bug#7756: 24.0.50; enhancements to package.el

[Stefan Monnier <monnier@iro.umontreal.ca>]

> @@ -338,9 +338,14 @@
>  	 (pkg-file (expand-file-name
>  		    (concat (package-strip-version package) "-pkg")
>  		    pkg-dir)))
> -    (when (and (file-directory-p pkg-dir)
> -	       (file-exists-p (concat pkg-file ".el")))
> -      (load pkg-file nil t))))
> +    ;; When one is creating a package and testing it out, it is easy
> +    ;; to forget to add the -pkg.el file.  When that happens, it would
> +    ;; be useful to provide feedback to the user rather than silently
> +    ;; failing.  That is what (error ...) below is for.
> +    (when (file-directory-p pkg-dir)
> +      (if (file-exists-p (concat pkg-file ".el"))
> +          (load pkg-file nil t)
> +        (error "'%s' file is missing!")))))

I agree with the intention, but I wonder: why do we pass the `noerror'
parameter to `load' in the first place?

> @@ -569,8 +574,16 @@
>  (defun package-unpack (name version)
>    (let ((pkg-dir (expand-file-name (concat (symbol-name name) "-" version)
>  				   package-user-dir)))
> +    ;; Delete the package directory if it exists already.  This may
> +    ;; not be useful to the end users.  However this is extremely
> +    ;; important for package developers as they experiment with which
> +    ;; files to include in a package.  If a file is removed from one
> +    ;; iteration to the next, then the presence of the unwanted elisp
> +    ;; file could cause problems by polluting the generated autoload
> +    ;; file.
> +    (if (file-directory-p pkg-dir)
> +        (delete-directory pkg-dir 'recursive))

Here, I also agree with the intention, but I'm a little uneasy blindly
removing a whole subdirectory without warning: I'd add a confirmation prompt.


        Stefan