Memory is not properly freed when removing data

[boesing]

Q	A
Bugfix	yes/no
BC Break	yes/no

Description

As of #5, a test was provided to ensure that memory is properly freed after calling flush.
Thats indeed not the case.

I've played around a bit aswell as in #5 was an assumption that it might be due to the ArrayObject by-reference pass of the key/value in AbstractAdapter of laminas-cache. I cannot confirm this as I've removed the ArrayObject and its still impossible to get memory freed.

I have lacks in knowledge of PHP memory handling and thus just want to provide the failing test here.
Help from the @laminas/technical-steering-committee and all others around would be highly appreciated.

Closing #5

[Xerkus]

on a cursory glance there are 2 issues:

• looks like a bug - memory_get_usage(true) with true provides allocated memory to the process. It is not the amount that is actually used within the allocated space.
• php does not immediately collect garbage. use gc_collect_cycles() in catch clause in test to immediately force GC run

[boesing]

• looks like a bug - memory_get_usage(true) with true provides allocated memory to the process. It is not the amount that is actually used within the allocated space.

Exactly. And that is the reason why we are allocating + 200 extra bytes which can be used.

• php does not immediately collect garbage. use gc_collect_cycles() in catch clause in test to immediately force GC run

I've already tried that - no difference.

[Xerkus]

Then again, the problem is that system allocated memory limit might be more important than the amount of actually utilized memory for some cases. Especially if app is run in environment where exceeding allocated memory limit will get app killed.

I have no idea how php manages memory, like does it always allocate new memory and releases/reuses only once each specific page is fully used? Does it reuse partially utilized pages?

[Xerkus]

I would say if you bounce off your more or less hard limit for the memory adapter you do not want to go and allocate more memory.

that makes me inclined towards wontfix and documentation note.

Also, do test with 4kb increments - the average page size

[boesing]

Closing in favor of #11