- Sep 27, 2016
-
-
Robert Goldman authored
Cleanups This branch contains a few cleanups as I proofread the entire code. Those currently in the pipe regard asdf upgrade, since the first file in asdf is upgrade.lisp. See merge request !23
-
Robert Goldman authored
-
Robert Goldman authored
- Sep 25, 2016
-
-
Francois-Rene Rideau authored
-
Francois-Rene Rideau authored
The previous use of defun (foo) () relying on with-upgradability was confusing to new developers reading the code. Explicitly using defun* and defgeneric* in these cases makes the code less surprising and less confusing to readers.
-
Francois-Rene Rideau authored
-
Francois-Rene Rideau authored
Instance-allocated operation slots are of the devil.
-
Francois-Rene Rideau authored
-
Francois-Rene Rideau authored
-
Francois-Rene Rideau authored
-
Francois-Rene Rideau authored
Also make traverse-actions and required-components simple defuns. It's unclear why they were in action. Maybe at some point in the past, they have been used earlier in the build.
-
Francois-Rene Rideau authored
Stop using make-instance directly in a few places. Use our convenience methods in other places.
-
Francois-Rene Rideau authored
-
Francois-Rene Rideau authored
Don't use operation-original-initargs in operation-forced. operation-original-initargs is deprecated, and so is operation-forced. The only, misguided, user of operation-forced is swank.asd, which conditionally loads the code only if the operation is forced; but the perform method that does the check will only be called twice if the operation is forced, anyway, so the correct answer is T. (And the first time around, it's OK to return T, too.)
-
Francois-Rene Rideau authored
locate-system doesn't cache its results anymore, so we don't need to invalidate bad results from the cache anymore.
-
Francois-Rene Rideau authored
Simplify find-system and a test accordingly. Also, group map-systems with other related functions.
-
Francois-Rene Rideau authored
The order in which the upgrade fixups were called wasn't very robust. We need to first get a list of systems being defined, then clear defined systems (but not asdf, and *neither uiop* that asdf depends on), then upgrade the configuration, and *finally* reload any of the systems being defined (including their defsystem-depends-on dependencies!), which means this all has to be coordinated within the same function.
-
Francois-Rene Rideau authored
Only redirect error-output in logging mode.
-
Francois-Rene Rideau authored
CCL, our previous default, was making `make -f Makefile-lisp-scripting load` unhappy, since CCL's background thread is eating rlwrap's input. Using single-threaded-ccl stops the input race condition, but CCL still puts the tty in raw mode which confused rlwrap and this requires adding a dependency. Still use CCL by default on Windows, since SBCL can't invoke CMD directly.
-
Francois-Rene Rideau authored
Declare a FTYPE for ENSURE-PATHNAME in uiop/pathname so that CCL won't complain about a forward reference during `make load l=ccl`. SBCL insists that a type (FUNCTION (T &REST T) T) doesn't cover a function with keyword arguments, and wants either an exhaustive list of keywords and types, or a &ALLOW-OTHER-KEYS. So just say FUNCTION, to make SBCL happy, too.
-
Francois-Rene Rideau authored
Selection: REQUIRE, 3.1.7, 3.1.5, 3.0.3, 2.26. Older doesn't make sense since we now always "punt" on 2.26 and earlier. More versions since then makes limited sense because the code base has been much more stable with respect to upgrade since then. You can always test upgrade from more versions with e.g. ./make-asdf.sh l=sbcl u=all u
-
Francois-Rene Rideau authored
When testing upgrade from old versions of ASDF, we must accommodate the need that some old versions have for us to wrap the calls in c-w-a-c, and the need for yet older versions for us not to call c-w-a-c, that didn't exist back then.
-
Francois-Rene Rideau authored
p-u-r-h was never called! Two functions were registered. Register them with *post-upgrade-cleanup-hook* instead. The hook was introduced with pre-release 2.27, possibly with the intent to be called by the main operate :around method when an asdf build is restarted in the middle. But that never happened.
-
Robert Goldman authored
%run-program: Overhaul of asynchronous process spawning The primary changes made in this branch can be summarised as follows: - Employ parameter-error and not-implemented-error - Add and use process-info class with a few corresponding getter functions - Export functions: process-info-pid, wait-process - Add functions: launch-program, close-streams, process-alive-p, terminate-process - Unify exit status codes - Add tests; unix-specific for now. See merge request !3
-
Robert Goldman authored
The code for %POSIX-SEND-SIGNAL -- which isn't invoked on Windows anyway, was crashing on Windows build -- because of differences in function signatures on CCL Windows/other. Fixed this by simply making %POSIX-SEND-SIGNAL a stub on Windows. I couldn't remove it altogether, because deciding whether to invoke it is a decision made at run-time, not compile-time.
-
Elias Pipping authored
-
- Sep 24, 2016
-
-
Elias Pipping authored
Even though the LispWorks documentation says otherwise, at least on Unix, run-shell-command does not accept file streams or socket streams
-
Elias Pipping authored
-
Elias Pipping authored
-
Elias Pipping authored
-
Elias Pipping authored
-
Elias Pipping authored
ECL: terminate-process will not work with ECL 16.1.2 and earlier
-
Elias Pipping authored
-
Elias Pipping authored
-
Elias Pipping authored
-
Elias Pipping authored
Previously, a process that was killed through signal 15 could have an exit code of 15, 143, or even "sigterm"
-
Elias Pipping authored
Some of the new features added here will not work on ABCL 1.3.3 and earlier.
-
Elias Pipping authored
-