Changing this behavior will adversely affect a lot of downstream modules. This change would have to be preceded by a deprecation warning for the duration of at least one major version, and it's probable that we wouldn't be able to actually change the behavior until a large number of modules updated first (no matter how many major versions that takes.)

I don't think the upsides of this (if there are any) are worth the breakage. -1.

No problem :-) Yesterday, when I was trying to understand the option mentioned in this comment, I simply used '' in fs.watchFile, expecting it to invoke the callback function, but it didn't. When I actually give an invalid path, it invoked the callback function. So, I went to the source code to understand why this happens. At this point, I didn't even imagine path.normalize call here would return the current directory. I checked the C++ part and couldn't find anything there and went into the uv source code to understand what it does and there also nothing interesting was done. So I introduced log statements in the C++ code and printed the error returned by uv_fs_poll_start here (Note: I am not sure why we are ignoring the error returned by uv_fs_poll_start). It was also printing zero. So, I started reading the fs.watchFile again, line by line and this time I figured that resolve returns the current directory, if an empty string is passed. I spent considerable amount of time figuring this out and I thought it would be not easy to debug for others also. Anyway, an empty string is not a valid filename. So, I thought issuing an error would be the best action here.

As the breakage is not justified by this PR, I am closing this and opening another PR #2106, targeting master, with the documentation changes. I hope that is okay.