News **** 0.15 ==== * Parser can now associate lexical location (file name and line number) with parsed objects. Added `call-with-lexical-origins' and `lexical-origin' functions. This adds memory overhead but is only enabled in interpreted mode, or other times it could be useful (e.g. when compiling) * The compiler enables line-numbering, and uses the information when it prints errors. It also prints errors in a more standard format (intended to mimic GCC), and distinguishes warnings from errors * Debugger is much improved, and supports emacs-style emission of line number tokens. Use the included `rep-debugger.el' elisp code to source-debug rep programs in Emacs! * New command line option `--debug'. When given, rep starts up in the debugger * Reformatted backtrace output. Also backtraces only ever include evaluated argument lists now. They also include lexical information when possible * Syntax errors include error description and line number * Now supports weak reference objects. New functions `make-weak-ref', `weak-ref', `weak-ref-set'. A weak reference is a pointer to another object. When that object is garbage collected, the pointer in the weak reference is set to false. * New `error helper' module. When an error is handled, this module is called and tries to print a human-understandable message explaining why the error may have occurred * REPL commands may now be specified by their shortest unique set of leading characters, e.g. `,o' instead of `,open' * Added an `#undefined' value. Returned by `%define' and the macros using it (`defun', `defmacro', etc...) * New function `table-size' in module `rep.data.tables' * `thread-suspend' returns true iff the timeout was reached (i.e. return false if `thread-wake' was used to unsuspend the thread) * Objects defined using the `object' macro now have an implicit `self' binding - the object representing their self (or their most derived self) * Added TIMEOUT parameter to `condition-variable-wait' and `obtain-mutex' functions * New `rep.threads.message-port' module, implements a simple message queue for threads * `log' function now optionally accepts a second argument, the base of the logarithm * Use gmp to generate random numbers when possible (if at least version 3 of gmp is found) [disabled in librep 0.15.1 - gmp seems to be buggy?] * The `string-replace' function may now be given a function as its TEMPLATE parameter * Bug fixes: - Signal an error if writes don't write all characters they were asked to. Also, some functions could write fewer characters than they were supposed to even if no errors occurred - Remembered that file sizes may not fit in fixnums - Don't preserve trailing slashes in results of canonical-file-name (to make the path canonical) - Don't signal an error when end of file is encountered immediately after reading `#\X' syntax - `current-thread' and `all-threads' will create a thread object for the implicit thread if there isn't one already - In C subrs that take optional integer arguments, signal an error if the given value isn't an integer or undefined (false). Also, accept all types of numbers where it makes sense to do so - Signal an error if end of file is read while parsing a block comment - Don't ever return a null object from `current-time-string' - Catch errors signalled during command line option processing, and pass them to the standard error handler - Right hand side of `letrec' bindings may now have more than one form - The `object' macro now evaluates its BASE-OBJECT parameter exactly once - Finally removed `define-value' - Ignore null lines (or lines which only have comments) in the repl - In the compiler, don't expand macros which have have been shadowed by local bindings - Don't print some compiler errors/warnings twice - Fixes for `mips-compaq-nonstopux' architecture (Tom Bates) - Fixed `,reload' and `,unload' repl commands not to try to remove non-existent structures 0.14 ==== * New module `rep.util.md5', has two functions for generating MD5 message digests (of files or strings) * Changes to the `rep.io.sockets' function: In the `socket-server' function the HOST and/or PORT arguments may be false, meaning to listen on all addresses and to choose a random unused port. New functions `socket-peer-address' and `socket-peer-port', these always returns the details of the far end of the connetion. `socket-address' and `socket-port' have been changed to always return the details of the local connection point. * New function in `rep.system' module, `crypt'. A wrapper for the system's `crypt' function (if it has one) * New function in `rep.threads' mdoule, `make-suspended-thread' * New module `rep.net.rpc', provides a text-stream based RPC mechanism for Lisp programs. Similar in some ways to untyped CORBA. (This is still in the experimental stage - its interface may change in forthcoming releases) * New functions in `rep.data' module, `list->vector' and `vector->list' * New macro `define-special-variable'. A combination of `defvar' and `setq' - it always makes the variable special and it always sets it to the given value * New module `rep.test.framework' implementing `assert', `check' and `test' macros. This provides a framework for implementing unit tests in Lisp modules (such that running the interpreter with the `--check' option will run all tests that have been set up to be autoloaded * Bug fixes: - When reading from strings, don't choke on zero bytes - When writing into sockets, be aware that the socket is in non-blocking mode - SDBM and GDBM modules now close any open databases before the interpreter exits - Fixed the `rep_parse_number' function not to require a terminating null character in the string when parsing bignums - Only define `Qrep_lang_interpreter' once - Don't assign vm registers to physical registers on 68000 architectures - it's been reported to crash - When running asynchronous subprocesses, open and initialize the pty slave before forking to avoid a race condition with the child process - Flush symbols from the module cache at another point - Fixes for Unixware - When compiling non-top-level `defvar' forms, add any doc string they have to the database 0.13.5 ====== * Tar file handling no longer requires GNU tar * The `defvar' special form can now take only a single argument * The reader now treats `#\return' characters as white space * Other miscellaneous bug fixes... 0.13.4 ====== * Don't restrict symbols exported from plugin libraries, some need to export symbols to work properly (this bug only seemed to appear on Solaris systems) * Added `rep_file_type' and `rep_guardian_type' to the list of symbols exported from librep * Fixed the `install-aliases' script (Peter Teichman) * New module `rep.threads.condition-variable' * Added `string-split' and `string-replace' to the gaol 0.13.3 ====== * Try to only export public symbols from `librep.so' and modules * When expanding file names translate `/..' to `/' * Set an upper bound on the allowed recursion depth when regexp matching, prevents the stack from overflowing in pathological cases * Added optional second arg to `readline' function, a function to call to generate completions. The `rl-completion-generator' method of supplying this function is deprecated * Fixed bugs when handling character-case in regexp module (Andrew Rodionoff) * Added an `premature-end-of-stream' error. This is signalled instead of `end-of-stream' when reading characters in the middle of a syntax form. The `end-of-stream' error is only signalled when the end of the stream is reached before encountering anything other than whitespace characters * Fixed bug of expanding declarations in the `define' macro expansion 0.13.2 ====== * Fix `define' so that it tracks bound variables and ignores shadowed keywords when traversing code * Added checks to compilation process for the kind of missing shared-library problems that many people see * Fixed the `install-aliases' shell script * New configure option: `--enable-full-name-terminator' 0.13.1 ====== * Added functions `remove-if' and `remove-if-not' * Various bug-fixes for non-linux or solaris systems (John H. Palmieri, Philippe Defert) * `#f', `#t', `#!optional', `#!key' and `#!rest' are now uninterned symbols. Keywords are interned in a separate obarray * Fixed bug of caching regexps even when their string has been modified * Fixed some bugs in the ftp remote file handler and the `pwd-prompt' function * Fixed `define' to ignore `structure' and `define-structure' forms 0.13 ==== * The end-of-list / boolean-false object is no longer the symbol `nil'. Instead there is a special object `()' fulfulling these two roles. For modules importing the `rep' module, the symbol `nil' evaluates to `()'. This allows the `scheme' module to be more compliant with the Scheme standard * Parameter list changes: - Deprecated `&optional' and `&rest', in favour of `#!optional' and `#!rest'. - Added keyword parameters. Use `#!key' to declare them. Keyword syntax is `#:PARAM'. For example: ((lambda (#!key a b) (list a b)) #:b 2 #:a 1) => (1 2) - `#!optional' and `#!key' parameters may now have default values, syntax is `(VAR DEFAULT)'. For example: ((lambda (#!optional (a 1)) a)) => 1 * The module namespace is now hierarchical. `.' characters in module names denote directory separators, e.g. `foo.bar' translates to the file `foo/bar' All module names prefixed with `rep.' are reserved for librep, other top-level names should be picked to be as unique as possible The existing modules have been renamed to fit this scheme (see the file `TREE' in the distribution for the hierarchy details). However, old module names will still work for the time being * The `rep' module no longer includes the `rep.regexp', `rep.system', `rep.io.files', `rep.io.processes' or `rep.io.file-handlers' modules. These need to be imported