The above bug was particular bad, because it would cause asdf
to spuriously be reloaded, which would interfere with poiu itself,
and cancel extensions such as cffi-grovel, etc.

Also, better fix posix-waitpid on allegro.

Also, add a README file.

1.29.14: Make it work with 2.32, factoring an action-effectively-done-p around action-already-done-p

when not forcing. Make it work properly with ASDF 2.26.172.

It so happens that we still need to compare actions by type of operation only.
Hide that under an abstraction.

Move deferred-warnings support to asdf-driver.

Make it work with CLISP 2.44 (modulo patch to cl-launch, now 3.20.1),
checking for LINUX symbols at runtime.
Graceful fallback to serial behavior on unsupported platforms.

* Using files, not pipes, to communicate results, to avoid deadlock due to full buffers.
* Major simplification by hooking into recent ASDF's perform-plan interface,
 rather than kludging our way around perform for parallel operations.
* Take advantage of ASDF 2.26.15's new internal traverse and action-stamp APIs.

This is a major rewrite of several subsystems of POIU.
NB: it looks like some race condition it making things hard for us on either SBCL or CCL.
This isn't stable. Try POIU 1.027 with ASDF 2.26 for stable results.

for incremental builds.
Another attempt at fixing to-go counts.

Get undefined-warning reports in the correct branch;
treat SETF function names correctly.

Fix normalize-dependencies to use proper parent namespace for non-flat systems.
Add deferred warning infrastructure by Douglas Katzman (SBCL only for now).

external-format, *compile-op-compile-file-function*, compile-flags).
For that, depend on ASDF 2.21 or later (2.23 or later preferred).
Also, override any with-compilation-unit when compiling in a fork.
If that causes warnings you don't like to be emitted...
use asdf-condition-control.

* have system depend on asdf.
* know how to recover when signals dropped.
* define parallel-load-system, parallel-compile-system.
* make test executable as a shell script.
* I can make clisp break mostly reliably on fare-utils
 if both module pure and stateful depend on base as well as interface(!)
 it then claims "Cycle detected".

instead have the new operations depend on both "internal" component-depends-on
dependencies AND on "external" component-do-first dependencies, both frobbed.