X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=NEWS;h=c14cbadd4256f836500e806f0fad7aefa80ad0f3;hb=5ce77b3465434e396aa2d7670138a7e7741f3dae;hp=ce1611ef41884c1c2c5d5f8dec52767402cf5ee2;hpb=6cc71ab8ffad49f43895ad0a1df6885c81876687;p=sbcl.git diff --git a/NEWS b/NEWS index ce1611e..1a13633 100644 --- a/NEWS +++ b/NEWS @@ -2002,6 +2002,16 @@ changes in sbcl-0.8.3 relative to sbcl-0.8.2: obscure ANSI requirements changes in sbcl-0.8.4 relative to sbcl-0.8.3: + * incompatible change: The --disable-debugger command line + option now clobbers the debugger at a more fundamental + level, by redefining #'INVOKE-DEBUGGER instead of by + rebinding *DEBUGGER-HOOK*. The main difference is that BREAK + is specified by ANSI to ignore *DEBUGGER-HOOK* and + INVOKE-DEBUGGER regardless. Under the old system, BREAK would + enter the debugger REPL and then suffer recursive errors + because *DEBUG-IO* is also messed up in --disable-debugger mode; + while under the new system, BREAK in --disable-debugger mode + terminates the system just as an unhandled error would. * fixed compiler performance when processing loops with a step >1; * bug fix: DOCUMENTATION now retrieves generic function documentation. Also, DOCUMENTATION and (SETF DOCUMENTATION) @@ -2021,6 +2031,8 @@ changes in sbcl-0.8.4 relative to sbcl-0.8.3: * fixed bug 285: TRUNCATE on bignum arguments, and indeed bignum arithmetic in general, is now much more reliable on the PPC platform. + * bug fix: LOGCOUNT on (UNSIGNED-BYTE 64) objects on the Alpha + platform now returns the right answer. * optimization: restored some effective method precomputation in CLOS (turned off by an ANSI fix in sbcl-0.8.3); the amount of precomputation is now tunable. @@ -2029,12 +2041,42 @@ changes in sbcl-0.8.4 relative to sbcl-0.8.3: performance of the compiler by about 20%. * optimization: performance of FILL (and :INITIAL-ELEMENT) on simple-base-strings and simple-bit-vectors is improved. + * optimization: the optimization of 32-bit logical and arithmetic + functions introduced in version 0.8.3 on the x86 has been + implemented on the mips, ppc and sparc platforms; an + implementation of the same facility, but for 64-bit arithmetic, + has been added for the alpha. * microoptimization: the compiler is better able to make use of the x86 LEA instruction for multiplication by constants. * bug fix: in some situations compiler did not report usage of generic arithmetic in (SPEED 3) policy. * bug 145b fix: compiler used wrong type specifier while converting MEMBER-types to numeric. + * bug fix: COMPILE-FILE must bind *READTABLE*. (reported by Doug + McNaught) + * bug fix: (SETF AREF) on byte-sized-element arrays with constant + index argument now works properly on the Alpha platform. + * bug fix: floating point exception treatment on the Alpha platform + is improved. + * bug fix: FILE-POSITION works much better on string input and + output streams. (thanks to Nikodemus Siivola) + * bug fix: many threading/garbage collection symptoms sorted. + SB-THREAD:INTERRUPT-THREAD now safe to call on a thread that + might be pseudo-atomic. + * internal change: Stopping for GC is now done with signals not + ptrace. GC is now done in whichever thread wanted it, instead of + in the parent. + * bug fix: GC hooks (missing since 0.8) reinstated, so finalizers + work again. + * bug fix: result form in DO is not contained in the implicit + TAGBODY. + * incompatible change: ICR structure is changed; the value part of + CONTINUATION is now called LVAR; corresponding functions are + renamed (e.g. SB-C::CONTINUATION-TYPE has become SB-C::LVAR-TYPE). + * added type deriver for ISQRT (thanks to Robert E. Brown). + * bug fix: better support for loading from the command line when an + initialization file sets (READTABLE-CASE *READTABLE*). (thanks + to Adam Warner) * fixed some bugs revealed by Paul Dietz' test suite: ** the RETURN clause in LOOP is now equivalent to DO (RETURN ...). ** ROUND and FROUND now give the right answer when given very @@ -2053,6 +2095,379 @@ changes in sbcl-0.8.4 relative to sbcl-0.8.3: ** unsigned addition of a 32-bit constant with the high bit set no longer causes an internal compiler error on the x86. ** LOGBITP accepts a non-negative bignum as its INDEX argument. + ** compiler incorrectly derived types of DPB and DEPOSIT-FIELD + with negative last argument. + ** byte specifiers with zero size and position no longer cause + an error during type derivation. + ** bignum multiplication on the Alpha platform now returns the + right answer. + * porting: The system now builds on SuSE AMD64, although it still + generates a 32-bit binary. + * .fasl file incompatibility: The fasl file version number has + been incremented (because of the changes to internal compiler + data structures referred to above). + +changes in sbcl-0.8.5 relative to sbcl-0.8.4: + * New code in contrib/sb-introspect (still probably not entirely + stable yet) provides some support for smart Lisp development + environments like SLIME. + * The conditions signalled for errors occurring when loading .fasl + files have been systematized (inheriting from SB-EXT:INVALID-FASL) + in a way which should help ASDF recover gracefully. + * The REQUIRE/PROVIDE behavior of *MODULE-PROVIDER-FUNCTIONS* + stuff has been cleaned up. If you code contrib/ stuff, this might + affect you, and you can look at contrib/README, contrib/STANDARDS, + and/or the 0.8.4.27 diff to check. + * In full calls the compiler now does not generate checks for declared + argument types for all arguments. + * various threading fixes + ** and some experimental patches which didn't make it into + the main tree for this release, but which are shipped in + contrib/experimental-thread.patch as a possible fix for some + failures (deadlock, spinning...) in GC-intensive multithreaded + applications. + * fixed PPC build problem (source code incompatibility of different + library versions): added offsetof() hackery which attempts to divine + where glibc maintainers put uc_mcontext today + * fixed bug 282: compiler does not trust type assertions while passing + arguments to a full call. + * fixed bug 261: compiler allows NIL or "no value" to be accepted for + &OPTIONAL VALUES type parameter. + * fix bug 214: algorithm for noting rejected templates is now more + similar to that of template seletion. (also reported by rydis on + #lisp) + * fixed bug 141b: printing backquoted information readably and prettily + inserts a space where necessary. + * bug fix: obviously wrong type specifiers such as (FIXNUM 1) or + (CHARACTER 10) are now reported as errors, rather than propagated + as unknown types. (reported by piso on #lisp) + * bug fix: the :IF-EXISTS argument to OPEN now behaves correctly + with values NIL and :ERROR. (thanks to Milan Zamazal) + * fixed bug 191c: CLOS now does proper keyword argument checking as + described in CLHS 7.6.5 and 7.6.5.1. + * bug fix: LOOP forms using NIL as a for-as-arithmetic counter no + longer raise an error; further, using a list as a for-as-arithmetic + counter now raises a meaningful error. + * fixed bug 213a: even fairly unreasonable CONS type specifiers are + now understood by sequence creation functions such as MAKE-SEQUENCE + and COERCE. + * fixed bug 46k: READ-BYTE now signals an error when asked to read from + a STRING-INPUT-STREAM. + * compiler enhancement: SIGNUM is now better able to derive the type + of its result. + * type declarations inside WITH-SLOTS are checked. (reported by + salex on #lisp) + * fixed some bugs revealed by Paul Dietz' test suite: + ** incorrect optimization of TRUNCATE for a positive first + argument and negative second. + ** compiler failure in let-convertion during flushing dead code. + ** compiler failure while deriving type of TRUNCATE on an + interval, containing 0. + ** ASH of a negative bignum by a negative bignum count now returns + -1, not 0. + ** intersection of CONS types now canonicalizes properly, fixing + inconsistencies in SUBTYPEP. + +changes in sbcl-0.8.6 