CMUCL 19e released
CMUCL Project
The CMUCL project is pleased to announce the release of CMUCL 19e.
This is a major release which contains numerous enhancements and
bugfixes from the 19d release.
CMUCL is a free, high performance implementation of the Common Lisp
programming language which runs on most major Unix platforms. It
mainly conforms to the ANSI Common Lisp standard. CMUCL provides a
sophisticated native code compiler; a powerful foreign function
interface; an implementation of CLOS, the Common Lisp Object System,
which includes multimethods and a metaobject protocol; a source-level
debugger and code profiler; and an Emacs-like editor implemented in
Common Lisp. CMUCL is maintained by a team of volunteers collaborating
over the Internet, and is mostly in the public domain.
New in this release:
* Supported system:
- Support for Mac OS X/Intel has been added for OSX 10.5 and
later.
* Issues
- One some openSUSE systems, cmucl will fail to start because it
can't map enough memory. This might be caused be a memory limit
that is set too low. Use "ulimit -a" to see the limit on
virtual memory. This can be adjusted with "ulimit -v <N>"
- Tracing on Mac OS X/Intel uses encapsulation, unlike other
platforms. This means tracing recursive functions may not show
the recursive calls. You can try setting
debug::*trace-encapsulate-default* to :default. Be warned that
under some conditions, tracing will cause an error such that you
can not get back to the repl and must terminate the process in
some other way.
* Feature enhancements:
- Support for dynamic-extent added for ppc. However, it suffers
from the same problems with dynamic-extent as other platforms,
so it is disabled by default. Tests indicate that it does work
in simple situations.
- PARSE-TIME recognizes the format produced by C asctime/ctime.
(This change may break some other less commonly used patterns.)
- PARSE-TIME recognizes and discards any microseconds.
- PARSE-TIME checks that a specified day of the week matches the
actual day of the week given in the date. An error is signaled
if they are inconsistent.
- New option to SAVE-LISP allows creating executable Lisp images
that do not require a runtime loader.
Syntax: (save-lisp "filename" :executable t)
Currently supported on FreeBSD and Linux; work on a Solaris
version is underway.
Limitations: depends on files in "library:" to dump new
executable images.
- CMUCL's version of CLX has been replaced with telent CLX.
- Preliminary support for external formats. Currently only
iso8859-1 and utf-8 are supported. Utf-8 support is limited
since CMUCL only has 8-bit characters.
- UNIX-MPROTECT added to access mprotect.
* ANSI compliance fixes:
- BOA constructors with &AUX variables are handled better now.
- SHADOW accepts characters now.
- Default initargs are now passed correctly to initialize-instance
and shared-initialize.
- Several issues in formatted output of floats have been fixed:
o ~,dF won't print out extra digits if the number is too small.
o ~E sometimes erroneously printed the overflow filler instead
of the number.
o ~G has changed so that ~E is chosen more often. This is seen
when printing large numbers. Instead of printing lots of
zeroes, ~E is used. ~G now matches what Fortran would do.
o Inconsistencies between print and ~E are now gone. (See Trac
ticket #1.)
o Some incorrectly printed results for ~E have been fixed. (See
Trace ticket #12.)
* Bugfixes:
- Floating-point traps are now handled on ppc. Previously, no
traps were signalled and SET-FLOATING-POINT-MODES did nothing.
- FILE-POSITION no longer returns negative values for mapped
file-simple-stream's.
- Potential Version numbers that start with a leading 0 are no
longer treated as version numbers. Hence, "foo.~1~" has name
"foo", type nil, and version 1, but "foo.~01~" has type "~01~"
and version :NEWEST.
- A bug in type derivation for EXPT has been fixed. (expt x y)
for x of type (double-float (0d0) 1d0) and y of type
(double-float (0d0)) now returns (or (member 0d0) (double-float
(0d0) 1d0)) instead of (double-float 0d0 1d0), i.e., -0d0 is not
in the range.
- On sparc, the decoding of a trapping FP instruction is correct
now. Previously the wrong instruction was decoded, which
produced the wrong operation and operands in the arithmetic
error handler.
- Fix issue with UNIX:UNIX-MMAP handling of "large" addresses that
appeared to be negative numbers.
- DOUBLE-DOUBLE-FLOAT fixes:
o Converting negative rationals to double-double-float's doesn't
produce wrong answers anymore.
o (float <negative bignum> 1w0) no longer returns a positive
result.
o Some issues with creation of DOUBLE-DOUBLE-FLOAT and (COMPLEX
DOUBLE-DOUBLE-FLOAT) have been fixed on sparc, ppc, and x86.
These seem to work, except there appears to be some bugs on x86
when compiling at speed 3 and safety 0.
o (INTEGER-DECODE-FLOAT <double-double-float>) was sometimes
returning the wrong integer value because the two components had
the wrong sign.
o Some issues with debugger printing out DOUBLE-DOUBLE-FLOATs and
(COMPLEX DOUBLE-DOUBLE-FLOAT)'s have been fixed.
o CLOS now recognizes that (COMPLEX DOUBLE-DOUBLE-FLOAT) is a
valid built-in class instead of a random object.
o Branch cuts for ASIN and ATANH for double-double-float's
should match the branches for double-float's.
o ATAN2 should correctly handle signed double-double-float
zeroes.
o FASL files containing -0w0 are now converted to -0w0 instead
of 0w0.
o SIN and TAN return -0w0 when the argument is -0w0.
