fix a bug for set_difference()

[wenqiang-gu]

@brettviren I found that, in the function Aux::untagged_traces(), the std::set_difference was given a std::unordered_set. However, the elements are actually assumed to be sorted in the std according to cppreference.

Here is a quick test:

  std::vector<size_t> all={0,1,2,3};

  std::unordered_set<size_t> tagged={0,1,2,3};  // fail
  // std::set<size_t> tagged={0,1,2,3};         // success, but maybe not recommended?
  // std::vector<size_t> tagged={0,1,2,3};      // success

  std::vector<size_t> untagged;
  std::set_difference(all.begin(), all.end(), tagged.begin(), tagged.end(), std::inserter(untagged, untagged.begin()));

Note that untagged_traces is used in Drifter, ChannelSelector, ManifySInk and FrameMerger. Given that most of our use cases are just retrieving all traces when the upstream simulation/data has no tag for traces. Therefore, this bug should not make an actual difference.

It makes a difference only when the upstream frame does have tagged traces, but we still want to retrieve some untagged traces. As an example in the above test, we have traces 0,1,2,3 and all of them are tagged. If we implement it properly, there should be no untagged traces, but the bug would give us untagged traces 0,1,2,3.

[brettviren]

Good catch, @wenqiang-gu !

I think I was originally worried about the loop over tags producing duplicate entries in the tagged set. But, if I read std::set_difference() correctly, any duplicates in tagged should not matter (maybe the check goes slightly slower) so your fix seems good to me.

[HaiwangYu]

Hi @brettviren and @wenqiang-gu, I do have a question. Are we sure the tagged is ordered? Otherwise, we may need to sort it first.
test: https://onlinegdb.com/IBdf7AeRt

[brettviren]

Oh, good question. It might be accidentally true but best to be defensive.

Either a std::sort() or the use of a (sorted) std::set would be good. @wenqiang-gu sorry to miss this and merge so quickly. Can you please add it?

[wenqiang-gu]

@brettviren I have made a new PR here: #179