I'm coming to ivy from flx-ido and found that filename matching was very different and less useful. I finally discovered the reason for the difference. When calling flx-score, flx-ido uses flx-file-cache, which is a filename cache defined in flx-ido.el. In ivy, the call to flx-score uses ivy--flx-cache which is a string cache defined in ivy.el. The difference in behavior is pretty big. Changing ivy--flx-cache to a filename cache would probably be too disruptive to people already familiar with the existing behavior. Maybe it would be possible add an official way to switch to using a filename cache rather than a string cache? In my current configuration I include (setq ivy--flx-cache (flx-make-filename-cache)) after ivy is loaded and the flex matching now works like I'm used to.

Here's an example of how matching is different. I have an sbt project with a source named something like "src/main/scala/org/fu/MySuperApp.scala". With a query of "ms" the string cache settles on the 'm' in main and the 's' in scala so all of my source code ends up with the same flx-score. The filename cache selects the 'M' and 'S' in the file name, which is a much more useful result.