commented:
every function should be nodiscard unless it returns void

  commented:
Really? Even things like std::set::insert (which returns an iterator
to the new element or the existing one and a boolean indicating
whether the element was inserted or already present)? I've used that
return value maybe once ever, out of all the times I've used that
function. It's pretty common (at least in C++) to have functions which
return something which some consumer might want but most consumers
won't.
Would you split that into two functions or just insist that every
caller explicitly indicate that they don't need the return value?

  commented:
Yeah exactly, just have two functions, one of them can call the other
one. Been trying out this way of doing things for 10 years now. I can
confidently say it's great. All upsides, no downsides.
It's easy to discard something at the callsite, too.

  commented:
m[k] for try emplace is actually a disgusting footgun, I wish they had
tried to fight this fight :(

  commented:
I'm with him on this.  If I see mymap[key] at the beginning of a line,
I'm going to expect an assignment of some kind and think it was
possibly a bug if I didn't see one.  The (void) will at least alert me
that it's intentional and save me the trouble of double-checking it.
Regarding old code bases, this seems like something you could add a
clang-tidy modernize rule to change to .try_emplace() or add the
(void) for you.
.