- 17 Feb, 2010 1 commit
-
-
Perform the same transformation in the load portion init FASL as COMPUTE-CLASSFILE-NAME did in creating the files..
-
- 14 Feb, 2010 7 commits
-
-
be inherited by deriving metaclasses. Note: this does definitely *not* complete the metaclass work.
-
-
-
-
-
-
Fixes ANSI LOAD.ERROR.1.
-
- 13 Feb, 2010 6 commits
-
-
readable class cast exceptions for NULL and UNBOUND values.
-
-
-
-
-
-
- 12 Feb, 2010 5 commits
-
-
-
-
Check that the cache entries still accesses an open ZipFile when it is handed out. Use SYS:DISABLE-ZIP-CACHE to disable the ZipCache entirely. Implemented some notion of thread synchronization, although we cannot guard against the case where two or more references to a ZipFile exist, and one thread closes the ZipFile.
-
Contributed by Dennis Lambe Jr.
-
Found by Paul Griffioen.
-
- 11 Feb, 2010 4 commits
-
-
Corrects failing COMPILE-FILE.* ANSI-TESTs under Windows.
-
If ASDF is loaded via (REQUIRE 'ASDF), all subsequent invocations of REQUIRE will search for a loadable ASDF system definitions if the default resolver mechanism fails. SYS::*MODULE-PROVIDER-FUNCTIONS* now contains a customizable list of module provider functions. Such a function takes a single argument of the module that should be resolved and loaded. There is a builtin resolver #'SYS::MODULE-PROVIDE-SYSTEM that implicitly called before any functions in this variable.
-
-
-
- 10 Feb, 2010 5 commits
-
-
Work to reinstate 12425 continues on branches/metaclass/.
-
Treat jars as zips in ZipCache which maintains an cache of all ZipFiles accessed via Pathname jars (which should be the entire system as Load now uses Pathname). ZipCache currently does not invalidate entries for any non-file resources due to deficiencies in the JVM that need to be corrected on a per-protocol basis. For instance, for HTTP we need an implementation that uses HTTP HEAD requests to get the Last-Modified header as opposed to re-fetching the entire resource as the JVM URLConnection does. SYS:REMOVE-ZIP-CACHE-ENTRY implements a way to invalidate ZipCache entries from Lisp. Used it in COMPILE-FILE to successfully recompile FASLs under Windows. Rewrite remaining Pathname Primtives in the informative stack trace style. Implement Debug.warn() which can be shut off with SYS::*DEBUG-WARN*. The intent here is to have a way to warn about Java side events which having potentially worrying side-effects during development which is by default not visible to end users (although it can be). Removed unused EXT:LAST-MODIFIED in favor of existing ANSI FILE-WRITE-DATE.
-
-
-
It is still the case that if anything loaded by 'boot.lisp' causes an error, the 'abcl.compile.lisp' task does not properly signal an error to the Ant build process. But this is no different from the previous behavior, so we adopt this as the conceptually simpler approach to writing Ant tasks (and the one that doesn't litter the filesystem with intermediate build files when the compilation process is being debugged).
-
- 09 Feb, 2010 3 commits
-
-
Found by: Paul Griffioen
-
reported by Blake McBride.
-
The java.net.JarURLConnection implementation *never* invalidates its cache even for files on the local filesystem, making it highly unsuitable to represent FASL sources. One can create a custom protocol handler, but in the manner oh-so-typical of Java, we cannot use the default implementation so a sizable amount of coding lies ahead. For the time being, we just disable caching.
-
- 08 Feb, 2010 6 commits
-
-
-
-
Undo inadvertent fix in last commit.
-
Phil Hudson suggested in Feburary 2009 that "[FastStringBuffer] should be removed with all references to it replaced with java.lang.StringBuilder once enough confidence in this change has been gained." After almost a year of using FastStringBuffer as a delagate for StringBuilder, that confidence has indeed been gained. One subtlety for use of StringBuilder: there is no StringBuilder(char) constructor, so use StringBuilder(String.valueOf(c)) to construct a new StringBuilder containing a single char. Otherwise that char will get promoted to an int, and you will invoke StringBuilder(int capacity) which will "swallow" the first character that you thought you were adding.
-
The way to not confuse ABCL under Windows that pathnames starting in a drive letter are not Lisp packages is solved by using the 'inputstring' attribute to the <java> task.
-
-
- 07 Feb, 2010 3 commits
-
-
r12420, and the bug r12420 was supposed to fix has been fixed, too.
-
-
The requirement that packed FASLS had to end in ".abcl" was introduced in the recent Pathname code as a optimization to avoid opening/closing an input stream on the assumption that renaming FASLs would not be expected to work. But COMPILE-FILE has an :OUTPUT-FILE arg that certainly can be used, so we relax this restriction. Found by dmiles in the ANSI test suite.
-