configuration values are not properly escaped, with hash signs being treated as comments

A simple test case for this problem:

    >>> from bzrlib.branch import Branch
    >>> b = Branch.open('.')
    >>> config = b.get_config()
    >>> config.set_user_option('foo', 'abc#def')
    >>> config.get_user_option('foo')
    u'abc'

I can reproduce this with bzr-1.0 and bzr.dev r3211. In both cases, the following gets written to the configuration file:

    foo = abc#def

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

James Henstridge пишет:
| ** Summary changed:
|
| - locations.conf gets rewritten, dropping necessary quotes
| + configuration values are not properly escaped, with hash signs being treated as comments
|
| ** Changed in: bzr
| Status: New => Confirmed
|

It's actually problem in ConfigObj library. I talked with Michael Foord about
this issue and this bug is fixed in upstream now.

Here is quote from our conversation:

- -------- Исходное сообщение --------
Тема: Re: ConfigObj: inline comments considered harmful
Дата: Tue, 15 Jan 2008 23:13:15 +0000
От: Michael Foord <email address hidden>
Кому: Alexander Belchenko <email address hidden>

Hello Alexander,

This problem is now fixed in SVN.

http://svn.pythonutils.python-hosting.com/trunk/pythonutils/

I'm making some further changes (minor bugfixes and possibly a couple of
new methods 'reload' and 'restore_defaults'). After that a new release
will follow shortly.

Many Thanks

Michael Foord

Alexander Belchenko wrote:
| Michael Foord пишет:
|> <email address hidden> wrote:
|>> Some one has sent you a message via your contact form. They filled
|>> in the
|>> following values :
|>>
|>> comment = We are using ConfigObj in Bazaar project and many bzr
|>> plugins. Recently I discover that your support of inline comments
|>> can create problems with write and then read back string with '#'
|>> sign in it.
|>>
|>> E.g. if I save option from my program, e.g.:
|>>
|>> message = bug #123 fixed
|>>
|>> and then I tru to read back config file, I get only 'bug' as value
|>> of 'message' option.
|>>
|>> It's bad. Any ideas why ConfigObj works in this way?
|>>
|> In order to allow inline comments that aren't part of the value.
|>
|>> Please, don't tell me that we must escape # signs when we write them
|>> to config, and unescape when read back. It's ugly.
|>>
|>
|> You can probably do:
|>
|> message = "bug #123 fixed"
|
| How this possible to do using ConfigObj API?
|

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFHpqD+zYr338mxwCURAsITAJ4ww/TRVzajjwaMPxsV94MGKHu/1gCeKTXe
JNrxivCkJm8OstiwGCmQSCY=
=CIEw
-----END PGP SIGNATURE-----

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

James Henstridge пишет:
| A simple test case for this problem:
|
| >>> from bzrlib.branch import Branch
| >>> b = Branch.open('.')
| >>> config = b.get_config()
| >>> config.set_user_option('foo', 'abc#def')
| >>> config.get_user_option('foo')
| u'abc'
|
| I can reproduce this with bzr-1.0 and bzr.dev r3211. In both cases, the
| following gets written to the configuration file:
|
| foo = abc#def
|

Today new version of ConfigObj 4.5.1 is released with fix for this issue.
We can update ConfigObj instance in bzr.dev.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFHq0kHzYr338mxwCURAsZAAJ9aAVio71qjLnILqVOBrCfclcl9DwCfeNgB
b+vvGdIYSsBj4zsswGw50fY=
=JS0+
-----END PGP SIGNATURE-----