explicitly * Doc strings are now indexed by module name as well as symbol name. The `define' macro now takes a doc string as its optional third parameter * Record constructors may include all lambda-list keywords (e.g. keywords and/or default values) * Incompatible virtual machine changes, hence bytecode files will need to be recompiled. Improvements include: - Only heap-allocate variables when absolutely necessary - Closure analysis to allow inlining of some types of `letrec' expressions - Added a `safe' virtual machine, which makes no assumptions regarding validity of bytecode, so is safe for untrusted code * Added an `unscheme' module. Another Scheme implementation, but the goal of this one is to integrate cleanly with the librep runtime environment, even if this is at the expense of R4RS compliance For example, in `unscheme' code, `#f => ()' and `#t => t'. This allows rep and unscheme functions to call each other without needing to convert any data * By default, it is now illegal to modify top-level variables that have not previously been defined * New macro `define-structures' to export multiple views of a single underlying environment * The librep runtime no longer handles the `--help' option itself, this should be done by scripts * Don't search `$LD_LIBRARY_PATH' for plugins, but prepend all directories in colon-separated `$REP_DL_LOAD_PATH' to `dl-load-path'. Similarly, the contents of `$REP_LOAD_PATH' is prepended to `rep-load-path' * `(/ X) => (/ 1 X)' * Extra string-manipulation functions: `string-replace', `string-split' (in the `rep.regexp' module) * `#f' and `#t' are now primitive symbols, not special objects * Special case tail-recursive calls to `apply', to ensure they get eliminated * The `0x123' and `0123' read syntaxes have been deprecated, use `#x123' and `#o123' instead * `#| ... |#' comments now nest correctly * New modules: `rep.i18n.gettext', `rep.vm.safe-interpreter', `rep.vm.assembler', `unscheme', `rep.data.objects', `rep.www.quote-url', `rep.www.fetch-url', `rep.util.ispell', `rep.util.base64', `rep.util.autoloader', `rep.io.sockets', `rep.util.time', `rep.net.domain-name' * Bug fixes, including: - Find size of `long long' type on AIX, IRIX and Solaris (Dan McNichol, Albert Chin-A-Young) - Never allow macros to be called as functions - Make bitfields unsigned (Albert Chin-A-Young) - Fixed bounds-checking when parsing non-base-10 fixnums - Thread fixes (and much lower thread-switch latency in many cases) - Fixed `DEFUN' macro for C++ (Matt Tucker); also fixed header files to avoid C++ keywords - Make error message for bytecode version mismatches more meaningful - Fixed: `default-boundp', `continuation-callable-p' - Only the evaluate the value of `defvar' forms if the symbol isn't already bound - Compile else-less `case' expressions correctly; eliminate tail-recursion in `cond' forms when possible - Various fixes in `scheme' module 0.12.4 ====== * Support building without GNU MP, `--without-gmp' option to configure. Use `long long' for non-fixnum integers (promote to floats when out of range); no exact rationals. There's also an option to disable continuations/threading (`--disable-continuations') * Sanitized function inlining: - Use `(declare (inline NAMES...))' to tell the compiler that it might be useful to inline the named functions - Don't even think about inlining across module/file boundaries (for now anyway) * Cleaned up the `gaol' module. Interface is essentially: `gaol-define', `gaol-define-special', `gaol-define-file-handler'. Added `gaol-open' to import complete modules. Still supports old interface * Be a lot more efficient when printing quoted strings and symbol names (for some streams there used to be a system-call per character!) Also, when quoting weird symbol names, be more intelligent * Removed code to autoload from modules (which didn't really work anyway) * Be more intelligent about deciding when to flush the module cache * Build fixes for IRIX (David Kaelbling) * Other miscellaneous bug-fixes and performance tweaks 0.12.3 ====== * New function `thread-join', waits for a specified thread to exit, then returns the value of the last form it evaluated * Added a rudimentary profiler (`,profile FORM' command in repl) * Reorganized `ring' module, sanitized the interface (preserving compatibility with old functions), also added a `ring->list' function * `rplaca' and `rplacd' (but not `setcar' and `setcdr') functions now return the cell being modified, not the value being stored into it, for compatibility with CL (Karl Hegbloom) * `unwind-protect', `catch', `condition-case': these special forms are now macros * When signalling `bad-arg' or `missing-arg' errors, try to include the function as the first element of the error data * `load' function now _only_ loads files without suffixes if NO-SUFFIX arg is non-`nil' (prevents picking up un-suffixed files by mistake, e.g. from the current