[asdf-devel] Is this a bug? or my error?

[Robert Goldman]

On 6/30/10 Jun 30 -2:11 PM, Faré wrote:
>>> The contents of laturk-unique.asd:
>>>
>>> (in-package :asdf)
>>> (defsystem "laturk-unique"
>>>   :version "0.1.0"
>>>   :name "laturk-unique"
>>>   :author "Louis A. Turk"
>>>   :description "Converts utf-8 text file to list of sorted unique
>>> words."
>>>   :components ((:file "packages")
>>>                (:file "iterate" :depends-on ("packages"))
>>>                (:file "collate" :depends-on ("iterate"))
>>>                (:file "laturk-unique" :depends-on ("collate"))))
>>
>> A couple of random suggestions:
>>
>> 1.  Don't put this into the :asdf package.  If you want to, create your
>> own package, using ASDF and COMMON-LISP, and put the definition in there
>> (there's some debate about whether this is strictly necessary, but
>> that's a subtlety we don't need to go into now).
>>
> I beg to disagree. If you're not going to define new bindings,
> and particularly if you're not even going to intern new symbols,
> there's nothing wrong with being in package ASDF.

I suppose that's true.  I usually just use a dedicated package every
time I make a system (it's part of a simple emacs template for .asd
files), and then I know I'm safe and don't have to think.

If one is not going to intern new symbols, etc., one is safe, as you
say.  But to determine whether or not you are safe you have to think.

There's enough thinking to be done already, so I'm willing to dispense
with it here! ;-)
> 
>> 2.  Don't rely on transitive closure for dependencies.  E.g., probably
>> collate depends on packages as well as iterate and similarly for
>> laturk-unique.
>>
> Here too I beg to disagree. Unless and until there is a tool for which
> this makes a difference, the only criterion for choosing what dependencies
> to explicitly include is what makes for a more maintainable result,
> at which point the author knows best.

Matter of style, I guess.  I like to leave it to ASDF to do the
transitive closure for me.  Then if I cut a dependency (e.g., I realize
that a file just contains DEFUNs that are not used at compile time), I
don't end up inadvertently losing a dependency that only was tracked
because of transitive closure.

As a matter of style, it's up to the author (as long as s/he gets it
right!), since it should not have any effect on someone who just uses
the library.

Best,
r