holland-discuss team mailing list archive

[Andrew Garner <muzazzi@xxxxxxxxx>]

I haven't had a lot of time this week to look at this in detail.  But
here are some inline comments.

On Thu, Sep 17, 2015 at 6:09 PM, Tim Soderstrom
<tim@xxxxxxxxxxxxxxxxxxxxx> wrote:
> I still haven't figured out the annoying lists quote issue noted below,

subprocess.list2cmdline is pretty basic and for complex stuff the
output is pretty bad.  I don't think it was really designed to handle
complex bits.  Per the subprocess module documentation "list2cmdline()
is designed for applications using the same rules as the MS C runtime"

I use one of three approach to deal with it:

- (most common) ignore it and treat the commandline output from
list2cmdline as "close enough" :)
- rely on some third party library that handles quoting args a little
better, like sarge
- roll my own quoting function.

You can look at the sarge.shell_quote implementation which is
relatively simple and see how it handles it:

https://bitbucket.org/vinay.sajip/sarge/src/1408b253c82da3dcaaeb7b1704a5e08d895dce34/sarge/__init__.py?at=default&fileviewer=file-view-default#__init__.py-68

To emulate list2cmdline I sometime do something like:

' '.join(map(sarge.shell_quote, args))

Which might give output like this:

rsync -avx --link-dest=/var/spool/holland/ssh/20150915_175342
--exclude=Private --exclude=.cache --exclude=Steam
--exclude=.thunderbird --exclude=.mozilla '--rsh=ssh -p' --bwlimit=1k
--one-file-system filedawg.moocow.home://home/tim/tmp
/var/spool/holland/ssh/20150915_175354

> I also disabled backup size estimation since that was proving to be really
> expensive, and could be inaccurate when using --one-file-system, not to
> mention won't work at all for remote sources (rsync's --dry-run could be
> used for that case). So for now it just returns 0, which is fugly. Need a
> more elegant solution I think.

Yeah.  I think the backup size estimation is one of the main
misfeatures in holland.   It can be really expensive and is almost
always inaccurate and leads to false failures or sometimes
underestimates.  I wouldn't feel too bad about always returning 0 and
just documenting the behavior.