From: CMUCL Project
Subject: [ANN] CMUCL 19a released
Date: 
Message-ID: <wziu0vj4885.fsf@melbourne.laas.fr>
========================== C M U C L  19 a =============================

August 3, 2004

The CMUCL project is pleased to announce the release of CMUCL 19a.
This is a major release which contains numerous enhancements and
bugfixes introduced since the 18e 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:
       
  * Feature enhancements:
     - Stack overflow checking added for FreeBSD/x86, Linux/x86 and
       Solaris.  Code that makes the control stack overflow (for
       example due to unbounded recursion) will lead to a
       STORAGE-ERROR being signaled, instead of making CMUCL crash.
     - LISP:FIND-CLASS and LISP:CLASS-OF now return CLOS classes.  The
       symbols LISP:STANDARD-CLASS, LISP:BUILT-IN-CLASS, and
       LISP:STRUCTURE-CLASS name CLOS classes.  It is no longer
       necessary to SHADOWING-IMPORT these from PCL when doing MOP
       programming, and it fixes some non-compliances as well.
     - Package locks: a package whose EXT:PACKAGE-LOCK is enabled
       non-NIL is protected against changes in its structure (list of
       exported symbols, use-list etc). Symbols in a package whose
       EXT:PACKAGE-DEFINITION-LOCK is enabled are protected against
       redefinition.
     - Weak hash tables have been added for x86 systems.
     - Local functions are now named (FLET <name> ...) or 
       (LABELS <name> ...).
     - Compiler checking control string and number of args to FORMAT,
       ERROR and WARN.
     - TRACE automatically using encapsulation when function-end
       breakpoints cannot be used.
     - INSPECT working on CLOS instances.
     - Callbacks from foreign code to Lisp.  Implemented for x86 and
       sparc.
     - Functions like GETHASH that are used in the implementation of 
       TRACE can now be traced.  See also
       DEBUG:*TRACE-ENCAPSULATE-PACKAGE-NAMES*.
     - New facility for encapsulating functions, FWRAPPERS;
       see also the CMUCL User Manual.
     - TRACE and PROFILE have been reimplemented using fwrappers,
       with the effect of tracing and profiling now working more
       nicely together.
     - When a profiled function is redefined, it is re-profiled.
     - A basic implementation of simple-streams added.  Some
       functionality may be missing.
     - UNIX:UNIX-LSEEK working with files > 2G on BSDs.
     - Type declarations on local bindings of special variables no
       longer being ignored.
     - DISASSEMBLE disassembling closure/funcallable instance
       functions.
     - (SETF EXT:WEAK-POINTER-VALUE) has been added.
     - CMUCL no longer entering an infinite loop when dumping
       circular constant lists to fasl files.
     - COMPILE-FILE-PATHNAME's BYTE-COMPILE keyword arg defaulting
       to *BYTE-COMPILE-DEFAULT* like COMPILE-FILE.
     - Added VOPs for CHAR=, CHAR<, and CHAR> when the second arg is a
       constant character.
     - Added missing VOPs for Sparc for array refs with a constant
       index. 
     - Lisp binary will also look at the PATH envvar to find out it's
       own location which is used to find other needed files.
     - Byte-compiled code being executed ca. 1/3 faster.
     - Heap overflow checking for x86 and solaris with gencgc.  When
       the heap is close to overflowing, a heap-overflow condition is
       signaled.  If an overflow happens again while debugging the
       condition, the user is returned to top-level without warning.
       Use lisp::*reserved-heap-pages* to set the number of pages to
       reserve.
     - C functions are displayed in backtraces now for x86/linux and
       sparc.  Previously, it just said "Foreign function call land".
       (From Helmut Eller.)
     - Source location recording is done better and handles macros
       including things defined by defclass, defvar, and defgeneric.
       From Helmut Eller.
     - Modular arithmetic implemented for x86 and sparc.  So 

         (ldb (byte 32 0) (+ x y))

       is a simple machine add instruction if x and y are
       (unsigned-byte 32).

