You should group these backports with the original patch you backport as to not leave master vulnerable.

You forgot the angle brackets in wallet/walletutil.cpp

Rebase on top of patches using the NULLPTR trick to make old versions of boost happy.

Use the NULLPTR trick to make boost happy

Try to make boost happy.

Just add friend class CSignatureCacheTest and define CSignatureCacheTest in the test and access whatever you want through it.

Remove call to synchronize now that it is not needed as we use RCUPtr.

In D2373#53771, @markblundeberg wrote:

In D2373#53753, @deadalnix wrote:

So the code looks good, but the new behavior is missing a test. You need to check that caching still works when flags like SCRIPT_VERIFY_MINIMALIF, but do hit a different key when SCRIPT_ENABLE_SCHNORR is passed.

Yes, I've been trying to think how to do this somehow. The only way to I can think of how to distinguish a correct cache hit from an unwanted cache miss would be to measure timing, or to add additional API that exposes the private cache. 🤔

Marking as request change, even though there is really no change to be made, just to clear the review queue.

This looks good to me but will need to wait for the release with the feature freeze.

Add a synchronize after removing the element from the tree. As it turns out, we may delete the element we just inserted and have to ensure all other threads are past it for correctness. Not a hue deal because this code is going away anyways, but worth fixing for correctness.

So the code looks good, but the new behavior is missing a test. You need to check that caching still works when flags like SCRIPT_VERIFY_MINIMALIF, but do hit a different key when SCRIPT_ENABLE_SCHNORR is passed.

Fix typo

The description is incorrect.

You should split this up in several diffs.

I don't understand what this task is about. What is inconsistent ? Why is this naming more consistent ? Who's going to maintain that test going forward ?