X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=NEWS;h=8bd29adc64aee5432cae24d9db76fb015fd462d4;hb=c593dc26733b179db6c12c7085ed76b762ac256b;hp=e1ac5629a1fcb3701a2c0825b82134318f257ad1;hpb=d56803388a0a0a35e1889596f928b367c650c15b;p=sbcl.git diff --git a/NEWS b/NEWS index e1ac562..8bd29ad 100644 --- a/NEWS +++ b/NEWS @@ -1,15 +1,262 @@ ;;;; -*- coding: utf-8; fill-column: 78 -*- + * minor incompatible change: SB-THREAD:JOIN-THREAD-ERROR-THREAD and + SB-THREAD:INTERRUPT-THREAD-ERROR-THREAD have been deprecated in favor + of SB-THREAD:THREAD-ERROR-THREAD. + * new feature: SB-THREAD:SYMBOL-VALUE-IN-THREAD provides access to symbol + values in other threads. + * new feature: SB-INTROSPECT:ALLOCATION-INFORMATION provides information + about object allocation. + * optimization: more efficient type-checks for FIXNUMs when the value + is known to be a signed word on x86 and x86-64. + * improvement: failure to provide requested stack allocation compiler notes + provided in all cases (requested stack allocation not happening without a + note being issued is now considered a bug.) + * optimization: compiler is smarter about delegating argument type checks to + callees. + * bug fix: on 64 bit platforms FILL worked incorrectly on arrays with + upgraded element type (COMPLEX SINGLE-FLOAT), regression from 1.0.28.55. + (thanks to Paul Khuong) + * bug fix: looping around HANDLER-CASE could silently consume stack space + on each iteration. (reported by "foobar") + * bug fix: better error signalling when calls to functions seeking elements + from lists (eg. ADJOIN) are compiled with both :TEST and :TEST-NOT. + (reported by Tobias Rittweiler) + * bug fix: regressions in DIRECTORY from 1.0.28.61: pattern matching of + directory components now works as it used to. (various prolems reported by + Michael Becker, Gabriel Dos Reis, and Cyrus Harmon) + * bug fix: :PTY option in RUN-PROGRAM was broken with stream arguments. + (reported by Elliot Slaughter, thanks to Stas Boukarev) + * bug fix: bogus undefined variable warnings from fopcompiled references to + global variables. (thanks to Lars Rune Nøstdal) + * bug fix: foreign function names should now appear in backtraces on + FC6 as well. (reported by Tomasz Skutnik and obias Rautenkranz) + * bug fix: SETF compiler macro documentation strings are not discarded + anymore. + +changes in sbcl-1.0.29 relative to 1.0.28: + * IMPORTANT: bug database has moved from the BUGS file to Launchpad + https://bugs.launchpad.net/sbcl + Bugs can be reported directly there, or by sending email to + sbcl-bugs@lists.sourceforge.net + (no subscription required.) + * minor incompatible change: under weak type checking policy integer + types are weakened less aggressively. + * minor incompatible change: SAVE-LISP-AND-DIE :TOPLEVEL function is now + allowed to return, which causes SBCL to quit with exit status 0. Previously + if the function returned with a small integer return value, that value + was accidentally reused as the exit status. + * new feature: SB-EXT:DEFINE-HASH-TABLE-TEST allows defining new arguments + to MAKE-HASH-TABLE :TEST, and MAKE-HASH-TABLE has been extended with + :HASH-FUNCTION argument. Refer to user manual for details. + * new feature: SB-EXT:DEFGLOBAL macro allows defining global non-special + variables. + * new feature: SB-EXT:GET-TIME-OF-DAY provides access to seconds and + microseconds since the Unix epoch on all platforms. + * new feature: SB-EXT:ALWAYS-BOUND proclamation inhibits MAKUNBOUND, and + allows the compiler to safely elide boundedness checks for special + variables. + * new feature: SB-EXT:GLOBAL proclamation inhibits SPECIAL proclamations for + the symbol, prohibits both lexical and dynamic binding. This is mainly an + efficiency measure for threaded platforms, but also valueable in + expressing intent. + * new feature: UNC pathnames are now understood by the system on Windows. + * optimization: the compiler uses a specialized version of FILL when the + element type