  * Numerous ANSI compliance fixes:
     - Many bugs in CMUCL's type system detected by Paul Dietz'
       ANSI test suite have been fixed.
     - Non-keyword keyword names in lambda lists are now accepted.
     - Condition slot initialization fixed for the case that two
       or more slots have the same :INITARG.
     - Initforms are now evaluated once only during condition slot
       initialization.
     - USE-VALUE, STORE-VALUE, CONTINUE behavior corrected in presence
       of multiple restarts having the same name.
     - RESTART-CASE's interaction with local macros fixed.
     - Interaction of COMPUTE-RESTARTS and RESTART-CASE fixed in
       presence of multiple restarts having the same name.
     - Condition slot readers/writers are generic functions.
     - &WHOLE, &REST, &BODY in macro lambda lists can be followed
       by destructuring patterns.  The previous behavior of &BODY
       when followed by (<body> <declarations> <doc>) is available
       through the new lambda-list keyword EXT:&PARSE-BODY.
     - ELT now signals an error in safe code when used on lists
       with an out-of-bounds index.
     - Accessing arrays in compiled code with out-of-bounds indices
       now signals a TYPE-ERROR.
     - DECLARE no longer being a special operator.
     - DEFINE-CONDITION accepts slot option :DOCUMENTATION.
     - Behavior of (... &KEY &ALLOW-OTHER-KEYS) fixed for 
       DESTRUCTURING-BIND and macros.
     - Stricter checking of keyword argument list in CLOS methods.
     - &ENVIRONMENT variable bound bound before other vars, regardless
       of where it appears in a lambda-list.
     - VALUES types no longer accepting &KEY or &ALLOW-OTHER-KEYS.
     - THE conforming to ANSI.
     - SLOT-EXISTS-P can be used with conditions.
     - REMOVE-METHOD always returning the generic function passed to it.
     - Standard methods for STANDARD-OBJECT and STRUCTURE-OBJECT
       defined on MAKE-LOAD-FORM.
     - Compliant MAKE-LOAD-FORM-SAVING-SLOTS.
     - DEFSTRUCT recording slot-names as specified by users.
     - Generic function call keyword argument checking according to
       ANSI 7.6.5.
     - CALL-NEXT-METHOD without arguments being called with original
       arguments when method parameters are changed with SETQ.
     - When CALL-NEXT-METHOD is called with arguments, an error is
       signaled in safe code when the set of methods applicable to
       CALL-NEXT-METHOD's arguments is different from the set of
       methods applicable to the original method arguments.
     - FIND-METHOD signaling an error when called with specializers
       not corresponding to the number of required arguments of the
       supplied generic function.
     - TYPE-OF returning KEYWORD for keywords, STANDARD-CHAR for
       standard characters, BOOLEAN for T, (INTEGER <N> <N>) for
       integers N, non-list types for functions, and CLOS classes for
       instances whose class doesn't have a proper name.
     - Class precedence list of NULL not having SYMBOL before LIST,
       precedence list of STANDARD-METHOD not having METHOD before
       STANDARD-OBJECT.
     - ADD-METHOD returning its first argument, the generic function.
     - DEFCLASS redefining a class with the given name only if the
       name is the proper name of an existing class.
     - KEYWORD package no longer having nickname "".
     - COMMON-LISP-USER package no longer having nickname USER.
     - COMMON-LISP package no longer having nickname LISP; LISP
       and COMMON-LISP are now separate packages.
     - SLOT-VALUE, (SETF SLOT-VALUE), SLOT-BOUNDP, SLOT-MAKUNBOUND
       returning values specified by the standard when SLOT-UNBOUND
       or SLOT-MISSING are called and return.
     - NTH and NTHCDR accepting bignums as first argument.
     - First arg of CERROR may be a function.
     - Lambda-lists of the form (X . Y) in DESTRUCTURING-BIND and macros
       are now equivalent to (X &REST Y).
     - LAST, BUTLAST, NBUTLAST accepting bignum counts.
     - WITH-OUTPUT-TO-STRING accepting :ELEMENT-TYPE.
     - FIXNUM no longer naming a function.
     - MAKE-STRING-OUTPUT-STREAM accepts :ELEMENT-TYPE.
     - Numerous LOOP fixes.
     - On FreeBSD, CLtS-required floating-point conditions are signaled.
     - The minimum rehash-threshold is 0.1, and is enforced silently.
       This is to prevent overflows and divide-by-zero errors when
       creating a hash-table with a too-small rehash-threshold.
     - MACROLET now accepts declarations, as prescribed by ANSI CL.
     - MAKE-PATHNAME signals an error on invalid combinations
       (:absolute or :wild-inferiors followed by :back or :up).
     - PEEK-CHAR skips whitespace based on the readtable instead of
       using hard-wired values.
     - PEEK-CHAR signals an end-of-file error if necessary when
       RECURSIVE-P is true.
     - FILE-LENGTH now signals a TYPE-ERROR when the stream is not a
       stream associated with a file.
     - Floats are now printed and read accurately, preserving
       read/print consistency.  (Was sometimes off by a bit in the
       least significant bit.)
     - Printing of rationals should generate correct results now for
       all cases of *print-base* and *print-radix*.
     - Symbol printer should print better with unusual settings of
       *print-case* and friends.
     - Stream element-type can go up to 1024 bits.

