First check is here

... but only also if wd is < 0

man inotify

IN_Q_OVERFLOW
Event queue overflowed (wd is -1 for this event).

then add an assert for this case, referencing the documentation?

asserts are IMHO pointless, nobody enables them, but it clutters the code. Anyone touching the code should and has to read the man pages etc anyway.

no, an assert is sort of "program by contract": you make it clear that this situation ought to not happen, and it if does, at least some people may notice that

Anyone touching the code should and has to read the man pages etc anyway.

and an explicit mention/comment in the code never killed anyone, nor made the code "cluttered"; also what is valid today might change in the future, so writing down why a check/assert was added make sure that people reading the changed documentation know why it made sense when it was added

After two years of maintaining baloo (and other software for much longer) I can assure you you are wrong.

If we checked all the conditions which are guaranteed by the API (null termination of strings, only complete events, bounds check of event->len ...) here and everywhere else, the could would be cluttered. And just picking an arbitrary condition to be more equal is nonsense.

This is kernel API, it does not change.

Comments are useful when some code is not obvious. Copying an explicitly spelled out condition from the man page does clutter the code. Otherwise we would have to copy the whole man page to have the code covered in its entirety.