The bug exists on lower branches, isn't it?
Can we mark the new class as internal to make this target 4.4?

Thanks, I picked the wrong base branch when creating the PR

I'm not sure we need to fix this "issue". In the original issue report, the reproducer in #39156 is a code that writes a string 26 millions of times. How is that realistic?

@javiereguiluz the original laravel issue was abotu a long-running command outputting lots of different things. The minimal reproducer is not meant to be a realistic case, but a minimal case.

The original Laravel issue report says that this problem occurs "After running it for a long time". How long is "long" considered here? Hours? Days? Weeks? I wouldn't try to fix this because for now it looks like a "non issue". (In PHP you must restart long-running processes from time to time to avoid these "memory leaks" or other issues, right?)

The script leaks. The fix is not that complexe...
I don't understand why we shouldn't fix it...

In PHP you must restart long-running processes from time to time to avoid these "memory leaks" or other issues, right?)

NOooo!!!! Why people believe that having a memory leak is the normal behavior?

I hope I won't offend you @javiereguiluz , but I'm really tired of people saying PHP sucks and have a bad memory management.
Running a PHP script for days is perfectly fine!

The original Laravel issue report says that this problem occurs "After running it for a long time". How long is "long" considered here? Hours? Days? Weeks? I wouldn't try to fix this because for now it looks like a "non issue". (In PHP you must restart long-running processes from time to time to avoid these "memory leaks" or other issues, right?)

If this is not the problem, our script can run forever without memory leaks.

This has nothing to do with the length of time the problem occurred. If the business needs it, the trigger point can be reached in a day. If this function is not used, it will never be triggered. What is the significance of this function? Knowing that there is a problem that can be solved, but through a method that is not a solution?

Thank you @jderusse.

@jderusse: Can you please have a look at b047064? When merging your changes to 5.1, I had to add an extra line to the test to make it pass again.

@jderusse: Can you please have a look at b047064? When merging your changes to 5.1, I had to add an extra line to the test to make it pass again.

Thanks for pinging. Yes, you fix is OK.
I'm not sure why it broke, but the idea of the tests remains the same:

1. warm the buffer
2. memorize the size of the memory
3. run the script several times
4. assert the size didn't changed

Yes, the test still proves that the leak is stopped. 👍🏻