is know in more cases, making eg. (UNSIGNED-BYTE 8) case + almost 90% faster. + * optimization: accesses to potentially non-simple arrays where element type + is known are 50% faster. + * optimization: compiler now generates faster array typechecking code. + * optimization: ARRAY-DIMENSION is now faster for multidimensional and + non-simple arrays. + * optimization: multidimensional array accesses in the absence of type + information regarding array rank are approximately 10% faster due to + open coding of ARRAY-RANK. + * optimization: result of (FILL (MAKE-ARRAY ...) ...) and (REPLACE + (MAKE-ARRAY ...) ...) can be stack allocated if the result of MAKE-ARRAY + form can be. + * optimization: result of call to VECTOR can now be stack allocated. + * optimization: MAKE-ARRAY with :INITIAL-CONTENTS is now vastly faster + as long as the resulting array is one-dimensional and has a known + element type. In particular, :INITIAL-CONTENTS (LIST ...) where the + length of the list matches the known length of the vector does not + allocate the list as an intermediate step. Ditto for VECTOR and simple + backquoted forms. + * optimization: MAKE-ARRAY can now stack allocate in the presence of + :INITIAL-CONTENTS and :INITIAL-ELEMENT as long as the result has a + known element type, and is known to be simple and one dimensional. + * improvement: SBCL now emits a compiler note where stack allocation was + requested but could not be provided (not in all cases, unfortunately) + * improvement: better MACHINE-VERSION responses. (thanks to Josh Elsasser) + * improvement: pretty-printing loop has been implemented properly. (thanks + to Tobias Rittweiler) + * documentation: CLOS slot typechecing policy has been documented. + * bug fix: FILE-AUTHOR no longer signals an error on Windows. + * bug fix: SB-SPROF could be foiled by foreign code not have a frame + pointer, leading to memory faults. (thanks to Bart Botta) + * bug fix: better floating point exception handling on x86/OpenBSD. + (thanks to Josh Elsasser) + * bug fix: exit status from QUIT when called under --script was lost + (reported by Hubert Kauker) + * bug fix: MAKE-ARRAY for non-zero :INITIAL-ELEMENT always used the + same implementation of FILL to initialize the array, even if a more + efficient one was available (reported by Stas Boukarev, thanks to + Paul Khuong) + * bug fix: potential miscompilation of array stack allocation on x86 and + x86-64. (reported by Time Tossavainen) + * bug fix: some forms of AND, OR, and COND resulted in expansions that could + result in their subforms being treated as top level forms. (reported by + James Knight) + * bug fix: On x86/x86-64 alien functions declared to return integers shorter + than a machine register could leave garbage in the high bits of the + result register (bug 316325). + * bug fix: disable address space randomization Linux/x86-64 as well, + not just x86-64. (reported by Ken Olum) + * bug fix: Attempting to DEREF an (ALIEN (* T)) would produce a WARNING and + generate incorrect code. + * bug fix: #201; type inference for CONS and ARRAY types could derive + wrong results in the presence of eg. RPLACA or ADJUST-ARRAY. + * bug fix: special variables with a proclaimed specific subtype of FUNCTION + could not be assigned to or bound with PROGV. (reported by Lorenz + Mösenlechner) + * bug fix: the value of CL:- in the inspector was the previous expression + evaluated rather than the expression being evaluated. + * bug fix: constants can no longer be locally declared special. + * bug fix: signals delivered to threads started from foreign land (read: + directly by pthread_create, not by MAKE-THREAD) are redirected to a Lisp + thread by blocking all signals and resignalling. + * bug fix: SHARED-INITIALIZE initialized unbound :ALLOCATION :CLASS slots + from :INITFORM, if any. + +changes in sbcl-1.0.28 relative to 1.0.27: + * a number of bugs in cross-compilation have been fixed, with the ultimate + result that building under (at