isNil will always be false unless M is an interface. It is best not to rely on such unexpected behavior.

What problem are you trying to solve here? Is it better to change trace.Metadata itself to a slice?

@SOF3 if trace.Metadata is nil，slice长度为1，identifiers序列化之后变成了字符串null，读缓存的时候没有办法再去判断这个null是个什么东西。 trace.Metadata 改为slice是可以，不过在读取缓存的时候依然有这个风险遇到未知类型。
不过any不就是interface{}吗，所以M不就一定是个interface{}，如果不是interface不应该会编译错误？如果是int等基础类型不为空应该是预期内的

M is struct{} when called from GetTrace(). any(struct{}{}) == nil is false.

I guess it meets your expectations, but checking metadata as nil randomly would be very strange (e.g. if M happens to be a thin pointer in the future, isNil will be false even for nil pointers).

I would prefer fixing this issue at a higher level, e.g. just making the trace.Metadata field an optional field or a slice, or just correcting it in the reader where it knows the real type of M.

ok, 我直接在读缓存的地方加个兼容吧

M is struct{} when called from GetTrace(). any(struct{}{}) == nil is false.

不过...这个好像是符合空指针判断逻辑的，这个case就不应该为nil

Yes, but when M is *Foo, any((*Foo)(nil)) == nil is also false, which could be confusing.

I mean, from a generic point of view, isNil doesn't serve any semantic purpose, so this logic is quite specific to one of the callers anyway.