Don't key upgrade attempt on source-registry, since things can also change
in the central-registry, etc.
Also rename required-files to direct-dependency-files
Improve the new convenience function pattern to work with operation initargs.
Implement getcwd on all platforms.
Actually catch sb-grovel messages.

Tests: make u l=ecl instead of make test-upgrade lisp=ecl
Tests: put the setup around the script, not inside;
Tests: start using globals, much easier for debugging

Also, have convenience methods for operation and component designators.
Enhance the incidental traversal functions.
Move compile-file error handling to lisp-build.

Passes tests on CCL and SBCL.

(equal #p"ccl:" #p"/") ==> T. Yikes.

A system named foo/bar will be looked up in a registered file foo.asd.
This makes it possible to sensibly name several systems in a .asd file
and still have asdf find them all by name.
We now use it to not have the asdf header be part of asdf-driver.

A new generic operation build-op makes it possible to deal with
the fact that the default operation for systems need not be load-op,
yet that the end-user does not want to have to know the exact operation
for each of the system he uses.

ensure-pathname has been improved again, and its clients tweaked.

remove-keys was made more sensible by using EQL rather than STRING-EQUAL.

:version need not take the first form in a file, but can take
a path to a subform. This makes it possible to get the exact subform
in a Lisp source file that has the string.
See SUB-OBJECT for how these paths work.

Cleanup of the semantics of ENSURE-FUNCTION on CONSes: use APPLY, not EVAL.

Some attempts at making things work for ABCL and CLISP.

* Pathname support has been cleaned up, yet again,
 extending and renaming COERCE-PATHNAME into PARSE-UNIX-NAMESTRING,
 also ENSURE-PATHNAME, many docstrings, and plenty of small enhancements.
* Bundles were refactored to take advantage of the new, more sensible,
 component-depends-on semantics.
* Image support was cleaned up again, and program-op was implemented
 to produce a standalone executable on supported platforms;
 a working hello-world-example is included amongst tests.
* The *compile-file-function* indirection was removed, with the
 ECL and MKCL object file support
 being folded into compile-file* instead.
* systems now can have NIL as component-pathname,
 instead of providing a non-sensible bogus pathname
 when no sensible pathname was determined.
 Most of our pathname functions have sensible default behavior
 when provided NIL instead of a pathname as an argument.

Settle on "restore-image" and "dump-image" for our API names.
Unbreak match-condition-p and move it to utility,
give static scoping to the generic muffler,
rename the compiler and loader mufflers.
Fix with-input.
Remove crud from script-support.

Package cleanup: every file in asdf itself now :use's asdf/driver asdf/upgrade.
Test script fix:
Allegro, like ECL *really* hates that we close *standard-input*,
so stop trying at all.
Tweak some utilities and debug utilities.

On Allegro, frob the autoload earlier and intern the gethostname later
so we don't unnecessarily autoload stuff.

On CLISP, don't undefine functions, just unintern everything in the beginning,
and it won't complain about disappearing methods anymore.

Also, only retrigger the upgrade attempt on source-registry change
if ASDF was not upgraded yet. This avoids unnecessary reloading of ASDF.

Finally, be sure to share more symbols between everyone - hopefully,
all the symbols that were used in ASDF and got recycled.
Slot names are a big one, notably.

Package tweaks to make CLISP complain less.
run-program/ and run-shell-command fixes and test updates.
wildening fix to issue found by CLISP.
! is used by EXT on CLISP, so use !p for defparameter.
More pathname madness for allegro.
Better upgrade.

* Fix bug found by CLISP regarding improper wildening.
* On CLISP, avoid nicknames in the package-definition-form.
* For ABCL avoid (a . #+foo b) pattern
* Also, test-clean-load now works on all lisps, not just SBCL

Tests: SBCL passes test-lisp and test-upgrade
Also:
* big refactoring of test infrastructure
* contrib/debug.lisp and (asdf/driver:asdf-debug) for debugging.
* remove aif and it, use the cleaner if-bind.
* backtrace support, with improvements from trivial-backtrace.
* integration of asdf-condition-control (originally from xcvb-driver)
* simplify the upgrade heuristic: always tries to upgrade once.
  The previous heuristic was too complex with too many failure cases.
  You are thus assumed to want a given version asdf
  if and only if it is in your source-registry.

This notably includes the former xcvb-driver:run-program/,
a much better replacement for run-shell-command.
This also includes image initialization, dumping, quitting.

Also, move wild-modules to a contrib/ directory.
Refactor tests somewhat, introducing a function touch-file.

Make default-directory a root pathname, remove it from source-registry.
Have it use lisp-implementation-directory.
Remove any attempt to use the "current directory".
Users who want that can still try to
	(pushnew '*default-pathname-defaults* asdf:*central-registry*)

Rename tmp/ to build/ and move asdf.lisp there. Update asdf.asd.
Rename implementation.lisp to compatibility.lisp.
Move *asdf-verbose* and asdf-message to upgrade.lisp

Avoids interspersed error output in CCL, or backtrace to terminal in SBCL.

New pathname functions to work around more logical-pathname issues.
See discussion with rpgoldman on asdf-devel, and see CCL ticket:953.
Also, tweak build to reflect recent refactoring of asdf.
WIP of define-package work needed for clean upgrade (still broken).

Add its forgotten source file test/lambda.lisp. Oops.
Skip one of its tests that mysteriously breaks on ECL_bytecodes.

Bug identified by avodonosov.

Also, fix make-pathname* to make CLISP happy.
Uniformly use "backward-compat*" in comments.
Rename operation-forced{,-not}-p to action-forced-p.
Shuffle some functions around.
Refactor operate into an :around and a regular method.

Remove debugging code from previous commit.

Be more selective in what warnings get quashed when upgrading ASDF.
Skip bundle test on ABCL, ECL-bytecodes.
Sleep more in test1.

where it ought to always have been, based on a properly cleaned up prepare-op
(previously introduced in 2.26.14 as parent-load-op).

This is a checkpoint of a mostly working system after major changes;
it requires further cleanup and fixes:
strangely, fails some tests on CCL but not SBCL.

Introduce find-operation as a companion to find-component,
subsuming make-sub-operation.
The results of component-depends-on are processed with find-operation and resolve-component-spec.
component-self-dependencies is fixed to play well with prepare-op.

Also, some internal renamings and signature changes:
parent-operation becomes upward-operation
parent-load-op becomes prepare-op
parent-source-load op becomes prepare-source-op
parent-component is distinguished from module
module-components becomes component-children
module-components-by-name becomes component-children-by-name
compute-module-components-by-name becomes compute-children-by-name
circular-dependency-components becomes circular-dependency-actions
component-load-dependencies becomes component-sibling-dependencies (with a stub)
introducing child-component, downward-operation.
refactoring visit-dependencies to include the loop in which it's always used.
refactoring force and force-not processing to use a function action-override-p;
force is achieved with a trivial method on operation-done-p for prepare-op,
and force-not as a trivial check in visit-action.

Tweak tests along the way. Tweak the documentation.

Failed to either rename operation-error or introduce action-error in its superclasses:
that breaks sb-grovel.