would have cause problems later when trying access the 'args' attributes.  But producing the error during the help formatting didn't help.
msg230187 - (view) Author: Wolfgang Maier (wolma) * Date: 2014-10-28 23:16
It doesn't seem to be the exact same problem, but still this very simple example with parentheses in metavar fails to format correctly:

import argparse
parser = argparse.ArgumentParser()
parser.add_argument("inputfiles", metavar = 'input_file(s)', nargs = "+", help = 'one or more input files')

args = parser.parse_args()


usage: -m [-h] input_files) [input_file(s ...]

positional arguments:
  input_file(s)  one or more input files

optional arguments:
  -h, --help     show this help message and exit

with the two occurences of input_file(s) being formatted wrong, but differently.
msg230198 - (view) Author: paul j3 (paul.j3) * Date: 2014-10-29 05:30
The patch with a major rewrite of '_format_actions_usage' handles this usage problem correctly:

    usage: [-h] input_file(s) [input_file(s) ...]

The error is the result of usage group formatter trying to remove excess `()`.  To give an idea of what is happening I replaced the missing () with {}.

    input_file(s) [input_file(s) ...]
    input_file{s) [input_file(s} ...]
msg262105 - (view) Author: Matt Pr (mattpr) Date: 2016-03-21 08:33
Same AssertionError is also caused by having certain "choices".

Python 2.7.10

global_options.add_argument('--field-sep', choices=[',',';','|','\t'], required=True, help='Field separator that separates columns in a row.')

Removing required=True or removing the tab (\t) from the options both work around this usage formatter issue for me.

I know this is an old issue but figured I would add another repro case to help whoever might work on this.
msg294575 - (view) Author: paul j3 (paul.j3) * Date: 2017-05-27 02:49
Any changes here including the latest pull request might conflict with

Argparser does not display closing parentheses in nested mutex groups

If someone uses nested mutually exclusive groups (which I don't think they should, but anyways ...), the usage formatter gets confused by the nested brackets, and weeds out the wrong one(s).

That issue hasn't been settle yet; my feeling is that fixing this assertion error properly is more important.
msg295214 - (view) Author: wim glenn (wim.glenn) * Date: 2017-06-05 22:02
May I ask, how to assign reviewers for CPython pull requests?  has been sitting there for 10 days and the only comment was from a bot.
Date User Action Args
2017-09-14 20:27:01wim.glennsetversions: + Python 3.6, Python 3.7
2017-06-05 22:02:45wim.glennsetmessages: + msg295214
2017-05-30 15:44:55wim.glennsetnosy: + wim.glenn
2017-05-27 02:50:00paul.j3setmessages: + msg294575
2017-05-27 00:54:04wim.glennsetpull_requests: + pull_request1912
2016-12-12 19:17:31paul.j3setpriority: normal -> high
2016-03-21 08:33:19mattprsetnosy: + mattpr
messages: + msg262105
2016-01-17 06:32:01martin.panterlinkissue25058 superseder
2015-05-05 21:08:44spaceonesetnosy: + spaceone
2015-05-05 21:06:35ned.deilylinkissue24089 superseder
2015-05-05 21:05:04ned.deilysetstage: patch review
versions: + Python 3.4, Python 3.5, - Python 3.2, Python 3.3
2014-10-30 06:27:45tshepangsetnosy: - tshepang
2014-10-29 05:30:38paul.j3setmessages: + msg230198
2014-10-28 23:16:39wolmasetnosy: + wolma
messages: + msg230187
2014-09-09 07:40:53Zacrathsetnosy: + Zacrath
2014-04-24 22:12:27paul.j3setfiles: - format_usage.patch
2014-04-24 22:12:12paul.j3setfiles: + format_usage.patch
2014-04-18 22:09:02paul.j3setmessages: + msg216812
2013-07-16 20:26:25paul.j3setfiles: + format_usage.patch

messages: + msg193187
2013-07-15 15:45:27vajraskysetmessages: + msg193112
2013-07-15 03:58:39paul.j3setmessages: + msg193072
2013-07-14 21:41:40paul.j3setmessages: + msg193068
2013-07-14 10:58:15vajraskysetfiles: + fix_and_unit_test_for_argparse_inner_bracket_bug.txt

messages: + msg193046
2013-07-14 09:42:30vajraskysetfiles: + unit_test_argparse.txt

messages: + msg193042
2013-07-14 03:44:13vajraskysetfiles: + unit_test_for_inner_bracket_metavar.txt

messages: + msg193022
2013-07-14 03:43:48vajraskysetfiles: - unit_test_for_inner_bracket_metavar.txt
2013-07-14 03:37:45vajraskysetfiles: + unit_test_for_inner_bracket_metavar.txt
nosy: + vajrasky
messages: + msg193021

2013-07-13 23:22:37paul.j3setnosy: + paul.j3
2012-03-18 22:41:11tshepangsetnosy: + tshepang
2012-02-25 10:33:25bethardlinkissue14046 superseder
2011-12-15 12:42:30bethardsetmessages: + msg149547
2011-04-29 09:14:52ysj.raysetcomponents: + Library (Lib), - None
2011-04-29 09:06:43ysj.raysetfiles: + issue_11874.diff

nosy: + ysj.ray
messages: + msg134741

keywords: + patch
2011-04-27 15:28:16xuanjisetnosy: + xuanji
2011-04-27 12:41:50manverusetnosy: + manveru
messages: + msg134558
2011-04-22 16:53:49merwoksetnosy: + bethard, merwok

versions: + Python 3.2, Python 3.3
2011-04-19 08:20:50htniemancreate