Danek Duvall wrote: > On Sat, Nov 08, 2008 at 11:16:57AM -0600, Tom Mueller wrote: > > >> Danek, >> If the pkg_client_name value is now being ignored, it seems that a program >> that was working correctly before this change will not continue to work >> correctly after the change. That doesn't seem to be compatible. >> > > I talked to Brock about this -- he considered this a compatible change, but > he can reconsider. Compatibility is a moot point if I'm changing all the > consumers. > > I don't have strong feelings about this. My inclination was to keep the version listed as compatible because I'd generally feel better not changing the API incompatibly this close to release. However, my advice was based on the mistaken belief that the old clients would continue to work, perhaps not ideally. With a fresh mind this morning, I see that the old clients will break because the client name wont be set, which will (can? did?) make the history module traceback. If that wasn't the case, I would continue to argue for calling this a compatible change. Now that my mind's working clearly, I see that because this will cause the old clients to throw tracebacks, it should be marked incompatible. Sorry for the confusion.
And Danek, the change to testutils looked good to me. Brock >> To make the program work correctly after the change, a statement would >> have to be inserted that sets the pkg.client.global_settings.client_name. >> > > True. > > >> Can't the pkg_client_name argument be used to set the global_settings >> value? >> > > It could be, but it wasn't clear to me how that ought to work. If and only > if the global_settings version is None, then set both? I guess that's > reasonable. > > Danek > _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
