Re: [PATCH] Re: package.el changes before the feature freeze

[Stefan Monnier]

I haven't read the whole patch, but here are some nitpicks.
The general idea looks fine, tho.  We'd need a ChangeLog with that,
it should describe the changes that are neither cosmetic nor simple
adjustments to the use of defstruct.


        Stefan


> +Slots:
> +
> +`:name'
> +Name of the package, as a symbol.
> +
> +`:version'
> +Version of the package, as a version list.
> +
> +`:summary'
> +Short description of the package, typically taken from the first
> +line of the file.
> +
> +`:reqs'
> +Requirements of the package. A list of (PACKAGE VERSION-LIST)
> +naming the dependent package and the minimum required version.
> +
> +`:kind'
> +The distribution format of the package. Currently, it is either
> +`single' or `tar'.
> +
> +`:archive'
> +The name of the archive (as a string) whence this package came."
> +
> +           name
> +           version
> +           (summary "No description available.")
> +           reqs
> +           kind
> +           archive)

Nitpick: the fields of the struct (which you can call "slots" if you
prefer, of course) don't have a ":" in front of their name.
[ I'd also prefer using fewer lines in the docstring, so the whole
  definition can hopefully fit within a tall-but-split frame.  ]

> -(defun package-activate-1 (package pkg-vec)
> -  (let* ((name (symbol-name package))
> -      (version-str (package-version-join (package-desc-vers pkg-vec)))
> +(defun package-activate-1 (pkg-desc)
> +  (let* ((name (package-desc-name pkg-desc))
> +      (version-str (package-version-join (package-desc-version pkg-desc)))
>        (pkg-dir (package--dir name version-str)))

Hmm... `name' in the new code is now a symbol whereas it was a string in
the old code.  Is that right?

> -    (load (expand-file-name (concat name "-autoloads") pkg-dir) nil t)
> +    (load (expand-file-name (concat (symbol-name name) "-autoloads")
> pkg-dir) nil t)

You can use (format "%s-autoloads" name) to make it work equally with
strings and symbols.

> +             (apply 'define-package-desc

BTW,please stick to the "package-" prefix.

> +                    name-string
> +                    version-string
> +                    summary
> +                    requirements
> +                    _extra-properties)))

Obviously you haven't played with lexical-binding yet, but the "leading
underscore" is used to denote variables/arguments that are not used, so
the above use of _extra-properties indicates that it should be named
`extra-properties' instead.

> -      (package-unpack name version))))
> +      (package-unpack (symbol-name name) version))))

All those make me wonder: do we need the `name' slot to be symbol?
Why not let it be a string?

> +       (make-package-desc :name name

I know it's the default, but I also prefer not to use the "make-" prefix
and use "package-" as the prefix instead.