| Re: [orbit-dev] Multiple bundle recipes for httpclient-* vs. a single one for httpclient-osgi? |
Hi Roland, > So this comes down to basically whether we bring 2 extra bundles > (httpclient-cache, httpcomponents-fluent), or whether we simply refer to > the httpclient-osgi (containing both) one in maven central and switch our > existing orbit-recipes generated bundle to use that GAV. exactly. Both approaches carry some risk of breakage: - Two extra bundles increase the risk of wiring issues (missing uses directives, etc.) - Using httpclient-osgi introduces some special OSGi code which so far has not been part of an Orbit bundle (whereas httpclient-cache and httpclient-fluent were present in their 4.3.x incarnation). > I can clearly see that maven central does contain finer grained and > aggregated dependencies for this stack so my argument about the finer > grained approach being more popular isn't right. I don't understand this sentence, sorry. > I suppose we need to file a separate CQ for httpclient-osgi since it > may container additional things on top of just httpclient-cache, and > httpclient-fluent ? Feel free to do so and when it's approved we can > just change the GAV being pulled in by orbit-recipes, and address the > osgi.bnd file with any changes. This will be targeted for M6 which > will be the last chance to make such changes. I just did a diff between the contents of org.apache.httpcomponents.httpclient_4.5.2.v20161115-1643.jar currently in Orbit and httpclient-osgi-4.5.2.jar from Maven Central. The latter contains indeed a bit more code: - A shaded org.apache.httpcomponents:httpclient-cache:4.5.2 - A shaded org.apache.httpcomponents:httpclient-mime:4.5.2 - A shaded org.apache.httpcomponents:fluent-hc:4.5.2 All this is expected and shouldn't cause any problems. But then there is also - A shared commons-codec:commons-codec:1.9 That this is shaded rather than an imported is odd. (Also, we would need a CQ for this, as Orbit contains only 1.6.0 of commons-codec.) And then there is this addition in httpclient-osgi: - The org.apache.http.osgi.* packages and an OSGI-INF/metatype directory I am not sure whether these would do any harm. I think I will try putting together recipes for both approaches, build AERI with either and see whether anything breaks when installed into M5. Does this sound reasonable? Best wishes, Andreas -- Codetrails GmbH The knowledge transfer company Robert-Bosch-Str. 7, 64293 Darmstadt Phone: +49-6151-276-7092 Mobile: +49-170-811-3791 http://www.codetrails.com/ Managing Director: Dr. Marcel Bruch Handelsregister: Darmstadt HRB 91940
Attachment:
signature.asc
Description: OpenPGP digital signature