filename-extension mishandles hidden files

[mbutterick]

A hidden file starts with a dot, but filename-extension will treat its name as an extension:

#lang racket
(filename-extension (string->path ".foo"))
> #"foo"

I would argue this is incorrect: that a filename extension is not merely a substring that starts with a dot, but a substring that starts with a dot and has some kind of filename in front of it (for which it serves as the extension).

Therefore, the fix would be to change the regexp pattern in the current filename-extension:

(define (filename-extension name)
  (let* ([name (file-name 'filename-extension name)]
         [name (and name (path->bytes name))])
    (cond [(and name (regexp-match #rx#"[.]([^.]+)$" name)) => cadr]
          [else #f])))

To have a dot pattern at the front, like so:

(define (filename-extension name)
  (let* ([name (file-name 'filename-extension name)]
         [name (and name (path->bytes name))])
    (cond [(and name (regexp-match #rx#".[.]([^.]+)$" name)) => cadr]
          [else #f])))

Because then:

#lang racket
(filename-extension (string->path ".foo"))
> #f

[mflatt]

I agree that filename-extension should not treat a leading "." as an extension separator, but I worry about backward compatibility. There's also path-replace-suffix and path-add-suffix, which have the same problem and same compatibility issues.

(Aside: why "extension" versus "suffix"? The root of this mismatch is ancient history and reflected in the function name add-extension-suffix. Outside of Racket, he term "extension" seems clearly to be the preferred one.)

One possibility is to introduce path-replace-extension, path-add-extension, andpath-extension` that do the right thing, and then document the old functions as deprecated. How does that sound?

[mbutterick]

Very good, thank you.

BTW not to induce mission creep, but why do these functions treat the extension/suffix as a byte string? It seems peculiar when everything else is geared toward Unicode.

#lang racket
(path->string (string->path "foo.λ"))
(filename-extension (string->path "foo.λ"))
> "foo.λ"
> #"\316\273"

[mflatt]

Unfortunately, paths are not composed of characters in general. On Unix variants, paths are composed of bytes. On Windows, they're composed of UTF-16 code units. Using a byte string accommodates both of those.

[mflatt]

Unfortunately, the name path-extension created conflicts for the "grip" package. I've renamed it to path-get-extension.