least) clisp should be much more reliable. + * minor incompatible changes: echo-streams now propagate unread-char to the + underlying input stream, and no longer permit unreading more than one + character. + * improvement: on x86/x86-64 Lisp call frames now have the same layout as C + frames, allowing for instance more reliable backtraces. + * improvement: the debugger REPL can now reference lexical variables + by name directly for code compiled with (DEBUG 3). + * improvement: errors from malformed declarations now have better source + paths associated with them. (thanks to Tobias Rittweiler) + * optimization: faster local calls on x86/x86-64 + * bug fix: some error messages for out-of-bound array indexes confused the + index and the bound. (thanks to Stas Boukarev) + * bug fix: pretty printing malformed DEFPACKAGE forms (thanks to Sidney + Markowitz) + * bug fix: running regressions tests in shells without OSTYPE set now works. + (reported by Harald Hanche-Olsen) + * bug fix: more robust static space exhaustion signalling from + MAKE-STATIC-VECTOR (thanks to Daniel Lowe) + * bug fix: (SETF DOCUMENTATION) for anonymous function now throws the + docstring away instead of storing it under names such as (LAMBDA (X)). + (reported by Leslie Polzer) + * bug fix: timers could go off in the wrong order, be delayed indefinitely + (thanks to Ole Arndt for the patch) + * bug fix: RESTART-FRAME and RETURN-FROM-FRAME stack corruption + * bug fix: the discriminating function for PRINT-OBJECT no longer preserves + potentially-invalid effective methods in its cache. + * bug fix: SB-INTROSPECT:FIND-DEFINITION-SOURCE now works with funcallable + instances as well (thanks to Paul Khuong) + * bug fix: using RUN-PROGRAM does not interfere with SB-POSIX:WAIT, + SB-POSIX:WAITPID and their C equivalents. + * bug fix: RUN-PROGRAM does not crash on Darwin when stressed. + +changes in sbcl-1.0.27 relative to 1.0.26: + * new port: support added for x86-64 OpenBSD. (thanks to Josh Elsasser) + * new port: support added for x86-64 Solaris. (thanks to Alex Viskovatoff) + * improvement: the system either recovers from stack exhaustion or dies + properly as opposed to leaving the user uncertain of whether the handler + trampled on some random memory next to the stack or having to rely on + --lose-on-corruption (which is still a good idea to use in production + because stack exhaustion can happen in signal handlers which will likely + lead to hangs.) + * bug fix: fix gc related interrupt handling bug on ppc (regression from + 1.0.25.37, reported by Harald Hanche-Olsen) + * bug fix: work around signal delivery bug in darwin (regression from + 1.0.25.44, reported by Sidney Markowitz) + * bug fix: fix ERROR leaking memory (reported by David Thompson) + +changes in sbcl-1.0.26 relative to 1.0.25: + * incompatible change: an interruption (be it a function passed to + INTERRUPT-THREAD or a timer function) runs in an environment where + interrupts can be enabled. The interruption can use + WITH-INTERRUPTS or WITHOUT-INTERRUPTS as it sees fit. Use + WITHOUT-INTERRUPTS to avoid nesting of interruptions and + potentially running out of stack. Keep in mind that in the absence + of WITHOUT-INTERRUPTS some potentially blocking operation such as + acquiring a lock can enable interrupts. + * incompatible change: GC-OFF and GC-ON are removed, as they were + always unsafe. Use WITHOUT-GCING instead. + * new feature: runtime option --disable-ldb + * new feature: runtime option --lose-on-corruption to die at the + slightest hint of possibly non-recoverable errors: running out of + memory, stack, alien stack, binding stack, encountering a memory + fault, etc. In the absence of --lose-on-corruption a warning is + printed to stderr. + * enhancement: detect binding stack exhaustion + * enhancement: detect alien stack exhaustion on x86/x86-64 + * improvement: generally more stable and reliable interrupt handling + * improvement: there is a per thread interruption