What these particular capabilities are about can be seen
in the LSP spec. I haven't checked, but I think this refers to
Eglot's capability to present meta information about a given
completion item and to do additional things elsewhere in the
buffer once a given completion item is chosen by the user.
So, typically you may be given the completion:
and if you do choose it, Eglot and LSP try to ensure that
an #include <iostream> appears in the top of your C++ file.
In general, normally both client and server advertise their
capabilities to each other so that the other side refrains from
doing anything that is not explicitly mentioned in these
advertisements. This is how there are really no "versions of
the protocol" (well there are, but they aren't used in the way
a normal versioned API is used). It doesn't always work that
way, and both clients and servers will both request stuff that
isn't possible and volunteer stuff that isn't needed.
It's indeed odd to find that Eglot starts advertising a
capability just by itself, without adding any new functional
code presumably backing that advertisement, so you're right
to be suspicious, Eli.
But it may be just that the advertisement was incorrectly
glossed over in the past (which wouldn't necessarily present
a problem since many servers don't bother to check
advertisements, as explained), or -- also possible -- that the
advertisement wasn't even specified in the past.
Most importantly, I would like Marcin to explain, if possible,
what actual problem with what server this is solving.
The reasoning and conclusion should be in the commit message
for this patch for future reference.