I hate them because they're an abstraction leak. The type of a file's content is different from the application that I want to use to manipulate it, which both are in turn different from the name I chose to give the file. That current file systems and the GUI representations on top of them are so badly designed as to confuse three unrelated concepts is a failing of software and as clear an indication as you could ask for of the triumph of Worse Is Better.
Why are they badly designed? There are a huge amount of bright OS hackers in the world, and a huge amount of mindshare on the problem. When the culmination of all this effort continues to be hilarious[1], I can only conclude that the problem is less tractable than we expected.
And in fact, that is what the people who work on the problem say[2]: that it is not tractable.
I'm not denying that fixing the problem requires a rethink. I guess I'm still just feeling beat up by the removal of the admittedly imperfect first class metadata from HFS in favor of file extensions. Roll that progress back, folks. Make NFS happy.
EDIT: I don't believe that the problem is tractable in a world where you need to maintain backwards compatibility; the solution then to me is ditch backwards compatibility. I almost always prefer elegant solutions to practical ones.
> the solution then to me is ditch backwards compatibility.
Backwards compatibility is a visible problem, but it isn't the True Problem with creatorcodes/metadata. The true problem is that a world with metadata is an unstable equilibrium, because what if just this once we get it wrong.
* Consider the case that the file's author got it wrong
* Consider the case of Python and RPython (or .js and .json, or .txt and .md) where a file can belong to multiple classes. Further consider the case where a system has a Python interpreter but not an RPython interpreter installed (or a JavaScript interpreter but not a JSON parser, or a text editor but not a MarkDown editor, or...)
* Consider the case where a document was saved with Excel 2013, but just happens to only use Excel 2010 features, because who really cares about whatever the heck they add in each new Office release
* Consider the case of a container format, like mkv or mp4, that does not alone completely describe the format of the data it contains
* Consider the case of sometimes-interchangable formats like m4a and m4b, mp4 and m4v, etc.
* Consider the difference between a storage format (.sqlite3) and the data that it contains (.localstorage, Firefox/Profiles/places.sqlite, ~/Library/Mail/V2/MailData/Envelope Index)
* Consider the case that that one day somebody just wants to make a "monkey-fighting file without filling out this monday-to-friday questionnaire" and either the "no really, it's a file" creator code is accepted by the International Body of Creator Codes, or it isn't and so there are five incompatible implementations.
As soon as you tolerate any of these (and probably the list is a lot longer than this), we are right back where we are now: trying to guess what kind of file it is by poking at it with a stick. Backwards compatibility is a red herring.
The conflation of file type and editor is orthogonal to the use of file extensions as type metadata. I agree that including the file's type in the name is suboptimal, but in a world where we're constantly sending files from one environment to a completely different one, it isn't entirely unreasonable either, as file names are the one truly universal way of identifying a file, and a file's type is part of its identity.
