Old News from 20e and later
FLOAT-TRAPPING-NAN-Pin favor of
FLOAT-SIGNALING-NAN-P. Signaling NaNs are also now printed with "Signaling" instead of "Trapping".
- FP exceptions are handled better on the ppc port.
- The compiler can now derive the type of
- Fixed an issue with multi-processing destroy-process doesn't take effect for a schedulable processbecause it has no run-reason and/or arrest-reason.
#105, #84, #101, #106, and #100have been fixed.
- A few issues with fdlibm not signaling FP exceptions have been fixed.
- Updated to ASDF 3.1.3.
- More consistent results for the special functions because all platforms now use fdlibm.
- Consing for the trig functions is greatly reduced since we use fdlibm instead of a Lisp implementation.
- Source location information has been added for
DEFINE-CONDITION. (From Helmut Eller.)
- The lisp binary is now compiled to use SSE2 on x86 machines. CMUCL won't even start now if you do not have SSE2.
CLEAR-OUTPUTwill now actually clear any pending output that has not been flushed. Previously,
- For Gray streams,
CLEAR-OUTPUTwill clear output instead of calling
- A critical bug in
KERNEL:DOUBLE-FLOAT-BITSon x86 has been fixed. In some situations,
KERNEL:DOUBLE-FLOAT-BITSwould cause a segfault.
There was no snapshot made for this month.
The 2014-06 snapshot has been released. See the release notes for details, but here is a quick summary of the changes between this snapshot and the previous snapshot.
- An error in printing
MOST-NEGATIVE-FIXNUMhas been fixed.
- A style warning is now issued when using
ECASEand friends. (Ticket
- Support for x87 has been dropped. This snapshot and future snapshots will be no longer include support for x87 and will require SSE2 support.
- Support for x87 is being dropped. This snapshot will be the last snapshot that includes support for x87. All following snapshots will only run on x86 chips that support sse2.
- Type derivation for
(log -0d0)must be
-infinity + i*pito be able to compute the special functions correctly on the branch cuts.
(log -0w0)returns values that correspond to
- More missing double-double float constants are added like
:ELEMENT-TYPEto specify the element type for the streams that are created.
- Lisp-unit (a Lisp framework for testing) has been added as a
(require :lisp-unit)to compile and load it. Unlike asdf and defsystem, precompild fasls are not included.
- Type derivation for
logwas not consistent with the values actually returned by
log. This is corrected now.
- Double-double floats were missing the constants describing various
limits such as
LEAST-POSITIVE-DOUBLE-DOUBLE-FLOAT. These are added now.
- When reading very small floats values, the reader will round to least-positive float if possible. Otherwise an error is still signaled.
- Micro-optimizations in arithmetic vops for x86/sse2, reducing the number of memory loads in some cases.
- Run-time checks for zero pages are now disabled; the lazy zeroing of heap pages has been the default for some time without problems, so the run-time check is disabled.
- A serious error in the complex multiply vop on sparc has been fixed.
- The trig functions
tanare faster now and more accurate.
- A bug in
CISintroduced in the 2013-12 snapshot has been fixed.
:RADIXkeyword arguments, like
- Improved speed for some SSE2 VOPS.
- Values on the branch cuts for the inverse trig and hyperbolic functions match the definitions. If there is a difference between the CLHS and formulas, we choose the formula.
- Updated support for Mac OSX 10.5 and gcc 4.8 on ppc.
Due to the bug in
CIS introduced in the 2013-12 snapshot, the
2013-12 snapshot is deprecated.
- The compile can derive the result types of
CISuses the faster C library routine
-?options to print out some help.
- ASDF 3.0.3.
- Optimizations for x86 (ported from sparc and ppc).
- Numbers will be printed with lower case letters when
CMUCL 20e released
Because of the release, there will not be a 2013-10 snapshot.