Avoid crash for reset/end/next/prev() on ffi classes

[dstogov]

Closes GH-9697

[TysonAndre]

Also, what's the motivation/context for overriding get_properties - if it's just reducing memory for typical workloads, it should be safe to override get_properties_for (which would keep memory low after var_export/debug_zval_dump/var_dump/json_encode)

Is there anything about FFI specifically that is a reason to avoid populating obj->properties

[TysonAndre]

What about ZTS builds? Technically, having multiple parallel threads calling reset() would be writing to this static array simultaneously.

That'd probably be writing the exact same value concurrently if multiple threads concurrently called reset(), which would be safe on most architectures, but may be unsafe in obscure architectures.

A bigger issue would be whether bugs in opcache, PECLs, the engine, or internal functions could actually write properties to this empty array - those would affect not only the given request, but all subsequent requests.

Doing something like ext/session and putting this in the request globals in rinit would limit the effect of bugs

[dstogov]

OK. You are right. Lets commit your solution.

Also, what's the motivation/context for overriding get_properties - if it's just reducing memory for typical workloads, it should be safe to override get_properties_for (which would keep memory low after var_export/debug_zval_dump/var_dump/json_encode)
Is there anything about FFI specifically that is a reason to avoid populating obj->properties

I think, FFI was developed before get_properties_for was added.
obj->properties for FFI objects just don't make sense.
If you like, you may refactor ext/ffi to use `get_properties_for``

[TysonAndre]

Also, what's the motivation/context for overriding get_properties - if it's just reducing memory for typical workloads, it should be safe to override get_properties_for (which would keep memory low after var_export/debug_zval_dump/var_dump/json_encode)

https://github.com/TysonAndre/pecl-teds/blob/1.2.6/teds_emptycollection.c#L422
https://github.com/TysonAndre/pecl-teds/blob/1.2.6/teds_util.c#L3-L8

E.g. for EmptySequence/EmptySet/EmptyMap in the teds pecl - https://github.com/TysonAndre/pecl-teds/tree/1.2.6#overview , I overrid get_properties_for to return null, and didn't override get_properties at all, since typical code paths in php 8.3 wouldn't really use get_properties that often with the various handler overrides, outside of rarely used or deprecated functions. Tests of expected behavior for empty collections pass