Bug in PATHNAME-AS-DIRECTORY

[edicl]

Bug in PATHNAME-AS-DIRECTORY

[Devon7]

(list :software (list (software-type) (software-version))
:implementation (list (lisp-implementation-type) (lisp-implementation-version))
:machine (machine-type)) =>
(:SOFTWARE ("Darwin" "18.7.0")
:IMPLEMENTATION ("Clozure Common Lisp" "Version 1.12.1 (v1.12.1) DarwinX8664")
:MACHINE "x86_64")

[stassats]

I don't think ~ is defined, it's an implementation specific thing.

[Devon7]

No input, defined or not, legal or not, should produce illegal output such as a "" directory component.
CL-FAD exists because everything to do with pathnames is implementation specific.
By widespread convention ~ is the user's home directory - follow the rule of least surprise.
#+CCL (setq pathspec (pathname pathspec)) seems to fix it in CCL,
now it runs (fad:walk-directory "~" …) ok but that's no stress test.
CCL does

	(pathname-name "~") => "~"
	(pathname-name #P"~") => NIL
	(pathname-name (pathname "~")) => NIL

therefore

	(fad:directory-pathname-p "~") => NIL
	(fad:directory-pathname-p #P"~") => #P"/Users/devon/"
	(fad:directory-pathname-p (pathname "~")) => #P"/Users/devon/"

which smells buggy to me, I'll grovel their sources and maybe talk to them too.

[stassats]

fad is a portability layer, ~ is not portable, so how it works is moot.

[Devon7]

The current behavior is compatible with filesystems that allow zero length names.
All others would prefer to signal an error rather than poison the caller with time bombs, e.g.,

(ccl:advise fad:pathname-as-directory
	    (let ((dir (:do-it)))
	      (assert (not (find "" (pathname-directory dir)
				 :test #'equal)))
	      dir)
	    :when :around
	    :name fad--pathname-as-directory-safety)

Oh, and a Merry Solstice to all, and to all a good night!