relative to sbcl-0.8.5: + * fixed a bootstrapping bug: the build process no longer assumes + that the various BOOLE-related constants have the same value in + host and target lisps. (noted by Paul Dietz' test suite on an + SBCL binary built from CLISP) + * The system can now be dynamically linked on the MIPS platform, + which enables dynamic loading of foreign code from Lisp. (thanks + to Ralf Baechle for discussions on the MIPS ABI) + * The system now records debugging information for its own source + files in a filesystem-position-independent manner, relative to + the "SYS" logical host. + * fixed a compiler bug: MV-LET convertion did not check references + to the "max args" entry point. (reported by Brian Downing) + * tweaked disassembly notes to be less confident about proclaiming + some instruction as an LRA. (thanks to Brian Downing) + * contrib update: SB-ACLREPL is now threadsafe; multiple listeners + now each have their own history, command character, and other + characteristics. (thanks to David Lichteblau) + * fixed some bugs revealed by Paul Dietz' test suite: + ** compiler failure in compiling LOGAND expressions including a + constant 0 + ** Implementation of ASH-MOD32 on X86 and PPC did not work for the + shift greater than 32. + ** FLUSH-DEST did not mark blocks for type check regeneration. + ** HANDLER-CASE failed to accept declarations in handler clauses + in some circumstances. + +changes in sbcl-0.8.7 relative to sbcl-0.8.6: + * When built with the :SB-FUTEX feature, threaded builds now take + advantage of the "fast userspace mutex" facility in Linux kernel 2.6 + for faster/more reliable mutex and condition variable support. + * Incompatible change (but one you probably shouldn't have been using + anyway): the interface and code for arbitrating between multiple + threads in the same user session has been redesigned. + * bug fix: GET-SETF-EXPANSION no longer throws an internal type + error when called without an explicit environment argument. + (thanks to Nikodemus Siivola) + * bug fix: buffered :DIRECTION :IO streams are less likely to become + confused about their position. (thanks to Adam Warner and Gerd + Moellmann) + * bug fix: Pretty printing backquoted forms with unquotations in the + argument list position of various code constructs such as LAMBDA + now works correctly. (reported by Paul Dietz) + * bug fix: Pretty printing unquotations no longer loses all + stream position information. + * optimization: performance of string output streams is now less + poor for multiple small sequence writes. + * optimization: performance of CSUBTYPEP in the presence of complex + expressions involving CONS and NOT many times has been improved. + (reported by Paul Dietz) + * ASDF-INSTALL bug fix: now parses *PROXY* properly. (thanks to + Sean Ross) + * SB-SIMPLE-STREAMS enhancement: simple-streams can now be used as + streams for the REPL, for the debugger, and so on. (thanks to + David Licteblau) + * DEFINE-CODITION is more efficient. (thanks to Brian Mastenbrook) + * fixed some bugs revealed by Paul Dietz' test suite: + ** the value of the :REHASH-THRESHOLD argument to MAKE-HASH-TABLE + is ignored if it is too small, rather than propagating through + to cause DIVIDE-BY-ZERO or FLOATING-POINT-OVERFLOW errors. + ** extremely complex negations of CONS types were not being + sufficiently canonicalized, leading to inconsistencies in + SUBTYPEP. + ** VALUES tranformer lost derived type. + +changes in sbcl-0.8.8 relative to sbcl-0.8.7: + * minor incompatible change: parsing of namestrings on a physical + (Unix) host has changed; numbers after the final #\. in a + namestring are no longer interpreted as a version field. This is + intented to be largely invisible to the user, except that the + meaning of the namestring "*.*.