Getting closer to release. The codebase is stable now, and I'm just making minor adjustments. Now at 2.26.164.
I'm going to release 2.27 as soon as all tests pass on all implementations. The magic feature is #+asdf3, and 2.27 should be considered a pre-release. Hopefully by next month, after some testing, we can declare the next one to be ASDF 3.0.
There were previously performance issues, with ASDF 2.26.152 being over 37 times slower than 2.26 at making a plan for iolib than it was before (3s vs .08s). After a day spent optimizing away a lot of truename queries and other pathname computations, ASDF is back at being 70% slower with default settings (while doing much more work behind the scene), and slightly faster if you disable all truename queries via (setf *resolve-symlinks* nil) -- but goodbye linkfarms if you do.
After the last round of upgrade cleanups, hot upgrade now works without punting on recent versions of ABCL or CMUCL. But still not on CLISP. There seem to be various optimizations in CLISP FASLs that defeat everything I've tried so far except punting by renaming away the ASDF package as the very first thing.
Many thanks to Stelian Ionescu, Stas Boukarev, Anton Vodonosov and Robert Goldman for their help with testing.
—♯ƒ • François-René ÐVB Rideau •Reflection&Cybernethics• http://fare.tunes.org You may easily play a joke on a man who likes to argue — agree with him. — Edgar Waston Howe