  * Numerous bugfixes:
     - NSET-EXCLUSIVE-OR returns the same results as SET-EXCLUSIVE-OR
       wrt to duplicate elements.
     - Forms like (MULTIPLE-VALUE-BIND 'LIST) no longer trapping
       an internal compiler error when compiled.
     - PARSE-INTEGER signaling PARSE-ERROR.
     - Compiler eliminating some type checks in safe code.
     - Byte-compiler and -interpreter not handling C:&MORE.
     - Some TYPE-ERRORs being signaled with unbound slots, which
       made them undisplayable.
     - Structure predicates no longer signaling an error when applied
       to obsolete instances.
     - A DEFTRANSFORM for array bounds checking had an off-by-1 error.
     - Compiler was not properly deriving the type of FTRUNCATE in all
       cases.
     - Fixed error wherein certain uses of 2 arg ATAN caused a
       compiler error on x86.
     - Compiler wrongly signaling a type error in VOLATILE-INFO-LOOKUP or
       COMPACT-INFO-LOOKUP for hash values equal to MOST-POSITIVE-FIXNUM.
     - FFLOOR and FCEILING return types sometimes being inferred as
       being the nil type.
     - PARSE-TIME accepting GMT offsets with a leading plus sign.
     - UNIX-MMAP not accepting a null fd, preventing anonymous
       mappings.
     - SIGPROF having the wrong value on GNU/Linux.
     - Wrong byte-code being generated for (APPLY #'+ <X> <Y>) and
       similar cases.
     - GCD sometimes returning a negative value.
     - LCM sometimes returning a negative value, or signaling division-by-zero.
     - FLOAT sometimes returning a float of the wrong type.
     - ROOM should handle the large heaps now without overflowing
       internal counters or signaling type errors.  Columns should
       line up better too.
     - READ-SEQUENCE was returning too soon when reading from files
       into simple-strings even though there were data available.
     - READ-SEQUENCE was returning the wrong value when reading into
       simple-strings and when the :START value was not zero.
     - CMUCL was computing the wrong value for some special functions
       when the argument was on the branch cut for the function.
     - MULTIPLE-VALUE-BIND accepts &REST and friends in the var list.
     - When constructing logical pathnames, all components will be
       automatically upcased.  Needed for print/read consistency, and
       fixes an issue with ASDF and logical pathnames.
     - Handle things like [abc] in namestrings better to match
       Unix-style usage.
     - ATANH was incorrectly computing the value of 1+i*y.
     - SOFTWARE-VERSION would hang on some Linux 2.6 kernels because
       it was waiting forever reading /proc files.  Hopefully fixed
       for all kernels.
     - Reloading of shared libraries happened in the wrong order.