*" has changed: it now refers to a + pathname with :TYPE :WILD :NAME #. This namestring + should usually be replaced by + (make-pathname :name :wild :type :wild :version :wild) + with the added benefit that this is more likely to be portable. + As a consequence of this change, the :IF-EXISTS :NEW-VERSION + option to OPEN now signals an error if the file being opened + exists; this may have an impact on existing code. + * fixed bug 190: RUN-PROGRAM should now work properly, respecting + signals received, on the PowerPC platforms (both Linux and + Darwin). (thanks to Pierre Mai for pointing out the location of + the error) + * several fixes on OS X: The system now builds and runs cleanly on + Panther (10.3), and works around sigreturn bug (no more SIGFPEs). + (thanks to Brian Mastenbrook) + * bug fix: DECODE-UNIVERSAL-TIME now accepts timezone arguments with + second-resolution: integer multiples of 1/3600 between -24 and 24. + (thanks to Vincent Arkesteijn) + * bug fix: functions =, /=, <, <=, >, >= did not check the argument + type when called with 1 argument; PEEK-CHAR checked type of + PEEK-TYPE only after having read first character from a + stream. (reported by Peter Graves) + * bug fix: the garbage collector now has much better locality + behaviour, and in particular no longer treats all memory as being + exhausted when many small objects point to each other in a deeply + recursive manner. + * bug fix: arrays specialized on (UNSIGNED-BYTE 15) are now + recognized as being TYPEP their class. + * bug fix: the PUSHNEW documentation string has been corrected. + (thanks to Vincent Arkesteijn) + * bug fix: defaulting of the value for the last of an atypically + large number of multiple values being bound was not being + performed correctly on the Alpha or PPC platforms + * optimization: implemented multiplication as a modular + (UNSIGNED-BYTE 32) operation on the x86 backend. + * optimization: SEARCH on simple-base-strings can now be open-coded. + (see also contrib/compiler-extras.lisp for inspiration for + teaching the compiler about the Boyer-Moore algorithm). + * value, returned by MAX (and MIN) called with several EQUALP, but + not EQL, arguments now does not depend on compiler settings. + * fixed some bugs revealed by Paul Dietz' test suite: + ** in stack analysis liveness information is propagated from + non-local entry points. + ** pathwise CAST removing failed when the CAST node did not start + a block. + ** INPUT-STREAM-P, OUTPUT-STREAM-P, STREAM-ELEMENT-TYPE and + OPEN-STREAM-P signal a TYPE-ERROR if their argument is not a + stream. + ** LOAD-LOGICAL-PATHNAME-TRANSLATIONS returns NIL if the logical + host is already defined. + ** RENAME-FILE works on streams instead of signalling an internal + type error. + ** PEEK-CHAR uses the current readtable when determining whether + or not a character is whitespace. + ** MERGE-PATHNAMES handles the case when the pathname does not + specify a name while the default-pathname specifies a version. + ** Pathnames now stand a better chance of respecting print/read + consistency. + ** Attempting to use standardized file system operators with a + pathname with invalid :DIRECTORY components signals a + FILE-ERROR. + ** OPEN :DIRECTION :IO no longer fails to work on non-existent + files. + ** DIRECTORY on logical pathnames is more correct. + ** CLEAR-INPUT, CLEAR-OUTPUT, FINISH-OUTPUT and FORCE-OUTPUT + signal a TYPE-ERROR if their argument is not a stream. + ** READ-BYTE and WRITE-BYTE signal a TYPE-ERROR if their stream + designator argument does not designate a stream. + ** OPEN-STREAM-P and INPUT-STREAM-P on synonym streams work by + examining the synonym. + ** STREAM-ELEMENT-TYPE and FRESH-LINE on broadcast-streams now + work as specified. + ** OPEN and WITH-OPEN-STREAM allow opening streams with + element-type larger than ([UN]SIGNED-BYTE 32). + +changes in sbcl-0.8.9 relative to sbcl-0.8.8: + * deprecation of old extension: *DEBUG-PRINT-LEVEL* and + *DEBUG-PRINT-LENGTH* are now deprecated in favor of the new, more + general SB-DEBUG:*DEBUG-PRINT-VARIABLE-ALIST* mechanism. (This + should matter to you only if you rebind the printer control + variables and then find you want different