I understand
you wanted this in the platform because you're using it
already.
...no, I am not so concerned about having it in platform for using it. I
now trust my module more than the existing implementation. I would like to
have it in platform to encourage others to look at it, use it, adopt it and
contribute to it, since I am really confident it is really better than the
current implementation.
I just want to be sure we're going to
maintain it since I can see it'll
require a lot more work before being to full switch to this implementation.
IMO, there is even more work to do to use the current right service, since
its own test cover only 52% of the right service itself and the
documentation is really sparse and almost unwritable ! (BTW, these tests
cannot be used on our bridge since these use implementation method that are
not published in the interface.)
Thanks
-Vincent
> Thanks
> -Vincent
>
> On Feb 15, 2012, at 11:17 AM, Denis Gervalle wrote:
>
>> Hi all,
>>
>> As you have probably notice, I have recently committed an
>> feature-security-authorization branch on platform. I am working on this
> for
>> a while now and it was the first step to share the outcome of this
large
>> refactoring of the initial work done
early last year by Andreas. Since
> the
>> code was not quality compliant with platform but the general structure
>> Andreas has build seems to me well appropriate, I have progressively
>> refactor its code to better fit our real needs. Here is what I have
been
>> done:
>>
>> 1) Split in to module api and bridge to allow breaking the currently
>> unavoidable dependency on oldcore. Now only bridge depends on oldcore,
> and
>> the api does not depends on bridge. As mush as possible has been
written
> in
>> the api (still some code to migrate), and some temporary internal
bridge
>> are used to access oldcore stuffs since
augmenting the existing
>> document-bridge does not seems appropriate IMO.
>>
>> 2) The initial enumeration of rights as been replaced by a Right class,
>> which could be seen has a pseudo enum, but could be augmented with new
>> rights. To register a new Right, you have to provide a RightDescription
> to
>> the AuthorizationManager. The description will define the default
state,
>> the tie resolution policy, the
inheritance policy, the list of entity
> types
>> for which the right is applicable, the implied rights and if the right
>> could be allowed in read-only mode. So new defined Right will benefit
the
>> whole logic of the AuthorizationManager
and currently existing one
could
> be
>> declaratively defined.
>>
>> 3) Large renaming to better distinguish stuffs, clarify comments and
>> prepare for future. I have voluntarily not taken existing names to
> clearly
>> split the old and the new api. In brief, the new right service is now
> named
>> AuthorizationManager. Internally, it manipulates SecurityReference (as
> well
>> as UserSecurityReference and GroupSecurityReference, to represent
> entities,
>> user and group), SecurityRule (representing a right object) and
>> SecurityAccess (representing an access level in the old nomenclature),
>> which are store in a SecurityCache using SecurityRuleEntry (a set of
> rules)
>> and SecurityAccessEntry (the access of a given user). The
>> AuthorizationManager delegate cache management to a SecurityCacheLoader
>> which loads rules using a SecurityRuleLoader ; and delegate itself the
>> computing of the access for a given user and a set of rules to an
>> AuthorizationSettler. This last one could be overridden to provide
> specific
>> decision that could not be done in declarative mode.
>>
>> 4) Refactoring was necessary to improve consistency and reduce
> complexity,
>> and simplify as much as possible; while extending the limitations to
> allow
>> more rights to be registered. This work has been a little bit opposed
to
>> the optimization done by Andreas, in
particular on memory usage. But
>> optimization is often the enemy of clean code.
>>
>> 5) Improvement were necessary to better mimic the existing
implementation
>> in some peculiar but necessary rules to
stay compatible with current
>> working wiki. I tend to reduce as much as possible what is not done
>> declaratively, but there are still some special cases, like delete for
>> creators, deny for other user on explicit allow and admin for wiki
owner
>> that are settle by the authorization
settler. My implementation should
be
>> almost compatible with the old one,
except for groups that are
currently
>> not checked from the entity wiki, but
only from the user wiki. This
needs
>> some more refactoring for which I feel
inconfortable with, some I'd
like
> to
>> share first.
>>
>> 6) The AuthorizationManager interface has been simplified, providing 2
>> methods for either checking or verifying an access right (the checking
>> methods throws while the verifying one return a boolean), and one to
>> register a new right.
>>
>> The existing RightService could be bridged on the new implementation
> using
>> the XWikiCachingRightService class in xwiki.cfg and the new API could
be
>> used side-by-side with the old
implementation as well. What should
still
>> really need to be improved is the unit
testing, currently some tests
are
>> still awful and incomplete. I already
refactor some of them, to
provide a
>> better coverage of essential part of the
code: the security cache and
the
>> default authorization settler. Obviously,
any help is welcomed.
>>
>> Since I already have an existing wiki using this implementation
>> successfully and using it for creating new rights for extensions, I
would
>> like to merge this new implementation as
experimental in platform to
have
>> it available for anyone who need it or
want to test it, and for you to
> use
>> in your new experimental development as well. Providing it in platform
> will
>> encourage it to be finalized and replace the existing implementation.
>>
>> Here is my +1 for the merge on 4.x,
>>
>> WDYT ?
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs