Local Package Aliases

Allows to define inside of a package aliases to refer other packages. Provides a reader macro $ to use the aliases (active only in packages having alias mapping, and does not affect other code). Portable.

(ql:quickload :local-package-aliases)

(defpackage com.my-company.some-library
  (:use cl)
  (:export #:func))

(in-package #:com.my-company.some-library)

(defun func () "hello")



(defpackage some-application (:use cl))
(in-package #:some-application)
(local-package-aliases:set #:com.my-company.some-library #:lib
                           #:some.other.library #:olib)

(read-from-string "$lib:func")
;; => COM.MY-COMPANY.SOME-LIBRARY:FUNC

($lib:func)
;; => "hello"

;; The aliases are not global; they are scoped only to the package
;; where they are defined:

(in-package #:cl-user)
(read-from-string "$lib:func")
;; => ERROR There is no package named "$LIB" .

;; The $ reader macro is non-terminating,
;; therefore it is only activated when $
;; is on the beginning of a token.
;; $ in the middle of a token has no special
;; effect.
(in-package #:some-application)

(read-from-string "just-a-$-symbol")
;; => JUST-A-$-SYMBOL

;; The $ designates alias reference only if the
;; current package has defined mappings. If there
;; is no alias mapping in the current package, then
;; $ is interpreted as usual:

(local-package-aliases:set) ; installs an empty alias mapping
(read-from-string "$-a-sybmol")
;; => $-A-SYMBOL

(in-package #:cl-user)
(read-from-string "$-a-sybmol")
;; => $-A-SYMBOL

;; However, CCL and CLISP signal an error when read
;; uninterned symbols whose name starts with a
;; macro-character, in our case $. For example #:$abcd.
;; See https://lists.clozure.com/pipermail/openmcl-devel/2013-February/009911.html
;; Therefore we do not recommend enabling the
;; $ macro globally, from your lisp implementation
;; init file - libraries that have such uninterned symbols
;; in their code will fail to parse.
;;
;; But for your own code, don't use symbols like #:$abcd
;;;and local-package-aliases will work.

To enable the $ macro in your lisp session:

(local-package-aliases:set-aliasing-reader *readtable*)

A macro character other than $ may be used. See the docstring for local-package-aliases:set-aliasing-reader for parameters description.

To return to the standard syntax:

(set-syntax-from-char #\$ #\$ *readtable* (copy-readtable nil))

To enable the $ macro during compilation of ASDF systems, use the :around-compile argument:

(asdf:defsystem #:some-application
  :depends-on (#:local-package-aliases
               #:com.my-company.some-library)
  :around-compile "local-package-aliases:call-with-aliasing-readtable"
  :components ((:file "some-application")))

SLIME support

SLIME uses the standard readtable rules for tokens such as pkg:symb and doesn't undersdand our aliases; thus, symbol completion, slime-edit-definition, and argument hints do not work out of the box for aliased tokens like $lib:func.

The solution we found is to hook into swank, and wrap evaluation of every SLIME request with temporary binding of the aliases defined in the current package as nicknames for their corresponding packages. So, during dynamic extent of every slime request, the aliases become real package nicknames and SLIME can handle them as usually.

It must be noted that this solution is not entirely transparent: when working from SLIME not only the reader follows the aliases, but also (find-package :$lib) will find the package. During normal run-time, only the reader knows about the aliases.

Functions hook-into-swank and unhook-from-swank enable/disable this SLIME support.

To have the SLIME support enabled automatically add the following to your ~/.swank.lisp:

(when (find-package :local-package-aliases)
  (funcall (read-from-string "local-package-aliases:hook-into-swank")))

Or this in ~/.emacs:

(add-hook 'slime-connected-hook
          (lambda ()
            (slime-eval '(cl:when (cl:find-package :local-package-aliases)
                            (cl:funcall (cl:read-from-string "local-package-aliases:hook-into-swank"))))))

Other Package Aliasing Approaches

Here is some information, solutions and ideas I encountered recently related to package aliases.

The solutions vary in whether:

• the aliases are only honored by the reader, or they also affect functions like cl:find-package, cl:find-symbol;
• the solution is portable Common Lisp, or relies on patches or language extensions;
• the aliases are scoped to a package, or somehow else;
• the solution is specific about aliasing scheme, or is a lower-level tool allowing to build various aliasing approaches.

cl-package-aliases - http://www.cliki.net/cl-package-aliases

Provides patches for 5 lisp implementations to introduce aliases. Aliases are scoped to package and visible both for reader and for standard functions like cl:find-symbol.

package-renaming - http://common-lisp.net/gitweb?p=users/frideau/package-renaming.git;a=tree

Tools based on cl:rename-package to temporary give packages desired short names/nicknames. Portable. To make the renaming local it is expected to be used with the ASDF's :around-compile argument.

CL language extensions

There were discussions to develop a CL language extension and propose it to CL vendors. The extension might be a hook called by CL to resolve package prefix, i.e. when encountering a token like pkg:symbol, call the hook with string "pkg", and the hook should return a package object, or maybe just a string designating real package name. Such hook may be called *package-prefix-resolver*.

Alternatively the hook may be passed the full token "pkg:symbol" and resolve names for both packages and symbols. Such hook may be named *parse-token-hook*.

There were considerations whether these hooks should be called only from the reader, or also by cl:find-sybmol and other functions. Sketch for a CDR: http://paste.lisp.org/display/133561 Discussions on the #lisp irc channel: http://ccl.clozure.com/irc-logs/lisp/2012-11/lisp-2012.11.05.txt http://ccl.clozure.com/irc-logs/lisp/2013-01/lisp-2013.01.06.txt

One more possible language extension would be to allow to fully substitute the lisp reader. In this case there might be a public library implementing fully compliant CL reader. Lisp implementations will delegate functions like cl:read, cl:read-delimeted-list, cl:set-syntax-from-char and others to the pluggable reader. The reader by default honors cl:*readtable*, cl:*package* and other variables, but also allows any custom hooks we need, such as described above parse-token-hook and/or package-prefix-resolver. It must be noted that interface between CL and such a pluggable reader will consist of many functions. Also, the reader should come with it's own implementations for all the reader macros, because standard reader macros are not implemented in terms of public cl:* functions, but use functions internal to the CL reader's implementation, such as read-token.

If speaking about pluggable reader, it's necessary to mention the reader-interception project: http://common-lisp.net/cgi-bin/gitweb.cgi?p=users/frideau/reader-interception.git;a=tree;js=1

It's a portable solution allowing to plug-in your own reader. It relies on the trick to look at the first character of input, configure this character temporarily as a reader macro, and then this reader macro may read the full input stream according to any rules.

Conclusion

The local-package-aliases approach with a reader macro seems to be a decent approach.

It is comparable by convenience with package-renaming.

In my opinion a form of package aliasing deserves to be introduced as a CL extension into all implementations.

The language extension may be specific, targeting only package aliases (like cl-package-aliases project proposes). It will encourage consistent coding practice across all the CL programs.

To simplify adoption of the extension by the CL implementations, I believe it should be enough to have aliasing only in reader. Calls to cl:find-package are rare, and we can pass full package names to it; on the other hand, if cl:find-package is unaware of aliases, it may complicate support by SLIME.

As for more low-level language extensions, like pluggable reader or various hooks, I would welcome them too, just to make Lisp more programmable and allow programmers to solve their needs simply.

Author

Anton Vodonosov, avodonosov@yandex.ru

License

MIT