Previously, they returned 0w0.
o Signed zeroes are handled better for addition, subtraction,
and multiplication. That is, the correct signed zero is
returned now.
o Overflow in addition, multiplication, and division returns
infinity if traps are disabled.
o EQL supports DOUBLE-DOUBLE-FLOAT's now.
o The printer and reader should now be consistent for
double-double-floats.
o Conversion of bignums and ratios to double-double-floats
should be more accurate.
o Double-double-float's should have print/read consistency now.
o TRUNCATE works now when given a DOUBLE-DOUBLE-FLOAT divisor.
o FORMATted output of DOUBLE-DOUBLE-FLOAT's should work.
- The assembler for ppc had some arguments for some instructions
in the wrong order, producing the wrong effect.
- When making displaced arrays, the element types are checked to
see if they are type equivalent. If not, an error is signaled.
- The reader for #= and ## has been enhanced to be much faster for
cases with a large number of shared objects. However, it is also
somewhat slower for simple cases.
- #p"..." now has a namestring and is treated as a pathname with
name ".." and type "".
- #p"..a" erroneously had directory (:relative). This has been
fixed so that directory = nil, name = ".", and type = "a".
- Compiling code using SIGNUM no longer causes a compiler error.
The defoptimizer for SIGNUM was missing a case for
double-double-float.
- MAPHASH no longer causes a type-error when the mapping function
calls (setf gethash) on the same table.
- NOTINLINE declarations are honored for local functions even if
they only have only one use. Previously, these would be inlined
anyway. This allows tracing of such functions.
- TRACE and UNTRACE should now work as expected for local
labels/flet functions. Untracing should work. Redefining a
function should automatically retrace the local functions if
they were traced previously.
- Callbacks should now work on systems where malloc'ed space does
not normally allow execution of code.
- The FLOAT-ACCURACY declaration has been removed. This should
have no affect on most user's code, unless they were using
this. The default is 53-bit (double-float) precision for
everything and the compiler takes care of the precision for the
appropriate ABI. It is the compiler's responsibility to make
sure single-precision floats are done correctly with
single-precision results. (This may be buggy currently.)
- The :PRECISION-CONTROL keyword in SET-FLOATING-POINT-MODES has
been removed.
- A compiler bug with type inferencing and deleting unsued
functions has been fixed.
- A compiler bug has been fixed where a source transformation was
incorrectly applied because the local variable happens to have a
function type and has the same name as a known function.
- On Darwin/x86, tracing with :encapsulate nil works better, and
tracing of labels/flets works better. However, there are still
issues with tracing. Returning to the repl after tracing
sometimes fails, and you must exit lisp completely.
- ~E should be faster now in some cases.
- The range over which sin/cos/tan would use the builtin
instruction on x86 has been reduced. Previously, values near
the limit would either produce result greater than 1 in absolute
value or cause a floating-point exception. Note that accuracy
is reduced when the builtin instructions are used for very large
arguments.
- Issues with compact info environments should be fixed. The
issue manifests itself when a new core is saved with lots of
functions (or other objects). The resulting core executes the
wrong code for some of the functions.
* Trac Tickets:
- #8 fixed so logs of bignums and ratios that won't fit into a
float can be computed, instead of signalling an error.
- #9 fixed. An error is generated if the number of days doesn't
match the number of days in the given month.
- #10 fixed. ROUND should now return correct answers for floats
bigger than most-positive-fixnum.
- #11 fixed. EQL handles double-double-float's correctly now.
- #1 fixed. prin1 and ~E should produce the same results.
- #12 fixed. (format t "~10,1,2,0,'*,,'DE" 1d-6) prints "0.1D-05"
now.
- #13 fixed. (format nil "~11,3,2,0,'*,,'EE" .9999) produces
0.100e+1 instead of 1.000e+0.
* Other changes:
- UNIX:UNIX-ERRNO accesses the thread errno value instead of the
global errno value.
- Floating point zero is now printed with an exponent of zero,
independent of any scale factor that might be used. Thus, we
get results like "0.0D+00" instead of "0.0D-5".
- CMUCL should now build and run on Redhat Fedora Core 9.
* Improvements to the PCL implementation of CLOS:
- Forward-referenced classes are allowed.
* Changes to rebuilding procedure:
- load-world.sh now supports a -p option to load the world without
PCL. This is mostly for cross-compiling which doesn't want PCL
loaded because it's not build during a cross-compile.
- make-dist.sh now defaults to bzip compression instead of gzip.
This release is not binary compatible with code compiled using CMUCL
19d; you will need to recompile FASL files.
See <URL:http://www.cons.org/cmucl/> for download information,
guidelines on reporting bugs, and mailing list details.
We hope you enjoy using this release of CMUCL!
Daniel Weinreb
It's great to see so much good work being done. I have updated some
items in my LispSurvey paper, for its next publication. Congratulations
on your release.
-- Dan