directory) * Fixed some bugs when reading rationals * Fixed bug of `gettext' module not redefining `_' binding in `rep' module * Fixed bug when building `rep-config' script (Mark Hewitt, Dan Winship) * Fixed bug of `rep_INTERN_SPECIAL' macro not looking for default values of special variables * Fixed interpreted versions of `min' and `max' when operating on non-numeric values * If unable to allocate heap space, just print an error and terminate the program (the low-memory handling doesn't currently work properly) * Fixed bug when extracting doc strings from `define' forms * Fixed bug when compiling structure definitions in non-top-level environments * Fixed bug of being unable to `load' empty files * When recursively macro-expanding, dereference identifiers in the correct module 0.12.2 ====== * The tar file-handler now caches the unpacked archive (wins big when loading sawfish themes) * The `gaol' module can now create multiple gaols, each with it's own namespace * More performance tweaks * Miscellaneous bug-fixes (more vm stack smashing, `defconst' never evaluates its constant) 0.12.1 ====== * Some virtual machine performance tweaks * Fixed nasty stack smashing bug (when using compiler declarations) * Some 64-bit cleanups (George Lebl) * Fixed non-ANSI C syntax (Sam Falkner) 0.12 ==== * Added a basic module system. Modelled after the Scheme48 system, but simpler. At its simplest, include a `define-structure' form in each file representing a module: (define-structure NAME INTERFACE CONFIG BODY...) The external definitions of this module can then be imported by other modules through their CONFIG statements, e.g. `(open NAMES...)'. Most modules will open `rep' to get the standard language definitions. `foo#bar' reads as `(structure-ref foo bar)' The `timers', `tables', `sdbm', `gdbm', `readline', `gettext', `ring', `mutex', `memoize', `lisp-doc', `disassembler', `compiler', `date', `cgi-get', `gaol' features are all now modules (this is backwards compatible, since modules may be imported using `require') See the "Modules" section of the manual for more details. * The repl now contains meta-commands for inspecting and configuring the module system (amongst other things) * Added a facility for creating new primitive types: `make-datum', `datum-ref', `datum-set', `has-type-p', `define-datum-printer' * Added an SRFI 9 compatible `define-record-type' macro for defining data structures (the `records' module) * Added fluid variables--a method of creating dynamically scoped bindings that fit well with lexically scoped definitions (`make-fluid', `fluid', `fluid-set', `with-fluids', `let-fluids') * Added a `queues' module providing a basic queue type * Added stream functions: `peek-char', `input-stream-p', `output-stream-p' * Interpreter now also eliminates tail-calls * Changed handling of inexact numbers to be compatible with the Scheme standard: - Many operations now produce inexact outputs if any of their inputs are inexact (e.g. `min', `max', `floor', `ceiling', `round', `truncate') - `eql' and `equal' no longer ignore exactness when comparing numbers. `=', `/=', `<', `>', `<=' and `>=' _do_ ignore inexactness. E.g. (eql 2 2.) => nil (= 2 2.) => t * Support most of Scheme's read-syntax for numbers (i.e. `#b', `#o', `#d', `#x' radix prefixes, and `#e', `#i' exactness prefixes). * Implemented Scheme's `string->number' and `number->string' functions * Included a basic R4RS Scheme implementation (module: `scheme'). Do `,new foo ,open scheme' to test it in the repl, use `(open scheme)' instead of `(open rep)' to use it within modules. The compiler also knows enough about Scheme to be able to compile it. Also, use the `-s' or `--scheme' options to load a file of Scheme code. * The debugger works better (and can be used to walk the stack history somewhat) * Last arg of `append' and `nconc' may be a non-proper-list now * Implemented the Scheme `do' macro for iteration * `define' supports curried functions. E.g. `(define ((plus a) b) (+ a b))', then `(plus 1)' evaluates to the function that adds one to its argument. * Many performance improvements: - Allocates less memory (so garbage collects less often) - Much faster at bytecode-to-bytecode function calling - Much reduced VM overhead (when compiled with GCC) * Compiler improvements: - Supports the `(declare CLAUSES...)' form. See the "Compiler Declarations" section of the manual for details on the actual declarations supported. - Is cleverer about detecting when to create new bindings when tail recursing, and when the old bindings can just be overwritten - Groks the module system, and the language of the module being compiled (so that it can compile both rep and Scheme code) - Splices bodies of top-level `progn' and `begin' forms themselves into the top-level (for when macros expand into multiple definitions) - Compiling already defined functions (or whole modules of functions) now (mostly) works - Coalesce and compile non-defining top-level forms * Many bug fixes (see ChangeLog files for details) 0.11.3 ====== * Fixed bug of throwing uninitialized errors when autoloading * Fixed bug of interpreting `(let () ...)' as a named let 0.11.2 ====== * Replaced many special forms by macros--`let', `let*', `function', `if', `and', `or', `prog2', `defmacro', `defun', `defconst', `define-value', `setq-default' * `let' now supports Scheme's named-let construct for iteration via tail recursion * Parse some standard Common Lisp and Scheme syntax: `#| ... |#' block comments, `#\C' or `#\NAME' characters (where NAME may be one of: `space', `newline', `backspace', `tab', `linefeed', `return', `page', `rubout'), and `#(...)' vectors * When comparing symbols, compare their names as strings * Implemented Scheme's `dynamic-wind' function * Fixed bug of sometimes evaluating function arguments in the environment of the callee not the caller * Fixed bug when calculating how long to sleep for when no threads are available * Fixed bugs in mutex implementation (Damon Anderson) * Work around bugs in Tru64 `RTLD_GLOBAL'; everything should work on Tru64 now (Aron Griffis) * Fixed bug of not saving current regexp state across continuations 0.11.1 ====== * The compiler now eliminates single-function tail calls (instead of leaving it to the virtual machine) * Updated to use libtool-1.3.4 * Miscellaneous bug fixes and minor changes 0.11 ==== * Better support for numerical computing. Now supports bignums, rational numbers (numerator and denominator are bignums), and floating point values as well as the original fixnums. Many new numerical functions supporting these types. Promotes and demotes hopefully as you'd expect (never demotes an inexact number to an exact number). Tries to follow the Scheme numeric system as much as possible * Supports "guardian" objects through the `make-guardian' function (as described in Dybvig's paper). These are a clean mechanism for allowing the programmer to control when arbitrary lisp objects are finally deallocated. Also added a new hook: `after-gc-hook' * The default error handler can now be redefined. If the variable `error-handler-function' contains a function then it will be called to handle the error, with arguments `(ERROR DATA)'. * New special form `case', switches on a key value and sets of constants * New function `call/cc' (also available through the alias `call-with-current-continuation'). Provides scheme-like continuation functions. Special variables are now deep-bound to support this correctly * Supports "soft" preemptive threads using continuations and a general "barrier" mechanism (used either for restricting control flow, or for receiving notification when control passes across a barrier) * Parameter lists in lambda expressions now support improper lists, as in scheme. E.g. `(lambda (x . y) ...)' * Implements the scheme `define' syntax, with support for inner definitions * The `tables' plugin implements hash tables, with extensible hashing and comparison methods; supports both strongly and weakly keyed tables * Included a GDBM binding; DOC files are now stored in GDBM files (SDBM has limits on datum sizes) * `put' and `get' functions now use `equal' to compare property names * Virtual machine / compiler improvements: - Variable references and mutations are classified by type: lexical bindings use (one-dimensional) lexically addressed instructions, global non-special bindings have their own instructions, everything else uses the original instructions. Similar classification when creating new bindings - Eliminate tail-recursive function calls wherever possible in compiled code (when the calling function has no dynamic state) Compiled lisp code will need to be rebuilt to run on the modified virtual machine. * When expanding macros, bind `macro-environment' to the macro environment it was called with. This allows macros to reliably expand inner macro uses * New hook `before-exit-hook'. Called immediately before exiting * `rep-xgettext' now has an option `--c'. This makes it output pseudo C code containing the string constants found * Fixed misfeature of interpreting filenames `FOO//BAR' as `/BAR'. Contiguous path separators are now merged (i.e. `FOO/BAR') 0.10 ==== * Updated support for dumping (freezing) lisp definitions to handle lisp-1 nature with closures. Also now generates C code instead of assembler for portability; creates a plugin that may be loaded through the REP_DUMP_FILE environment variable * Plugin `.la' files may now contain rep-specific settings: `rep_open_globally=yes' and `rep_requires='FEATURES...'' * New function `define-value'. A combination of `set' and `defvar', but without implying dynamic scope * `load' scans AFTER-LOAD-ALIST for plugins as well as lisp libraries * `(if t)' now evaluates to `nil' not `t' * Fix regexp bug in matching simple non-greedy operators (Matt