I was trying to run SC on Equinox instead of Felix last night just to
see how they compare so I discovered the FileAccessService and
HistoryService use org.apache.felix.servicebinder. Anyway, the
question isn't about Equinox, it was just the reason why saw the
dependency in the code.
Since there are just two services which use it and all the other
services don't, I thought it wouldn't be a bad idea to remove the
dependency on servicebinder in version 1.0. I've already removed it in
my local tree and I've also cleaned up the HistoryService. I cannot
say how it affects the performance because I'm already in the middle
of delaying the ConfigurationForm implementations (but I'll run a
comparison of the combined changes to trunk once I'm done).
I know the plan in the long term may be to go in the direction of
servicebinder. But Emil talked about an alternative to servicebinder
which was supposed to have more advantages. Besides, only using it for
1.0 in two of the numerous services we have doesn't much sound like
the price to pay for the distribution size increase, the runtime
overhead because of its loading, the XML parsing it performs, etc.
What I'm trying to say is that it may be a good idea but we should
rather evaluate the current alternatives and think of a way to apply
the chosen one to the whole tree (probably after 1.0 in order to get
it properly tested).
What do you think?