I think we can do it in an alternative approach. We can remove the Debug field in vm.Config, but define Tracing() function(or some other better name) instead. If the tracer is configured, then we dump the execution statistics into tracer.

// Config are the configuration options for the Interpreter
type Config struct {
	// Debug                   bool      // Enables debugging
	Tracer                  EVMLogger // Opcode logger
	NoBaseFee               bool      // Forces the EIP-1559 baseFee to 0 (needed for 0 price calls)
	EnablePreimageRecording bool      // Enables recording of SHA3/keccak preimages
	ExtraEips               []int     // Additional EIPS that are to be enabled
}

// Tracing returns an indicator if evm execution tracer is configured.
func (config *Config) Tracing() bool {
	return config.Tracer != nil
}

I'm slightly in favor of dropping Debug as Gary suggested. We can check vmConfig.Tracer != nil to avoid the method invocation overhead.

It seems so far tracer is the only use-case for Debug. Even if it were other cases (e.g. mocking caller as per #26414) they won't be necessarily invoked in the same place as tracing. E.g. see:

https://github.com/ethereum/go-ethereum/pull/26414/files#diff-64508d317d86e7a4a5294d76455a5e74148e26883da9e8b4ffc9bbfa3cc8550eR289-R290

Ok, I'm fine with @rjl493456442 's idea, but then we need to turn it into a local variable in the interpreter loop

I think we can do it in an alternative approach.

Implemented in #27048, PTAL

Closing in favour of #27048