News: what happened in 2002
- A new version of GerdsPCL has been uploaded to the experimental directory of the download mirrors. This version includes an optimization that can make class slot access as efficient as for structures!
- Documentation: a new document giving hints on using the PCL MOP has been put online.
- Binaries for x86-linux, x86-freebsd and sparcv9-solaris from current source have been uploaded to the binaries directory of the download sites.
- A new GerdsPCL tarball is available in the experimental directory of the download sites. This release can only be used with a recent CMUCL built from sources after 2002-12-13.
- Binaries featuring the "gerdspcl" version of PCL have been uploaded to the experimental directory of the download sites. There are binaries for sparcv9-solaris and x86-linux. The gerdspcl branch contains a number of bugfixes and speed enhancements that have not yet been integrated into the CMUCL version of PCL, but that likely will be included in a future release.
- There has been work on reviving the Linux/Alpha port of CMUCL; experimental binaries should be available in a few days.
- An updated version of the CMUCL User's Manual has been uploaded to the documentation section, in HTML, PDF and Postscript formats.
- A tarball containing the revamped source to PCL (the CLOS implementation in CMUCL) is available in the experimental directory of the download sites. This contains all the changes that have been made by Gerd Moellman. Please test that the optimizations and code deletions do not break your code.
The profiler now reports precise measurements of allocation on x86
platforms using the generation garbage collector (previously, the
reported information was imprecise). Non-gencgc platforms already
had precise allocation measurement, and are not affected by the
change. A new function
WITH-SPACEREPORTallows breaking up profiling reports into sections. A CMUCL binary with this change for x86-Linux is available in the experimental directory of the download sites.
- Some Solaris-sparcv9 binaries with the linkage table changes have been uploaded to the binaries directory of the download sites.
- The linkage-table feature has been ported to SPARC by Raymond Toy. This means that it works on all currently supported platforms.
- A Karatsuba multiplier implementation has been added for bignums, which should speed up multiplications for bignums with more than 512 bits.
- Binaries for x86-linux, x86-freebsd and sparcv9-solaris from current source have been uploaded to the binaries directory of the download sites. The x86 builds have the recent foreign-linkage changes enabled, and the sparcv9 builds have the recent 64-bit integer support enabled.
- Initial support for 64-bit integers on sparc-v8plus architectures (basically UltraSPARC machines) has been added by Raymond Toy. This allows the use of unboxed arithmetic on much larger fixnums, which can make certain algorithms much more efficient.
- On FreeBSD and Linux on x86 architectures, there has been a change to the way that foreign symbol addresses (function and data references) are resolved. An extra level of indirection, via a linkage table, has been added to the resolution process. This means that it is possible to dump images that contain references to symbols in shared libraries, and that image files are no longer dependent on the addresses of symbols in the lisp runtime. This work is due to Tim Moore.
- There have been a large number of bugfixes, ANSI compliance fixes, unused code deletion and cleanup of the PCL code in CMUCL, thanks to Gerd Moellmann. See the cmucl-imp mailing list for details of all the improvements that have been made.
- The format of APROPOS output is more verbose, showing more information on whether each symbol names a special variable, a constant, a symbol macro, and alien variable, a macro, a special operator, a function, etc.
- The nasty bug in inappropriate code deletion when using FLET or LABELS has been fixed, thanks to Alexey Dejneka.
- Binaries for NetBSD 1.5.2 on x86 have been uploaded to the experimental directory of the download site. They have known problems with floating point traps.
- Binaries from current source have been uploaded for the following platforms: sparcv9-solaris, x86-linux, x86-freebsd. Both normal and small images have been generated.
- Some reduced-footprint binaries have been uploaded to the experimental directory on the download site and mirrors. These include a lisp runtime built without debugging symbols, a runtime image (no compiler, no CLOS) and a small development image.
- Updated extra tarballs for 18d on x86/linux. If you use Hemlock on this platform you should reinstall the extra tarball; the previous ones were broken.
- CMUCL version 18d released. See the release notes for details of all the crunchy new features and bugfixes.
- Preparation for 18d release in around a week. Added documentation for the hierarchical packages feature.
- Fix for dynamic loading in FreeBSD 4.0 and later. This involves moving the static space up, so a cross-compile is required to bootstrap these changes.
- Allow users to disable tail-call optimization by setting a debug optimization quality higher than 2.
Changes to the compile-time behaviour of
DEFCONSTANT, to fix its behaviour when it is not at the toplevel.
- Add the ability to select rounding precision for x86 floating point operations.
- Anoncvs access is available again.
Experimental NetBSD/x86 binaries are available in the
experimentaldirectory of the download sites.