This will slow down the VM somewhat, although it's hard to say whether that makes a big difference without measuring. Essentially any handler that allocates memory could potentially throw out-of-memory errors, which are likely most of them.

AFAIK the main reason for the SAVE_OPLINE in error handling is that the current opline is replaced with an exception handler, which then later restores opline. If opline refers to an old opline then the exception handler will jump back to the wrong location.

The out of memory errors just do bailout, where an accurate opline is not really necessary. Thus the additional overhead in the VM might not be worth it.

Please let me know if there's something I'm missing!

I had the same concerns, but I wanted to measure the impact of adding them.

Right now I'm seeing a stable ~1% regression on the Symfony blog benchmark (only this one, interestingly), and since the crash is fixed in #10003, I will probably not pursue this.