  * Other changes:
     - CREATE-REQUEST-SERVER has an additional :reuse-address keyword
       argument. 
     - Removed obsolete alias from MAKE-VECTOR to MAKE-ARRAY
     - the signature of *COMPILER-NOTIFICATION-FUNCTION* has changed;
       it now has an extra argument that describes the nature of the
       error or warning. 
     - File versioning occurs when the version component is :NEWEST,
       and Emacs style file versions are created:  foo.bar.~N~, where
       the larger N is newer.  foo.bar is the newest version.
     - Converting namestrings to pathnames now get a version component
       of NIL instead of :NEWEST.
     - *DEFAULT-PATHNAME-DEFAULTS* has a version component of
       :UNSPECIFIC instead of :NEWEST so versioning is disabled by
       default to preserve current behavior.  Set the version to
       :NEWEST to create versions.
     - LDB backtrace on x86.
     - DYNAMIC-EXTENT support on x86 and sparc.  Stack-allocation can
       be controlled by the setting of
       EXT:*TRUST-DYNAMIC-EXTENT-DECLARATIONS*.  See also the CMU User
       Manual.  Be warned that all bets are off if a stack-allocated
       object or parts of it escape; CMUCL might crash or format your
       hard disk.
     - Generational GC (gencgc) available for Sparc/Solaris on an
       experimental basis.  Despite the name and feature, it is not
       conservative. 
     - Search in /usr/lib/cmucl for the lisp core file as well, to
       support x86/Linux
     - The trap for floating-point underflow is turned off by default
       for all platforms now.
     - A continuable error is signaled when defining a logical
       pathname host that has the same name as a search-list host. 
     - A better sxhash function for string is implemented on sparc.
       This is based on the one-at-a-time hash from
       http://burtleburtle.net/bob/hash/doobs.html. 
     
  * Numerous improvements to the PCL implementation of CLOS:
     - Gerd's PCL has been added, which fixes numerous bugs and ANSI/
       AMOP non-compliances, and adds various new optimizations (also
       see the CMU User Manual).
     - PCL's class hierarchy has been changed to conform to AMOP, for
       instance, GENERIC-FUNCTION is now a subclass of
       STANDARD-OBJECT.
     - ENSURE-CLASS-USING-CLASS's argument list has been changed
       to conform to AMOP.
     - COMPUTE-EFFECTIVE-SLOT-DEFINITION now takes a slot name as
       argument, for AMOP compliance.
     - SUBTYPEP works in presence of forward-referenced classes.
     - Accessing slots of specialized parameters in (SETF
       SLOT-VALUE-USING-CLASS) methods works.
     - DIRECT-SLOT-DEFINITION-CLASS and EFFECTIVE-SLOT-DEFINITION-CLASS
       AMOP compliant.
     - ALLOCATE-INSTANCE working with structures defined with DEFSTRUCT.
     - With (DEFGENERIC FOO (&REST X &KEY)) (DEFMETHOD FOO (&REST X) X),
       (FOO 1) now signals an error.
     - Methods with &OPTIONAL arguments detecting too many args.
     - DEFMETHOD and DEFGENERIC detecting doubled lambda variables.
     - DEFGENERIC :ARGUMENT-PRECEDENCE-ORDER option checking.
     - SLOT-VALUE, (SETF SLOT-VALUE), SLOT-BOUNDP working with conditions.
     - (SETF SLOT-VALUE) working on read-only structure slots like
       in other implementations.
     - The declaration identifiers SLOTS, AUTO-COMPILE,
       NOT-AUTO-COMPILE have been moved from PCL to EXT.
     - Support for inlineing methods in effective methods; see the 
       CMU User Manual.
     - Wrong results could be returned from generic functions when
       all methods returned constant values, and some methods were
       qualified, or the generic function didn't use the standard
       method combination.
     - For standard method combination and method combinations defined
       with the short form of DEFINE-METHOD-COMBINATION: If there are
       no applicable primary methods when a generic function is
       called, the generic function PCL:NO-PRIMARY-METHOD is called.
       The default method of this generic function signals a
       continuable error of type PCL:NO-PRIMARY-METHOD-ERROR.
     - Effective methods are no longer precomputed at method load
       time, if the generic function has more than 
       PCL:*MAX-EMF-PRECOMPUTATION-METHODS* methods (default 100).
     - The loading of methods with invalid qualifiers no longer
       signals an error, for standard's conformance.  Instead, a
       warning is printed at method load time.  An error is
       signaled if the generic function is called with arguments such
       that the method is used.
     - Redefining a generic function with a different method
       combination now recomputes effective methods.
     - Slot access and gf-call optimizations are no longer performed
       for method parameters being assigned to in a method body, because
       we can't tell what type of object the method parameter actually
       contains.

