Small memory fixes

[marschall]

We analysed the memory usage of our application and found a lot of empty LinkedHashSet/Collections.synchronizedSet in the following three cases:

• org.springframework.beans.factory.annotation.InjectionMetadata.injectedElements
• org.springframework.beans.factory.annotation.InitDestroyAnnotationBeanPostProcessor.LifecycleMetadata.initMethods
• org.springframework.beans.factory.annotation.InitDestroyAnnotationBeanPostProcessor.LifecycleMetadata.destroyMethods

This fix should address this issue by:

1. Creating a LinkedHashSet of the "right" size, eg. the default capacity is 16 but we know the exact capacity that we need in advance
2. If the argument is empty then we use Collections.emptySet() which is a constant so no additional memory is used. Since it's immutable we also save ourselves the Collections.synchronizedSet wrapper.

Should I additionally open an issue?

[cbeams]

Hi Phillipe, thanks for the submission. We've just added a contributor guidelines doc to the wiki. Please read through that and follow the instructions, particularly with regard to the CLA and commit comment formatting. Note that it's fine for you to collapse these two commits into one given that they are addressing the same concern.

[marschall]

• I signed the CLA. Do you need my confirmation number?
• I created an issue, SPR-9264
• I made a new commit with correct comment formatting that also collapse both commits into one (I hope) and updates the year in the license header.

Is there any way to update this pull request or should I close it and make a new one?

[cbeams]

Just do a forced push (-f) against your existing small-memory-fixes branch

[marschall]

Ok, there we go, it seems to have worked.

There is a small issue with the patch though. InjectionMetadata.checkConfigMembers and LifecycleMetadata.checkConfigMembers both synchronize of the underlying collection. In the case of new LinkedHashSet this is fine but in the case of Collections.emptySet() we synchronize on a global variable that is accessed from other places. This could be avoided if we first check for #isEmpty() before synchronizing. Let me know if you want me to update the patch accordingly.

[marschall]

I did sign the CLA.

[cbeams]

Merged. Note that the commit comment was updated slightly, a line of trailing whitespace removed, and else blocks formatting changed from:

} else {

to

}
else {

for consistency.

Thanks!