Let\'s say I build an application on top of net.aserve
and bordeaux-threads
. My package declaration might look like this:
(defpackage :
Use quickproject (accessible via (ql:quickload :quickproject)
) to create a system for your application. As z_axis described, you can then fill the list of dependencies in the defsystem
declaration (if you missed any when you called quickproject:make-project
).
If you create your new project in the local-projects path of you Quicklisp installation, you can quickload your project too (even if it's not part of the Quicklisp distribution yet). Quickloading your project will of course download the dependencies (if they are part of the Quicklisp distribution), then load them.
I had exactly the same question and I agree I should not force a package manager on a user. Before quicklisp's time I was using clbuild and it puts all .asd files into a systems/ directory. As long as the `systems/' directory is in asdf:central-registry, one can simply (require "a-package"), at least in SBCL and CCL, to load all relevant packages. The new clbuild2 retains this feature if you do install-from-upstream, and its integrated quicklisp does respect the separately installed-from-upstream packages, but quicklisp installed packages don't expose their .asd files anymore.
So my solution is to write a shell script that scans all quicklisp installed packages, usually under dists/quicklisp/software/, and link all .asd files there to a central place. In this way one doesn't need to load quicklisp into the cl image if one only wants to use quicklisp installed packages. I hope quicklisp could ship this feature by default.
If you don't want to include a quicklisp call in the deployed source code at all, separate the quickproject system definition file from the rest of the source.
At the top of the source, just before the defpackage
call, add the necessary (require ...)
's for your package dependencies. This guarantees that those lisp packages are loaded (somehow) before proceeding, but does not specify 'how' those packages get loaded. They could be loaded by running the ql:quickload :my-package
call (using quickproject), which would first load the dependencies, and then run through the require calls when loading the source. Or possibly a user could load the source directly (without calling ql:quickload
), and the dependencies would be loaded during the require call, if those dependencies can be found on the *module-search-path*
. This technique, as you said, would allow the end user to use whatever build tool he/she wants to load your source.
After experimenting with this for a few minutes, it seems that quicklisp latches into the require function call, so that if quicklisp is installed, and (require :bordeaux-threads)
e.g., is called, lisp will use quicklisp to download and install that dependency. This is a very nice feature (IMO), because it allows the Common Lisp standard require
function to act as the interface layer, and abstracts the specific build tool used to satisfy the dependency. Quicklisp can latch into the require, asdf latches into it (IIRC), etc.
So to answer your question, quicklisp calls should not go anywhere in the deployed source code, and requires
should be used to ensure that dependencies are loaded before the package definition file is evaluated. If someone has quicklisp installed before loading the package definition file, those requires will be satisfied by using quicklisp to download and install the dependencies. If someone has asdf installed, those dependencies will be satisfied with that build tool. And if someone already has the dependencies installed (using some other technique), the requires will simply be passed over.
In your asd file, you should define the depends realtion as below:''
(asdf:defsystem #:aserve
:serial t
:depends-on (#:hunchentoot :hunchentoot-cgi
#::bordeaux-threads
#:parenscript)
...)
After then you just need to (ql:quickload :aserve) .