[RSA] Creation of EndpointDescription too strict wrt objectClass property

[bjhargrave]

Original bug ID: BZ#2715 From: Jan Willem Janssen <janwillem.janssen@luminis.eu> Reported version: R6

[bjhargrave]

Comment author: Jan Willem Janssen <janwillem.janssen@luminis.eu>

The current implementation of #verifyObjectClassProperty as defined in EndpointDescription is too strict as it only accepts arrays of string values. This while the examples in the RSA specification v1.0 use non-array values for the objectClass property (for example, see page 322 of OSGi Enterprise r5).

The #verifyObjectClassProperty method should be made more lenient when it comes to the given value: either a non-null string-value or a non-null, non-empty, array of strings should be acceptable for the objectClass property.

[bjhargrave]

Comment author: @bjhargrave

For a ServiceRegistration, objectClass is always a String[]. So at runtime objectClass must be a String[]. See Table 122.1 where objectClass is defined as a String[]. objectClass is never a String.

The example on pg 322:

is wrong and that is what should be fixed. It should be:

com.acme.Foo

to signify it is an array.

[bjhargrave]

Comment author: @bjhargrave

(In reply to BJ Hargrave from comment BZ#1)

It should be:

com.acme.Foo

to signify it is an array.

Actually, that is how to do it for a DS xml file. For the endpoint description schema, it should be:

com.acme.Foo

[bjhargrave]

Comment author: Jan Willem Janssen <janwillem.janssen@luminis.eu>

Indeed objectClass is an array of strings, I probably was confused by the convenience method on bundle context that accepts a single string as objectClass.

Nevertheless, the example on page 322 should be modified to reflect this.

[bjhargrave]

Comment author: @bosschaert

Fixed in commit da8a6fabd25a129d3f6e762726f681593662de46