Re: [equinox-dev] [prov] Some provisioning changes released

[John Arthorne <John_Arthorne@xxxxxxxxxx>]

Yes, that's the fix. I released this
to CVS already, but didn't get back to the mailing list because Pascal
and I got caught up in discussions.  You'll need one more fix that
Pascal released to InstallableUnit about an hour ago. With all the latest
from HEAD, the metadata generation and install tests pass.

David R Stevenson <dstevens@xxxxxxxxxx>
Sent by: equinox-dev-bounces@xxxxxxxxxxx

06/15/2007 12:24 PM

Please respond to
Equinox development mailing list <equinox-dev@xxxxxxxxxxx>

To	Equinox development mailing list <equinox-dev@xxxxxxxxxxx>
cc	Equinox development mailing list <equinox-dev@xxxxxxxxxxx>, equinox-dev-bounces@xxxxxxxxxxx
Subject	Re: [equinox-dev] [prov] Some provisioning changes released

Thanks, Tom. And thanks for the reference to the LDAP filter spec.

The bug is in the mergeFilters method of the Transformation class in prov.resolution:
 
       private
String mergeFilters(String filter1, String filter2) {
                if
(filter1 == null
&& filter2 == null)
                    
   return
null;
                if
(filter1 == null)
                    
   return
filter2; 
                if
(filter2 == null)
                    
   return
filter1; 
                return
"(&("
+ filter1 + ")
(" + filter2 + "))";
// bogus! 
        } 

replacing the bogus line by 

                return
"(&"
+ filter1 + filter2 + ")";

gets past the errors I was seeing. 

Since the two string args are apparently supposed to be filters (I assume
well-formed) and the definition of a filter is 

        <filter> ::= '(' <filtercomp> ')'

(ie. always parenthesized) then I believe the replacement should be a correct
solution. 

If there are cases where we pass malformed filter conditions like "os=foo"
"arch=bar", then the fix has to be more forgiving and the missing
parentheses. 

        - Dave

Thomas Watson/Austin/IBM@IBMUS
Sent by: equinox-dev-bounces@xxxxxxxxxxx

06/15/2007 06:31 AM

Please respond to
Equinox development mailing list <equinox-dev@xxxxxxxxxxx>

To	Equinox development mailing list <equinox-dev@xxxxxxxxxxx>
cc
Subject	Re: [equinox-dev] [prov] Some provisioning changes released

This is not a bug in FilterImpl 

This is not a valid filter according to rfc 1960 http://www.faqs.org/rfcs/rfc1960.html:
(&((&(osgi.os=win32)(osgi.ws=win32)(osgi.arch=x86))) ((version>=3.3.0.v3339)))

There are too many nested (()) parentheses.  A '(' char must be followed
by an operation char ('&' | '|' | '!') or an evaluation item e.g (osgi.os=win32).
 The filter should be changed to the following to be valid:

(&(&(osgi.os=win32)(osgi.ws=win32)(osgi.arch=x86)) (version>=3.3.0.v3339))


Tom

David R Stevenson/Redmond/IBM@IBMUS
Sent by: equinox-dev-bounces@xxxxxxxxxxx

06/14/2007 07:41 PM

Please respond to
Equinox development mailing list <equinox-dev@xxxxxxxxxxx>

To	Equinox development mailing list <equinox-dev@xxxxxxxxxxx>
cc	equinox-dev@xxxxxxxxxxx, equinox-dev-bounces@xxxxxxxxxxx
Subject	Re: [equinox-dev] [prov] Some provisioning changes released

Hmm. I picked up your changes and Pascal's. I regenerated the metadata.
When I attempt to install (an SDK) I get  a number of InvalidSyntaxExceptions
in FilterImpl$Parser: 

FilterImpl$Parser.parse_attr() line: 1627        

                    
 if
(length == 0) { 
                    
         throw
new
InvalidSyntaxException(NLS.bind(Msg.FILTER_MISSING_ATTR,
String.valueOf(pos)),
filterstring);
                    
 } 

when,  for example, the filter string is 

 (&((&(osgi.os=win32)(osgi.ws=win32)(osgi.arch=x86))) ((version>=3.3.0.v3339)))

and the message is 

      Missing
attr at character 3 

and the iu being processed is 

       org.eclipse.rcp.featureIU
3.3.0.v20070405a-8w8eDpEbsN3KkG_gP8JZLGb 

similar for platform.featureIU with filter string 

      (&((&(osgi.ws=win32))) ((version>=3.2.100.I20070319-0010)))

The problem appears to be with the handling of the compound conjunction.
A version filter only or an _expression_ like 

      (&(osgi.os=hpux)(osgi.arch=ia64_32))

parses without exception. 

Bugzilla reveals 3 bugs matching FilterImpl, none for this behavior. Is
it a bug or is this type of _expression_ not supported.

      - Dave

John Arthorne <John_Arthorne@xxxxxxxxxx>
Sent by: equinox-dev-bounces@xxxxxxxxxxx

06/14/2007 01:45 PM

Please respond to
Equinox development mailing list <equinox-dev@xxxxxxxxxxx>

To	equinox-dev@xxxxxxxxxxx
cc
Subject	[equinox-dev] [prov] Some provisioning changes released

I released a couple of small to the provisioning incubator:

- New local metadata repository. If you create a metadata repository pointing
to a local file system directory, it will traverse the directory in search
of "content.xml" files, and aggregate them together. This makes
it easier to cobble together metadata from various places without creating
multiple metadata repositories 

- To allow for experimenting with groups that define filters, I added some
bits to the feature parser so that features can define extra filters on
their imports and includes. If you run a feature with filters through the
metadata generatory, they get turned into filters on the RequiredCapability.
This was simpler than hand-crafting filter-enabled groups manually._______________________________________________
equinox-dev mailing list
equinox-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/equinox-dev
_______________________________________________
equinox-dev mailing list
equinox-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/equinox-dev
_______________________________________________
equinox-dev mailing list
equinox-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/equinox-dev
_______________________________________________
equinox-dev mailing list
equinox-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/equinox-dev