queue, + interruptions are executed in order of arrival + * improvement: a repeating timer reschedules itself when the it has + finished, but expiration times are spaced equally. If an + expiration time is in the past it will trigger after a short grace + period that may give a chance to other things to run. + * optimization: slightly faster gc on multithreaded builds + * optimization: faster WITHOUT-GCING + * bug fix: when JOIN-THREAD signals an error, do it when not holding + important locks so that the debugger/handler doesn't produce + recursive errors or deadlock. + * bug fix: real-time signals are not used anymore, so no more + hanging when the system wide real-time signal queue gets full. + * bug fix: INTERRUPT-THREAD on a dying thread could produce memory + fault. + * bug fix: finalizers, gc hooks never run in a WITHOUT-INTERRUPTS + * bug fix: fix random memory faults related to interrupts on alpha + * bug fix: fix random memory faults related to interrupts on sparc + * bug fix: fix deadlocks related to starting threads + * bug fix: fix deadlines on locks on futex platforms + * bug fix: restore errno in signal handlers + * bug fix: fix deadlocks related to hash tables + * bug fix: fix deadlocks in pcl + changes in sbcl-1.0.25 relative to 1.0.24: * incompatible change: SB-INTROSPECT:FUNCTION-ARGLIST is deprecated, to be removed later. Please use SB-INTROSPECT:FUNCTION-LAMBDA-LIST instead. * new feature: SB-INTROSPECT:DEFTYPE-LAMBDA-LIST allows retrieval of DEFTYPE lambda lists. (thanks to Tobias Rittweiler) + * enhancement: MUTEX-VALUE is to be superseded by MUTEX-OWNER that has a + better name and does not return values so stale on multiprocessor systems. + Also, HOLDING-MUTEX-P was added for about the only sane usage of + MUTEX-OWNER. + * improvement: unithread builds keep track of MUTEX-VALUE. * improvement: reading from a TWO-WAY-STREAM does not touch the output stream anymore making it thread safe to have a concurrent reader and a writer, for instance, in a pipe. * improvement: GET-SETF-EXPANDER avoids adding bindings for constant arguments, making compiler-macros for SETF-functions able to inspect their constant arguments. + * improvement: COMPILE-FILE reports times with millisecond accuracy + (thanks to Luis Oliveira) + * optimization: CHAR-CODE type derivation has been improved, making + TYPEP elimination on subtypes of CHARACTER work better. (reported + by Tobias Rittweiler, patch by Paul Khuong) + * bug fix: setting alien structure fields of type struct by value now + computes the right offset for the memory copy. + * bug fix: compilation problem involving inlined calls to aliens with + result type VOID. (reported by Ken Olum) + * bug fix: #235a; sequential inline expasion in different policies no + longer reuses the functional from the previous expansion site. + * bug fix: DEFTYPE no longer breaks on bodies consisting of a single + unquoted symbol. Regression from 1.0.22.8. (reported by Ariel Badichi) + * bug fix: named ENUMs in multiply-referenced alien record types no longer + cause an implied type redefinition. Regression from 1.0.21.29. + * bug fix: sign-extension of small signed return values now works with C + code compiled with gcc 4.3 or newer. (reported by Liam Healy) + * improvements to the Windows port: + ** SB-BSD-SOCKETS now works from saved cores as well. (reported by Stephen + Westfold, thanks to Rudi Schlatte) changes in sbcl-1.0.24 relative to 1.0.23: * new feature: ARRAY-STORAGE-VECTOR provides access to the underlying data @@ -243,7 +490,7 @@ changes in sbcl-1.0.19 relative to 1.0.18: type is not know sufficiently well a compile-time are now compiled correctly. (reported by John Morrison) * bug fix: compiler no longer makes erronous assumptions in the - presense of non-foldable SATISFIES types. + presence of non-foldable SATISFIES types. * bug fix: stack analysis missed cleanups of dynamic-extent arguments in non-let-converted calls to local functions. * improvements to the Windows port: