Interesting. For reference: https://github.com/ropensci/rtika

But, but, rJava 😭 ...

I think that the CRAN version does not use rJava, instead it offers an install_* function and pushes system calls out to a CLI. So although you do need java on your system, you don’t need working rJava.

So, via magic numbers? Even those can fail. For example, .ODS and .XLSX files can have the same signatures. And they're both .zip files anyway, so that's another level of ambiguity.

On the other hand, it doesn't take that much time for rio::import() to try and fail at reading an incorrectly specified file, and doesn't introduce any new dependencies. So all it would take is a wrapper function on rio::import() that first tries the default parsing and if it fails, iterates over the entire list of supported file types calling rio::import() until it either succeeds or they all fail.

I have a proof of concept here (though many formats not yet loaded and discussion/brainstorming needed before submitting it as a PR)

https://bokov.shinyapps.io/anyfile/

I'm not super keen on trying to parse with every imaginable import function. That might produce some kind of unanticipated weird behavior if one of those underlying functions changes to start supporting different kind of file or if we add future functionality that changes the deterministic order of import attempts.

We could add a separate function that does that, though, like try_import().