========================== C M U C L 19 e ============================= 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 " - 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 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 ) 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 for download information, guidelines on reporting bugs, and mailing list details. We hope you enjoy using this release of CMUCL!