impl(otel): add OTelContext infrastructure

[dbolduc]

Part of the work for #12880

For more context on OTelContext, Googlers should see go/cloud-cxx:otel-async-context-dd or go/cloud-cxx:otel-async-context-slides

---

This change is 

[codecov]

Codecov Report

Attention: 2 lines in your changes are missing coverage. Please review.

Comparison is base (cd6da18) 93.59% compared to head (58d0c31) 93.60%.
Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff            @@
##             main   #12887    +/-   ##
========================================
  Coverage   93.59%   93.60%            
========================================
  Files        2064     2067     +3     
  Lines      179808   179917   +109     
========================================
+ Hits       168300   168409   +109     
  Misses      11508    11508            

Files	Coverage Δ
...oogle/cloud/internal/opentelemetry_context_test.cc	100.00% <100.00%> (ø)
google/cloud/internal/opentelemetry_context.cc	96.29% <96.29%> (ø)
google/cloud/internal/opentelemetry_context.h	92.30% <92.30%> (ø)

... and 4 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[coryan]

On second thought... maybe we should keep these in a multiset or some other data structure where std::find() can run faster?

Even a std::list would be better as the .erase() for vector is $O(n)$, whereas std::list::erase is $O(1)$. You still would have a $O(n)$ search.. that is where using std::multiset<> would give you $O(ln(n))$

[dbolduc]

We would only see duplicated Context objects in the list in practice when the futures are executed immediately. In that case we would basically take the current stack and duplicate it.

I think this was a tad lazy, so now we will check if we are in the same thread, and if so, we will not touch the stack.

---

Re: time complexity....

Now that the Context objects in the stack are all unique, and we can only compare equal, we cannot beat $O(n)$ search. (Also, I will remind that n <= 5).

Updated to use a std::list for the better erase.

[devbww]

I see v.erase(it, v.end()), which seems to be O(n) no matter what container you use.

[alevenberg]

If you are erasing from the back, do you need to use find anyway?

This is what I thought when I first looked at the doc

deque stack;

while(stack.back() != c) {
stack.pop_back();
}

(no need to make changes, just thought I'd leave my thoughts)

Updated: saw convo in chat nevermind

[dbolduc]

If you are erasing from the back, do you need to use find anyway?

This is what I thought when I first looked at the doc

deque stack;

while(stack.back() != c) {
stack.pop_back();
}

Yeah, that is true, and how I was thinking about it. Your API might be better.

The OTel Detach() implementation will find the element in their stack (searching from the back) and pop every element along the way (assuming the element is in the stack). I was copying that, but it was probably over cautious.

I think we can assume that the thing we are trying to pop is either back(), or it isn't found in the stack. I don't think we would leak memory.

[alevenberg]

I think your impl is good. What if it doesn't find it on the stack? My suggestion would delete the whole stack anyway

[coryan]

If you think it is either back() or not there the code would be simpler (don't forget the empty collection case).

Otherwise, consider std::find_if(stack.rbegin(), stack.rend(), ...);

[coryan]

Seems like this is some kind of "Remove Common Suffix", something like:

~ScopedOTelContext() {
...
  DetachOTelMany(std::move(context_));
}

void DetachOTelMany(std::vector<opentelemetry::context::Context> c) {
  auto& v = ThreadLocalOTelContext();
  auto p = std::mismatch(c.rbegin(), c.rend(), v.rbegin(), r.end(), [](auto const& a, auto const& b) {
    return a == b.first;
  });
  std::erase(v.rend(), p.second);
}

https://en.cppreference.com/w/cpp/algorithm/mismatch

[dbolduc]

TIL: std::mismatch

Seems like this is some kind of "Remove Common Suffix"

That is what I am trying to do. But I am struggling to erase using the reverse iterators.

[coryan]

https://godbolt.org/z/raG9K4ocW though I think it does not matter for small collections.