Fix GH-17727: JIT SEGV on OOM in dtor when creating backtrace

[nielsdos]

This became visible after GH-17056 was merged, but technically the lack of setting the opline is also present on lower branches. We set the opline to mirror the SAVE_OPLINE() from ZEND_INIT_STATIC_METHOD_CALL().

Targeting master, but I can merge into 8.4 (except the test) and make equivalent changes to 8.3 if wanted.

[dstogov]

I suspect, now we have to keep consistent EX(opline) state before each memory allocation (including possible indirect allocations)...

Why this might be a problem before (for PHP-8.4)?

[nielsdos]

I suspect, now we have to keep consistent EX(opline) state before each memory allocation (including possible indirect allocations)...

This is probably true

Why this might be a problem before (for PHP-8.4)?

I remembered this discussion: #12648 where they wanted to add SAVE_OPLINE before allocations to get the right (non-NULL, up-to-date) opline when an OOM happens. Out of sync opline seemed to have caused crashes in 3rd party extensions before and also in the core PHP (fixed in #10003 by adding a NULL check). Perhaps we can just add a NULL check although that may make an inconsistency between JIT and VM.

[dstogov]

I don't see a big problem adding SAVE_OPLINE() in this single place, but I wouldn't like to add it before each potential memory allocation.

[nielsdos]

but I wouldn't like to add it before each potential memory allocation.

I agree, I also would not like that, and I expressed that in the linked PR already.
I think however, we are "lucky" that many places already have the opline set (or at least non-NULL) before allocating.

[dstogov]

I agree, I also would not like that, and I expressed that in the linked PR already. I think however, we are "lucky" that many places already have the opline set (or at least non-NULL) before allocating.

I don't object against this commit.