Krai) * Borrowed guile's bouncing parentheses for readline (Ceri Storey) * New C functions `rep_load_environment' and `rep_top_level_exit' * `defvar' allows symbols to be redefined in protected environments if they haven't also been defined by unprotected environments * Detect GCC's with broken `__builtin_return_address' functions (George Lebl) * Try to use libc `gettext' implementation, but only if it looks like it's the GNU implementation 0.9 === * Support for using GNU readline (give configure the `--with-readline' option) * New functions: `letrec', `caar', ..., `cddr', `caaar', ..., `cdddr', `in-hook-p', `make-variable-special' * Changed `unless' to have the Common Lisp semantics--return `nil' when the condition evaluates true, not the value of the condition * Fixed/added some compiler optimisations * Fixed `rep-xgettext' script to remove duplicated strings and to search exhaustively * `add-hook' forces the hook variable to be special (in case it wasn't declared using `defvar') 0.8.1 ===== Fixed some documentation bugs; fixed some build problems 0.8 === * Default scoping is now lexical, only variables declared using `defvar' are dynamically scoped. * There is now only a single namespace for symbols (excepting property lists), this means that the `fset', `symbol-function' and `fboundp' functions have been removed This allows all elements in procedure-call forms to be evaluated equally (as in scheme), so things like: ((if t + -) 1 2) now work. Related to this, function names (i.e. symbols and lambda expressions) are no longer dereferenced by any operations taking functions as arguments. Only built-in subroutines and closures are considered functions. This means that where before you'd write something like: (mapcar '+ '(1 2 3)) this is now illegal; the `+' function must be evaluated: (mapcar + '(1 2 3)) * `lambda' is now a special form evaluating to a closure (as in scheme); this means that the following are exactly equivalent: (lambda (x) x) == (function (lambda (x) x)) == #'(lambda (x) x) An alternative method of enclosing a lambda expression is to use the `make-closure' function. * `gaol' module providing semi-safe environment for untrusted code to evaluate in * Support for i18n through `gettext' module; also support for `%1$s' type format specifiers * New functions `string-equal' and `string-lessp' 0.7.1 ===== * Added `--with-rep-prefix' option to autoconf AM_PATH_REP macro * Fixed bug when inserting a new timer before an existing timer * Fix the malloc tracking code * Fix dlmalloc for FreeBSD * Use install when installing, not cp * Some fixes for compiling with SUN's C compiler on Solaris 0.7 === * Added file handler for read-only access to the contents of tar archives, access files like `foo.tar.gz#tar/bar' * `process-id' function now returns pid of lisp interpreter when called with zero arguments * Added (untested) support for loading dynamic objects via `shl_load' (HP-UX) * Added (untested) support for systems that prefix symbol names in dynamic objects with underscores * Fix bug when compiling `last' function * Fix bug of not closing files in the `load' function 0.6.2 ===== * Added `autoload-verbose' variable; set it to `nil' to turn off the messages when autoloading * Fix problems when `--prefix' option has a trailing slash * Updated libtool files to version 1.3.3 * Initial (incomplete) support for building under Tru64, from Aron Griffis 0.6.1 ===== No new features; minor portability tweaks and build changes. Fix bug of trying to load directories as Lisp scripts 0.6 === * Add `unsetenv' function * `system' now uses `process-environment' * Workaround compiler bug with GCC 2.95 on sparc * Fix build problem where libsdbm.la can't be located 0.5 === * New function `set-input-handler', registers an asynchronous input handler for a local file * Don't abort on receipt of unexpected `SIGCHLD' signals * Upgrade libtool to version 1.2f * The `rep' binary by default always loads a script named `rep', not named by it's `argv[0]' (this breaks under the newer libtool) 0.4 === * Sending a rep process a `SIGUSR2' prints all debug buffers * Added `--with-value-type', and `--with-malloc-alignment' configure options. Also added code to automatically detect the first of these options. * Fixed some 64-bit problems * Removed the difference between static and dynamic strings 0.3 === * New compiler command line option `--write-docs' 0.2 === * The variables `error-mode' and `interrupt-mode' control where errors and user-interrupts (i.e. `SIGINT' signals) are handled. The three possible values are: `top-level', `exit' and `nil' (denotes the current event loop). * Fixed bug where all dynamic types were erroneously `symbolp'. * `SIGINT', `SIGHUP' and `SIGTERM' signals should now be caught more successfully. * Added a new directory to `dl-load-path': `LIBEXECDIR/rep/ARCH' to contain third-party shared libraries. 0.1 === First public release.