[asdf-devel] optimization settings

Robert Brown robert.brown at gmail.com
Thu Mar 20 12:20:06 UTC 2014


Here's the problem I want solved.  A library author wants to
compile some source files with high optimization settings.  How
can this be done?  The author could put

    (declaim (optimize ... ))

at the top of those files, but then any code compiled later may
accidentally get the new settings.  If my library depends on
others, I don't want the act of compiling them to change whatever
optimization settings I prefer.

If ASDF is going to manage readtables, why doesn't it manage
compiler optimization settings?  Maybe it should manage neither.
Can the around method technique handle both?

Bob


On Wed, Mar 19, 2014 at 6:22 PM, Robert Goldman <rpgoldman at sift.net> wrote:
> Robert Brown wrote:
>> Does ASDF set compiler optimization settings before compiling
>> each file when building?  I'm concerned that changing the
>> optimization setting inside one source file could cause the
>> change to stick and be used for other files.  I don't think the
>> CL standard mandates that "(declaim (optimize ... ))" applies
>> only to the current file.
>>
>> The issue is similar to that of readtables ...
>
> To the best of my knowledge (i.e., I'm not looking it up again right now!) you are right that the standard does not mandate file-scope for DECLAIM, and I have a *vague* memory that some implementation does allow it to leak.
>
> I am reluctant to add this to the set of things that ASDF (mis)manages for the programmer.  E.g, I think SBCL has a "policy" setting that is intended to provide a default setting for the optimization parameters.
>
> However, you COULD do this yourself, I believe, by writing an :AROUND method for PERFORM COMPILE-OP on your own system.  I'm actually not sure what the right mechanism is.  Would it be:
>
> (defmethod perform :around ((op compile-op) (c my-optimized-files))
>   (locally (declare (optimize (speed 3) (good-looking 11)))
>     (call-next-method)))
>
> I'd have to make sure that LOCALLY is going to do what I expect here....
>
>



More information about the asdf-devel mailing list