Refine performance (2nd)

[XenoAmess]

refine performance for demo at apache/maven#2481

including 3 parts:

1. fix the bug about using MMap when value be collection, but the old codes didn't copy the collection, thus may cause appending to a list value in a DoneMMap, thus makes infinity long value, thus makes the run slow in super multi module project.
2. BfDependencyCollector can share executorService between resolvers, to reduce the create/close/create/close of the executorService, which is expensive.
3. intern some strings to make use of the new hash cache in String class in jdk.

Following this checklist to help us incorporate your
contribution quickly and easily:

• Your pull request should address just one issue, without pulling in other changes.
• Write a pull request description that is detailed enough to understand what the pull request does, how, and why.
• Each commit in the pull request should have a meaningful subject line and body.
  Note that commits might be squashed by a maintainer on merge.
• Write unit tests that match behavioral changes, where the tests fail if the changes to the runtime are not applied.
  This may not always be possible but is a best-practice.
• Run mvn verify to make sure basic checks pass.
  A more thorough check will be performed on your pull request automatically.
• You have run the integration tests successfully (mvn -Prun-its verify).

If your pull request is about ~20 lines of code you don't need to sign an
Individual Contributor License Agreement if you are unsure
please ask on the developers list.

To make clear that you license your contribution under
the Apache License Version 2.0, January 2004
you have to acknowledge this by using the following check-box.

• I hereby declare this contribution to be licenced under the Apache License Version 2.0, January 2004
• In any other case, please file an Apache Individual Contributor License Agreement.

[XenoAmess]

@cstamas please have a look.
the most severe part is the first part.(MMap part)

[cstamas]

-1: we can run on monsters as well. This is not an "improvement" on alg, just "throwing more" on it. In other words, doing this may hit "remote end" as well, think of https://en.wikipedia.org/wiki/Tragedy_of_the_commons like Central is.

[XenoAmess]

@cstamas we increase the number because we try to merge the executers. see changes in several lines under that in the same file.
before, each instance have one seperated executor，and thus have 2 problems

1. when running with like T2C, I afraid there might be very many of instance in parallel, thus many small executors in parallel
2. the start and stop of the small executors be time costing.(especially adding worker threads)

Still，if you don't like mixing idea for merging the executors of every instance, we can remove this part. I just don't think it good to start many small thread executors, and I still think a bigger single executor is better in this case.

[rmannibucau]

think we should use forkjoin pool by default, it is alsmot certainly created by a build these days and will avoid the overhead of custom pool with almost no benefit

[XenoAmess]

think we should use forkjoin pool by default, it is alsmot certainly created by a build these days and will avoid the overhead of custom pool with almost no benefit

yes it would be far better than using many small pools.
But as for forkjoin pool, I have no idea why we should use forkjoin pool but not executer here. IMO fork join pool is use for works who have relationship/dependency, and the tasks here seems not quite dependent on each other..?

[cstamas]

https://issues.apache.org/jira/browse/MRESOLVER-283

[XenoAmess]

https://issues.apache.org/jira/browse/MRESOLVER-283

Oh, I see #213 and the related comments...
Interesting.
Good, now I kind of getting interested in the fork-join-pool solution now.

@cstamas Is there any news since the original executor pr closed? should I start another pr(and split this from this pr) for that?

[cstamas]

No, nothing new since. It was just a faint idea, and did not move anywhere yet.

[XenoAmess]

No, nothing new since. It was just a faint idea, and did not move anywhere yet.

Um. Seems it is good enough a thing that worth my time. Would have a try when next weekends.
I agree on split it out of this pr, it is a bigger thing than my original thought.
Though I do agree it seems a better way than I provided in this pr.
Please focus on reviewing other part of this pr then.

[gnodet]

I think the intern logic (apart from the context in trace requests) would be better placed inside Maven when parsing the POM files.
It should be fairly simple, as it's just a matter of passing a ContentTransformer as an argument to this call.
We should intern all enums (scopes, packaging) and at least groupId, artifactId, version. The models are kept in memory throughout the whole build anyway, so it makes sense to intern quite a few values.

[gnodet]

I think the intern logic (apart from the context in trace requests) would be better placed inside Maven when parsing the POM files. It should be fairly simple, as it's just a matter of passing a ContentTransformer as an argument to this call. We should intern all enums (scopes, packaging) and at least groupId, artifactId, version. The models are kept in memory throughout the whole build anyway, so it makes sense to intern quite a few values.

I've raised apache/maven#2495. I was quite sure the work was done already, but after investigation, it had not been plugged in.

[XenoAmess]

I think the intern logic (apart from the context in trace requests) would be better placed inside Maven when parsing the POM files. It should be fairly simple, as it's just a matter of passing a ContentTransformer as an argument to this call. We should intern all enums (scopes, packaging) and at least groupId, artifactId, version. The models are kept in memory throughout the whole build anyway, so it makes sense to intern quite a few values.

yes, I agree.
interning is more important in mordern jdk versions because they add a cache for the hash of the string, thus a same string object calling hash methods for several time would be far faster than different objects.
(and we use hashmap based maps quite a lot)

I've raised apache/maven#2495. I was quite sure the work was done already, but after investigation, it had not been plugged in.

glad to see it. so I will remove this part from this pr too.