"Deferred by Instruction" stats don't add up.

[markshannon]

Bug report

Bug description:

The "Deferred by instruction" stats should add up to the total "Not specialized" total in Specialization effectiveness but they don't come close.

In the current latest stats "Not specialized" total is ~8.8 billion, but the "deferred by instruction" adds up to only ~4 billion.

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Linked PRs

• GH-123232: Fix "not specialized" stats #123236
• GH-123232: Factor BINARY_SLICE and STORE_SLICE to handle stats properly for tier 2. #123381

[markshannon]

It looks like we are treating any instruction with a cache as specializable.
It used to be true than any instruction with a cache was specializable, but it isn't any more.

[markshannon]

The "deferred by instruction" counts won't quite add to the "not specialized" total as they don't include successful specializations.
But they should be close.

[brandtbucher]

I think the new stats are still incorrect in tier two (the macro uses the opcode local variable, which is meaningless). And they won't build with the JIT, since opcode isn't defined in the JIT template.

[markshannon]

I think the new stats are still incorrect in tier two

These are tier 1 stats. They are (mostly) meaningless with tier 2 active as tier 2 doesn't specialize, so messes up the ratios.

And they won't build with the JIT

Fixed by #123381