  * Improvements to Hemlock, the Emacs-like editor:

  * Changes to rebuilding procedure:
     - When rebuilding CMUCL, *features* do not need to be set by
       hand.  As files are loaded, the appropriate *features* are set
       so that the resulting build has the same set of *features*.
       There should also be fewer warnings about internals.h changing,
       since that now just contains things in sys:*runtime-features*.
       Use sys:register-lisp-feature and
       sys:register-lisp-runtime-feature to set *features* and
       sys:*runtime-features*, respectively.
     - Pierre Mai's build-tools have been incorporated.  The scripts
       are in the src/tools directory.  These are slightly modified
       versions of his scripts.  Should make building CMUCL easier.

  * Deprecated features:
     - EXT:ENCAPSULATE and associated functions; use fwrappers instead.
     - The sparc port with the Cheney stop-and-copy garbage collector
       will no longer be supported after this release.  The supported
       GC will be gencgc.
     - The sparc port will only be supported on Ultrasparc machines
       (Sparc V9).

This release is not binary compatible with code compiled using CMUCL
18e; 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!

From: Rob Warnock
Subject: Re: [ANN] CMUCL 19a released
Date: 
Message-ID: <Kb-dnRUMJ8m1WI3cRVn-iA@speakeasy.net>
CMUCL Project  <··················@cons.org> wrote:
+---------------
| The CMUCL project is pleased to announce the release of CMUCL 19a.
+---------------

Yippee!

For those of us who've been using 19a-pre3, are there any
significant differences between 19a-pre3 & 19a-release?


-Rob

-----
Rob Warnock			<····@rpw3.org>
627 26th Avenue			<URL:http://rpw3.org/>
San Mateo, CA 94403		(650)572-2607
From: Eric Marsden
Subject: Re: [ANN] CMUCL 19a released
Date: 
Message-ID: <wzihdrj3zfz.fsf@melbourne.laas.fr>
>>>>> "rw" == Rob Warnock <····@rpw3.org> writes:

  rw> For those of us who've been using 19a-pre3, are there any
  rw> significant differences between 19a-pre3 & 19a-release?

  only minor changes:

     - DEFTYPE no longer emits spurious efficiency notes (when
       compiling optimized for speed)

     - fixed signalling of SIMPLE-STREAM-ERRORs (I/O errors on
       fd-streams were not being reported correctly)

     - improvements to the User's Manual

   I think that's all.

-- 
Eric Marsden                          <URL:http://www.laas.fr/~emarsden/>
From: Rob Warnock
Subject: Re: [ANN] CMUCL 19a released
Date: 
Message-ID: <SrmdncQ6lK2weI3cRVn-hw@speakeasy.net>
Eric Marsden  <········@laas.fr> wrote:
+---------------
| >>>>> "rw" == Rob Warnock <····@rpw3.org> writes:
|   rw> For those of us who've been using 19a-pre3, are there any
|   rw> significant differences between 19a-pre3 & 19a-release?
| 
|   only minor changes: ...[elided]...
+---------------

Thanks!


-Rob

-----
Rob Warnock			<····@rpw3.org>
627 26th Avenue			<URL:http://rpw3.org/>
San Mateo, CA 94403		(650)572-2607