Re: [Configobj-develop] Problem with unrepr mode

[Jeffrey B. <jef...@ea...>]

On Sunday 21 September 2008 14:01:48 Michael Foord wrote:
> Jeffrey Barish wrote:
> > On Sunday 21 September 2008 13:25:27 Michael Foord wrote:
> >> Jeffrey Barish wrote:
> >>> On Sunday 21 September 2008 11:30:36 Michael Foord wrote:
> >>>> Jeffrey Barish wrote:
> >>>>> On Sunday 21 September 2008 04:17:50 Michael Foord wrote:
> >>>>>> Jeffrey Barish wrote:
> >>>>>>> On Saturday 20 September 2008 18:55:17 Jeffrey Barish wrote:
> >>>>>>>> Configobj was working fine until I switched to using unrepr mode.
> >>>>>>>> Now it works fine on one platform, but on another the same code
> >>>>>>>> produces the following error:
> >>>>>>>>
> >>>>>>>>   ...
> >>>>>>>>   File "/usr/lib/python2.5/site-packages/configobj.py", line 1272,
> >>>>>>>> in __init__ self._load(infile, configspec)
> >>>>>>>>   File "/usr/lib/python2.5/site-packages/configobj.py", line 1355,
> >>>>>>>> in _load raise error
> >>>>>>>> configobj.ConfigObjError: Parsing failed with several errors.
> >>>>>>>> First error at line 3.
> >>>>>>>>
> >>>>>>>> Line 3 is:
> >>>>>>>>
> >>>>>>>> v = 0.77304964539007093
> >>>>>>>>
> >>>>>>>> I am using version 4.5.3 on python 2.5.2.  Considering that the
> >>>>>>>> same rc file works fine on the other platform, I am at a loss as
> >>>>>>>> to what to do. Any suggestions?
> >>>>>>>
> >>>>>>> The problem appears to arise because there is no compiler module on
> >>>>>>> this platform (Nokia N800).
> >>>>>>>
> >>>>>>> Do I have any options other than returning to code like
> >>>>>>>
> >>>>>>> if x == 'True'
> >>>>>>
> >>>>>> unrepr mode relies on the compiler module - so it won't work on a
> >>>>>> platform that doesn't provide it.
> >>>>>>
> >>>>>> You could use validate and a configspec to convert types instead.
> >>>>>>
> >>>>>>
> >>>>>> Michael
> >>>>>
> >>>>> OK.  I learned how to use validate.  It works nicely for converting
> >>>>> types and I am also doing range checking, so it's a better solution
> >>>>> anyway.
> >>>>>
> >>>>> One problem, though.  In my configuration file I have a section with
> >>>>> colors, but I don't know a priori what colors it will contain.  For
> >>>>> example:
> >>>>>
> >>>>> [colors]
> >>>>> color1 = [143, 188, 143]
> >>>>> color2 = [70, 130, 180]
> >>>>>
> >>>>> and so on.  Since I don't know what the keys are going to be, I
> >>>>> presume that there is no way for me to validate the entries.  Is that
> >>>>> true?  It would be nice to be able to test that values are between 0
> >>>>> and 255 and to convert the strings to integers and the brackets to a
> >>>>> list. Currently I am doing
> >>>>>
> >>>>> color1 = 143, 188, 143
> >>>>>
> >>>>> and not validating.
> >>>>
> >>>> You're correct that ConfigObj doesn't support validating arbitrary
> >>>> entries in a section. It is a worthy feature request though (perhaps
> >>>> "* = check()").
> >>>>
> >>>> I'll add it to my list - but as I am *still* finishing a book it will
> >>>> be quicker if someone else implements it. It will need to be in the
> >>>> ConfigObj.validate method rather than in the validate module.
> >>>>
> >>>> Michael
> >>>
> >>> Here's a worse problem: after using config.reload(), all values revert
> >>> to strings.  I have to revalidate after every reload to get values back
> >>> into their correct types.
> >>
> >> Hmmm... that sounds like correct behaviour to me. Validation is an
> >> explicit step.
> >>
> >> Michael
> >
> > You make a good point.  But consider: If I provided a configspec when I
> > created the ConfigObj and I then validated the ConfigObj, would it ever
> > happen that I would not want the ConfigObj validated again when I call
> > reload?  If someone really wanted the values to revert to strings, it
> > seems to me that he would create a new ConfigObj (without a configspec)
> > rather than call reload.  I see reload as a shortcut for refreshing the
> > values using the same procedure I used initially to create them.
>
> On the other hand, validation requires a validator - and having reload
> do an implicit validate would require us to keep a reference to the
> validator used the first time. This may or may not keep a lot of memory
> in use unnecessarily or even potentially cause us to reuse an object not
> in a valid state.
>
> I see reload as returning the ConfigObj instance to the state
> immediately after loading - which would be before validation. Is calling
> validate again really such a big burden?
>
>
> Michael

It's not a problem.  Thanks for explaining the issues.
-- 
Jeffrey Barish