bindings in the + debugger than in the ordinary execution of your program.) + * The runtime build system has been tweaked to support building + (on SPARC/SunOS) using a C compiler which invokes Sun's own + assembler and linker. (thanks to Nikodemus Siivola) + * Unbound, undefined, undeclared variables now trigger full + WARNINGs, not just STYLE-WARNINGs, on the assumption that this is + more often programmer error than deliberate exploitation of undefined + behaviour. + * optimization: the hash algorithm for strings has changed to one + that is less vulnerable to spurious collisions. (thanks to Juho + Snellman) + * optimization: VECTOR-POP, VECTOR-PUSH-EXTEND and REPLACE do less + needless bounds checking. (thanks to Juho Snellman) + * optimization: implemented multiplication as a modular + (UNSIGNED-BYTE 32) operation on the PPC backend. + * fixed some bugs revealed by Paul Dietz' test suite: + ** ADJUST-ARRAY now copies the datum in a zero rank array if + required. + ** ADJUST-ARRAY no longer adjusts non-adjustable arrays. + ** MAKE-STRING-INPUT-STREAM accepts NIL as a value for :END. + ** MAKE-STRING-INPUT-STREAM functions correctly for strings with + element-type NIL. + ** CLEAR-INPUT accepts NIL and T for its (optional) stream + argument. + ** Ratios can now be printed correctly with *PRINT-BASE* bound to + 2, 8 or 16. + ** ECHO-STREAMs no longer attempt to echo the end of file value to + their output stream on EOF from read. + ** CONCATENATED-STREAM-STREAMS discards constituent streams which + have been read to end-of-file. + ** CLOSE works as expected on the null CONCATENATED-STREAM, and on + STRING-STREAMS. + ** Printing symbols with *PRINT-CASE* :CAPITALIZE respects the + description of determination of which consecutive characters + constitute a word. + ** Printing the "Space" character with escaping on now yields "#\\ ", + rather than "#\\Space", as mandated by ANSI 22.1.3.2. + ** Reading floating-point numbers with *READ-BASE* set to a number + less than 10 works correctly. + ** Reading floating-point numbers with *READ-BASE* set to a number + more than 10 works correctly. + ** Printing with *PRINT-READABLY* targets the standard readtable, not + the readtable currently in effect. + +changes in sbcl-0.8.10 relative to sbcl-0.8.9: + * Support for the forthcoming 2.0 version of the NetBSD kernel + running on x86 hardware has been added. (thanks to Perry + E. Metzger most immediately, and others for their past work) + * SBCL now runs on OpenBSD 3.4. (Thanks to Scott Parish; 3.4 is the + current release version; SBCL's OpenBSD support had been broken + since about the time of OpenBSD's switch to ELF binary format.) + * [placeholder for DX summary] + ** user code with &REST lists declared dynamic-extent, under high + speed or space and low safety and debug optimization policy. + * The manual has been converted to Texinfo format and the debugger + chapter from the cmucl manual has been added. + * A facility has been added to extract documentation strings from + sbcl and store them as Texinfo-formatted snippets for inclusion in + the manual (via Texinfo's @include directive) + * bug fix: compiler emitted division in optimized DEREF. (thanks for + the test case to Dave Roberts) + * bug fix: multidimensional simple arrays loaded from FASLs had fill + pointers. (reported by Sean Ross) + * bug fix: PROFILE output is printed nicely even for large numerical + values. (thanks to Zach Beane) + * bug fix: streams with element-type (SIGNED-BYTE ) for + greater than 32 handle EOF correctly. + * bug fix: on X86 an immediate argument of the IMUL instruction is + correctly printed in disassembly. (thanks to Lutz Euler) + * bug fix: class slots in redefined classes preserve their old + values. (thanks to Bruno Haible and Nikodemus Siivola) + * bug fix: compilation of funcalls of CXX+R (e.g. CDDR) now + succeeds. (reported by Marco Baringer) + * fixed some bugs revealed by Paul Dietz' test suite: + ** READ-SEQUENCE now works on ECHO-STREAMs. + ** RATIONALIZE