On Tue, Feb 26, 2013 at 3:00 PM, Robert Goldman rpgoldman@sift.info wrote:
I got through almost all the upgrade tests before failing with allegromodern trying to upgrade from 1.85:
; registering #<system :asdf @ #x1000afe932> as asdf ; Fast loading /Users/rpg/lisp/asdf/build/asdf.fasl TEST ABORTED: Attempt to fast load a non-fasl file: #P"/Users/rpg/lisp/asdf/build/asdf.fasl" Script failed upgrade FAILED for allegromodern from 1.85 using method 'load-asdf-lisp'load-asdf-system
I can see why this fails!
When I look at ~/lisp/asdf/build/asdf.fasl in emacs I see this:
#!/usr/local/bin/sbcl --script # FASL compiled from "/Users/rpg/lisp/asdf/build/asdf.lisp" using SBCL version 1.1.3.7-63e78fc
which suggests to me that we may be leaking something from one test to another...
Oh, I see.
Usually, we do not create a build/asdf.fasl, because the output-translations layer moves it away.
However, when testing antique version 1.85, we do create that file. And then, we have the clash between asdf.fasl from sbcl and asdf.fasl from allegro, just like we used to have in the bad old times — precisely because we're testing an upgrade from the bad old times.
To counter that, the first upgrade target is load-asdf-lisp-clean, but in this case, it wasn't cleaning enough.
I've pushed a fix to the repository, and my manual testing suggests the tests are working after in a way that they weren't before: make u l=sbcl ASDF_UPGRADE_TEST_TAGS=1.85 make u l=allegro ASDF_UPGRADE_TEST_TAGS=1.85
—♯ƒ • François-René ÐVB Rideau •Reflection&Cybernethics• http://fare.tunes.org Any sufficiently advanced misquotation is indistinguishable from an original statement. — John McCarthy, misquoted