For the 1.x - where the issue is the most pressing for us - I will try to open a backport PR.

I did think about this but didn't re-add the filtering, primarily because tests stay green. In practice everywhere it is used in Sinatra itself it is only the first caller location that gets used/preserved. Since Sinatra is now shadowing a Ruby builtin, there are 2 courses of action possible:

• To preserve full backwards compatibility - we would need to implement the "in-hoise" caller_locations under a different name (something like sinatra_caller_locations) so that external tools - Bootsnap and RSpec to name a few - would get the builtin implementation, always
• Reimplement caller_locations with all the possible overrides and argument semantics as per Kernel implementation, and then try to "spice" it with the customised Sinatra filtering
• Aim for the intersection of required functionality for Sinatra and the builtin - what this PR proposes

What would be your preference? IMO ending up in a situation where you override a kernel method is undesirable because these usually have quite subtle implementations

Aa you said, our tests passed, but it doesn't mean that your change is safe for all public APIs.
We have some implicit public APIs and it tightens us, but it means the developer may depend on that.
So I have to say the first idea isn't safe changes.

Actually your proposal has a important meaning for improving our ecosystem so I'd like to consider merging this change but it's not now.

I'll go back to the change next week, and then I'm going to rethink the necessity and its history at that time.

Btw, thanks for your contribution.

@namusyaka Did you arrive at a decision by any chance? We are nearing a Rails 5 upgrade and also there is something that might have to happen on the bootsnap side. Should we monkeypatch in the meantime?

Please monkey-patch at this time if necessary.
It's worth to get your change for improving sinatra next major or minor release such as v3.0 or v2.1.

FWIW, I just ran into this when using ActiveSupport's cattr_accessor inside a class inheriting from Sinatra::Base.

In Rails 6.1 ActiveSupport expects a caller_locations method that can take two arguments, but the one defined in sinatra does not take any arguments. So using cattr_accessor breaks with ArgumentError: wrong number of arguments (given 2, expected 0).

@namusyaka can we move forward with this in v3.0? I think this would be a good reason to release a v3 along with other updates that have been sitting.

@jkowens Is this planned to release anytime soon?

@namusyaka @olleolleolle I think I'm good with merging this into the unstable branch for v3.0. Do you have any thoughts or concerns?

Yay!