Robert Goldman wrote: > Daniel Herring wrote: >> On Sun, 29 Nov 2009, Robert Goldman wrote: >>> I'm actually surprised to see that cl-utilities wants to load the swank >>> asdf definition, too.... >> >> It was McCLIM that wanted swank (IIRC for a simple IDE). Cl-utilities >> requires sb-rotate-byte. >> > > OK, I'm confused. Why is what we see here unusual? I'm looking at the > transcript and what seems to happen is that SG initially causes the > various and sundry McCLIM sysdefs to be loaded (they are all in > mcclim.asd), and then asks for the cl-utilities asdf definition to be > loaded. > > From there I don't see anything particularly odd in the transcript. We > get CL-UTILITIES, a boatload of CLIM systems, and SWANK. SWANK is > imperatively loaded by McCLIM. The CLIM systems are all defined in > mcclim.asd, which has been loaded by an SBCL initarg. > > So what's weird here? What's the seemingly-unrelated software that we > are looking up? is it esa-mcclim? I don't know about this because > esa-mcclim.asd is not in the mcclim release tarball.
Ok. Cl-utilities only intends to load sb-rotate-byte. Unfortunately, FIND-SYSTEM actually has to load ASDF files to find the definition it needs; by default, it reads each .asd in *asdf-c-r* until it finds the one it needs. Many system definitions are declarative; loading them causes negligible side effects. However, some asd files are more involved, in some cases even dumping fasls. I think Samium was objecting to the registration of unrelated systems as part of ASDF's FIND-SYSTEM traversal. To me, the other side effects in some asd files are a bigger issue. There would be several benefits if FIND-SYSTEM were to simply PROBE-FILE for a system definition in each of the configured paths (e.g. path1/system.asd, path2/system.asd) until one is found. Better efficiency, no pollution from unrelated packages, etc. - Daniel _______________________________________________ asdf-devel mailing list [email protected] http://common-lisp.net/cgi-bin/mailman/listinfo/asdf-devel
