[Rpm-maint] PATCH: Accessing file and policy lists from python (and C)

[Panu Matilainen]

On 09/21/2012 08:13 PM, Alec Leamas wrote:
> On 09/21/2012 01:42 PM, Panu Matilainen wrote:
>> On 09/21/2012 01:59 PM, Alec Leamas wrote:
>>> On 09/21/2012 07:02 AM, Panu Matilainen wrote:
>>>> On 09/15/2012 05:47 PM, Alec Leamas wrote:
>>>>> When rpm-build parses the specfile, it also parses the file list and
>>>>> (nowadays?) also the policy. These are stored in specPackage objects
>>>>> attached to the spec. However, these items are not visible on the
>>>>> rpm.specPkg interface, neither from C nor from python.
>>>>>
>>>>> Is there a particular reason these items aren't visible?
>>>>
>>>> It's simply because access to the flat, unpopulated file list doesn't
>>>> seem useful at all to me, and not exposing something means not having
>>>> to commit to an interface for it.
>>>>
>>>>> I have som pronlems with this related to the fedora-review tool [1],
>>>>> which basically forces us to use some hand-crafted specfile parsing ,
>>>>> since the file lists are needed. I would certainly prefer using
>>>>> rpmlib's
>>>>> own parsing for that. However, then external access to the fields
>>>>> fileList, fileFile and policyList is needed. The enclosed patch
>>>>> arranges
>>>>> this.
>>>>
>>>> I'm curious: what exactly are you doing / planning to do with the file
>>>> list?
>>>>
>>>>
>>> [cut, lot's of]
>>>
>>> It's fedora-review... we are  reviewing the packages, comparing the
>>> raw, unexpanded list with various guidelines sections.
>>
>> I guessed that much, but what kinds of things exactly? I'm mostly just
>> curious, but also wanting to see if there's something that rpm could
>> help with.
>
> Just an example, a snippet from current tests below. There are more of
> these, but I doubt it makes sense to paste it here.
>
> Yes, this is released  code. But the I would very much prefer to use
> rpm's parsing instead of the hand-crafted parser used today.
>
>
> _init__(self, base):
>          GenericCheckBase.__init__(self, base)
>          self.url = 'http://fedoraproject.org/wiki/' \
>                     'Packaging/Guidelines#Configuration_files'
>          self.text = '%config files are marked noreplace or the reason' \
>                      ' is justified.'
>          self.automatic = True
>          self.type = 'MUST'
>
>      def run(self):
>          rc = self.NA
>          extra = ''
>          for pkg in self.spec.packages():
>              for line in self.spec.get_files(pkg):
>                  if line.startswith('%config'):
>                      if not line.startswith('%config(noreplace)'):
>                          extra += line
>                      else:
>                          rc = self.PASS
>          self.set_passed(self.FAIL if extra else rc, extra)
>

Thanks, also found the other checks in fedora-review now, which are more 
or less what I expected to see, just wanted to be sure you didn't have 
some use-case I couldn't imagine at all.

The "problem" with these kind of checks is that they only cover very 
limited ground and will miss eg anything that goes in through "%files -f 
<manifest>" etc, which can be a lot. Anyway, what I suppose you'd 
*really* want (rather than the flat file list) is the parsed result of 
all those directory modes, config & flags etc. Which of course is 
available through the package once its built... but rpm itself doesn't 
bother parsing the file lists before the package has been built, as the 
entire file list or parts of it can be generated as a part of the build.
And this is basically why the flat file list from spec is not currently 
exported: its simply not all that useful.

That said, I dont see much harm in exporting it either, eg through 
rpmSpecGetSection().

	- Panu -