Closed tdunning closed 2 years ago
Thanks, and it is great to see this moving forward!
Hand-rolled to maven I personally think is totally acceptable assuming it is cryptographically signed and compiled with Java-8 or older. I am not certain the normal Java or Scala-accepted pathways here really so I will have to defer to someone more involved in those communities.
Yes. It would definitely be signed and compiled with Java 8.
Let's make and popularize a punch list of issues blocking the release if any, then. I don't know of any.
On Thu, Mar 25, 2021 at 9:56 AM Chris Nuernberger @.***> wrote:
Thanks, and it is great to see this moving forward!
Hand-rolled to maven I personally think is totally acceptable assuming it is cryptographically signed and compiled with Java-8 or older. I am not certain the normal Java or Scala-accepted pathways here really so I will have to defer to someone more involved in those communities.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/tdunning/t-digest/issues/163#issuecomment-807093125, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAB5E6QYIGTXUINR3BOKNATTFNTLXANCNFSM4ZZXHYIQ .
I think it depends on if we intend to further address the AVL tree pathway which I don't recommend at this point. If there is no intention to do so then there are no blocking issues that I saw and there is clear benefit to a point release as it addresses at least a few issues while not introducing any new known ones.
My point list is empty.
Other candidates that don't show up include:
some more quality tests that compare (well) against the sketches in Apache DataSketches
a response to the paper that showed a pathological data case:
https://arxiv.org/abs/2102.09299
On Thu, Mar 25, 2021 at 11:23 AM Chris Nuernberger @.***> wrote:
I think it depends on if we intend to further address the AVL tree pathway which I don't recommend at this point. If there is no intention to do so then there are no blocking issues that I saw and there is clear benefit to a point release as it addresses at least a few issues while not introducing any new known ones.
My point list is empty.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/tdunning/t-digest/issues/163#issuecomment-807231699, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAB5E6QIQDSHY7BVJHBI3V3TFN5R7ANCNFSM4ZZXHYIQ .
So.
The situation is that Travis runs are clean.
I added more quality tests and resolved a few lingering AVLTree issues.
Nobody on twitter spoke up about blockers.
A we ready to pull the trigger?
My only key issue (array index out of bounds) was resolved.
Chris said this:
@cnuernber Let's continue the discussion about the point release here.
I think that your points are really excellent all around. If you don't incorporate them into the serialization document, I will go ahead and do so.
The key here, then, is just the point release. My feeling has been that the changes so far are important and don't make anything worse. That is basically an echo of what you came up with.
I have been wondering for some time whether the hand rolled release through mvn central directly is the best path. It certainly is doable and there are big benefits to being boring and stable, but I hate being the only one in the game.
Do you have any thoughts about modern patterns of releasing jars? I haven't kept up with fashions in quite a while.