works more according to its specification. (thanks + to Bruno Haible) + +changes in sbcl-0.8.11 relative to sbcl-0.8.10: + * new feature: the SB-EXT:MUFFLE-CONDITIONS declaration should be + used to control emission of compiler diagnostics, rather than the + SB-EXT:INHIBIT-WARNINGS OPTIMIZE quality. See the manual for + documentation on this feature. The SB-EXT:INHIBIT-WARNINGS + quality should be considered deprecated. + * fixed bug 320: Shared to local slot value transfers in class + redefinitions now happen corrently from superclasses as + well. (reported by Bruno Haible) + * fixed bug 316: SHIFTF now accepts VALUES forms. (reported by Bruno + Haible) + * fixed bug 322: DEFSTRUCT :TYPE LIST type predicates now handle + improper lists correctly. (reported by Bruno Haible) + * fixed bug 313: source-transform for was erroneously + applied to a call of a value of a variable with name . + (reported by Antonio Menezes Leitao) + * fixed bug 307: The obsolete instance protocol ensures that + subclasses are properly obsoleted. (thanks to Nikodemus Siivola) + * on X86 and Alpha fixed bug 298, revealed by Paul F. Dietz' test + suite: SBCL can remove dead unknown-values globs from the middle + of the stack. + * added a new restart to *BREAK-ON-SIGNALS* handling to make it + easier to resume long computations after using *BREAK-ON-SIGNALS* + to diagnose and fix failures. (thanks to Nikodemus Siivola) + * fixed bug reported by PFD in lisppaste #747 (and Bruno Haible from + CLISP test suite): WRITE-TO-STRING is not constant-foldable. + * fixed bugs in COMPLEX type specifier: UPGRADED-COMPLEX-PART-TYPE + is now consistent with (COMPLEX ); bugs in treatment of COMPLEX + MEMBER and UNION types have likewise been fixed. (thanks to Bruno + Haible) + * fixed a (fairly theoretical) bug in string printing: if + *PRINT-READABLY* is true, signal PRINT-NOT-READABLE if the string + does not have array-element-type equal to the most general string + type. + * fixed bug: SET-PPRINT-DISPATCH does not immediately resolve + function name. (thanks to Nikodemus Siivola) + * fixed bug: compile-time format string checker failed on + non-closed ~{. (reported by Thomas F Burdick) + * fixed bug: as reported by Kalle Olavi Niemitalo on #lisp IRC, + don't warn on legal loop code involving more than one aggregate + boolean. (thanks to Nikodemus Siivola) + * fixed bug: as reported by Peter Graves on #lisp IRC, passing a NIL + in keyword position to MAKE-PACKAGE was not being reported as + invalid. + * fixed bug: as reported by Juan Ripoll on cmucl-imp, + MULTIPLE-VALUE-BIND should be able to lexically bind lambda list + keywords. + * optimization: rearranged the expansion of various defining macros + so that each expands into only one top-level form in a + :LOAD-TOPLEVEL context; this appears to decrease fasl sizes by + approximately 10%. + * fixed some bugs revealed by Paul Dietz' test suite: + ** MAKE-INSTANCES-OBSOLETE returns the class name when called with + a symbol. + ** Fixed an optimization bug related to inheritance of initforms + from local to shared slots. + ** FILE-POSITION works as specified on BROADCAST-STREAMs. + ** CAST optimizer forgot to flush argument derived type. + ** print/read consistency on floats is now orders of magnitude + more likely. (thanks also to Bruno Haible for a similar report + and discussions) + ** removed stack cleaning in the cleanup part of UNWIND-PROTECT. + ** IMAGPART is specified (infelicitously) to return (* 0 ) + for objects of type REAL. Make it so. + ** SXHASH is specified (infelicitously) to respect similarity, + which means that (SXHASH 0.0) must equal (SXHASH -0.0). Make + it so. (thanks to Markus Ziegler) + ** on the Alpha, the compiler succeeds in compiling functions + returning a known number of arguments greater than 63. + ** fixed handling of invalid NIL arguments in keyword position + in local calls. planned incompatible changes in 0.8.x: * (not done yet, but planned:) When the profiling interface settles