Sam Yates

Hi Thorsten, I strongly believe throwing away the genericity of the sample_record is a bad idea; not everything we have now is best represented by a vector of doubles, and...

Just on a few of the points raised in description and your notes above, before I look more thoroughly at the probe-demo code. * The necessity of alignment between the...

I guess I can't really see how the current design (in broad strokes; details of course can be improved and simple_sampler also is a mess) is less frustrating than the...

Just to expand on the first part of the reply: if we represent a set of `mlocation`s as `mcable`s, something that gets mcables expecting a set of mlocations would then...

On the point of probes defined over multisets: it's not a matter of protecting users from some weird behaviour; it's still there because it is consistent with sampling more than...

I like having the type map (as implemented with the typedefs) from probe address type to metadata and value type. I'd use any_ptr in preference to std::any in the blob...

Ah, I did misunderstand what you meant there with the table, sorry! Even heterogenous-in-time data could be represented this way although I agree that that is kinda mad. I think...

Just for context, the locking was put in place so that samplers could be changed dynamically in response to observed samples, and so in fact could be called by multiple...

I always hate to see functionality go away, even if no one is using it :D Nonetheless, in this case it isn't a code maintenance concern but a performance and...