bzr 2.1.0 on windows: bzr eats backslashes in command-line arguments

[Alexander Belchenko]

C:\Temp\1>bzr ci -m "move to \\ACAD"
Committing to: C:/Temp/1/
added bar
Committed revision 2.

C:\Temp\1>bzr log -l1
------------------------------------------------------------
revno: 2
committer: Alexander Belchenko <email address hidden>
branch nick: 1
timestamp: Sat 2010-02-27 13:28:14 +0200
message:
  move to \ACAD

NOTE: in log message there should be 2 backslashes, but there is only one!

I discover this while doing commit with qcommit and my first guess was: something wrong with Qt. But then I've managed to reproduce the problem with plain bzr. I think there is bug in new windows-specific command-line parser. For me it's a critical regression.

[Alexander Belchenko (bialix)]

Corresponding part of .bzr.log:

0.219 bazaar version: 2.1.0
0.219 bzr arguments: [u'ci', u'-m', u'move to \\ACAD']
0.266 looking for plugins in C:\work\Bazaar\plugins
0.484 looking for plugins in C:/Program Files/Bazaar/plugins
0.578 encoding stdout as sys.stdout encoding 'cp866'
0.641 opening working tree 'C:/Temp/1'
0.687 preparing to commit
[ 1792] 2010-02-27 13:28:14.953 INFO: Committing to: C:/Temp/1/
0.719 Selecting files for commit with filter None
[ 1792] 2010-02-27 13:28:14.953 INFO: added bar
[ 1792] 2010-02-27 13:28:15.171 INFO: Committed revision 2.
0.937 Transferred: 0KiB (0.0K/s r:0K w:0K)
0.953 return code 0

[Alexander Belchenko (bialix)]

To avoid your guesses, plain python works fine here:

C:\Temp\1>python -c "import sys; print sys.argv" "move to \\ACAD"
['-c', 'move to \\\\ACAD']

[Gordon Tyler (doxxx)]

This may be a result of the new cmdline module. It tries to replicate the backslash parsing behaviour of CommandLinetoArgv Win32 API function which in some cases will reduce 2N backslashes to N slashes.

[Alexander Belchenko (bialix)]

Gordon Tyler пишет:
> This may be a result of the new cmdline module. It tries to replicate
> the backslash parsing behaviour of CommandLinetoArgv Win32 API function
> which in some cases will reduce 2N backslashes to N slashes.

I think this is completely wrong.

[Gordon Tyler (doxxx)]

I cannot reproduce this using current bzr.dev (r5059). I only get it using bzr 2.1.0, but the cmdline stuff was only merged ito bzr.dev and not 2.1.0. So, I'd say that the new cmdline stuff fixed it even ;)

[Gordon Tyler (doxxx)]

Sat 2010-02-27 09:57:08 -0500
0.134 bazaar version: 2.2.0dev1
0.134 bzr arguments: [u'commit', u'-m', u'move to \\\\acad']

[Alexander Belchenko (bialix)]

Gordon Tyler пишет:
> I cannot reproduce this using current bzr.dev (r5059). I only get it
> using bzr 2.1.0, but the cmdline stuff was only merged ito bzr.dev and
> not 2.1.0. So, I'd say that the new cmdline stuff fixed it even ;)

So then it should be backported.

[Gordon Tyler (doxxx)]

On 2/27/2010 9:59 AM, Alexander Belchenko wrote:
> Gordon Tyler пишет:
>> This may be a result of the new cmdline module. It tries to replicate
>> the backslash parsing behaviour of CommandLinetoArgv Win32 API function
>> which in some cases will reduce 2N backslashes to N slashes.
>
> I think this is completely wrong.

It's in the handling of backslashes followed by a quote.

See: http://msdn.microsoft.com/en-us/library/bb776391(VS.85).aspx

[Alexander Belchenko (bialix)]

OK, I see. But in my case following rule is apply:

n backslashes not followed by a quotation mark simply produce n backslashes.

Anyway, it seems I have to address other cases when 2n backslashes becomes n backslashes, otherwise our qsubprocess becomes unhappy.

[Alexander Belchenko (bialix)]

I've added workaround in QBzr to make qsubprocess more robust against such kind of errors. But the root cause in bzr core is not fixed yet, and should be fixed as well.

[Perry (perryrapp)]

I said I'd test 2.2b2 and follow-up on my bug (https://bugs.launchpad.net/bzr/+bug/569050) which was marked as duplicate to this, so I'm following up here.

Unfortunately, I was unable to test 2.2b2, because the install failed, so I do not have a data point for 2.2b2. I'll post the failed install as a separate bug.

Is 2.2b2 expected to fix this?

[Alexander Belchenko (bialix)]

Perry пишет:
> Is 2.2b2 expected to fix this?

AFAIK, it should. But I'm not sure. :-/

--

All the dude wanted was his rug back

[Martin Pool (mbp)]

I'm told this is fixed in trunk but needs to be merged back to 2.1. But it can't directly be merged back because it needs another fix.