Re: Discussions about concerns over User API changes

2020-02-28 Thread Kirk Lund
Ok, thanks! I've submitted https://github.com/apache/geode/pull/4748 to revert the change. I'll be offline for a while so feel free to merge it if it passes. Thanks, Kirk On Thu, Feb 27, 2020 at 5:19 PM Owen Nichols wrote: > While changing a void method to have a return type does not break

Re: Discussions about concerns over User API changes

2020-02-28 Thread Udo Kohlmeyer
Another affect is code deployment onto/into the server, which could/would reference a change (binary) API. Users generally don't recompile the code they redeploy. The NoSuchMethod is now harder to track down. --Udo On 2/28/20 8:59 AM, Anthony Baker wrote: If I run the

Re: Discussions about concerns over User API changes

2020-02-28 Thread Anthony Baker
If I run the japi-compliance-checker [1] against the 1.12 release branch and develop this change pops up as a binary incompatibility. As Owen noted, this would require recompilation to avoid NoSuchMethod errors. One effect this could have is that a library built on top of Geode (e.g. Spring)

Re: Discussions about concerns over User API changes

2020-02-27 Thread Owen Nichols
While changing a void method to have a return type does not break source compatibility, it appears likely to break binary compatibility[1]. So, if you are compiling your client from source, it will compile successfully against either Geode 1.12 or Geode 1.13. But if your client was already

Discussions about concerns over User API changes

2020-02-27 Thread Kirk Lund
Remember, if there are any concerns about recent backwards-compatible changes to Geode user APIs, they should be brought on the dev list. Also, backward-compatible changes are by definition safe and ok for a user API because it won't break the user's code. Here's an example of a user API that I