** Pure 0.57 2013-02-14
This release sports some improvements in the build system, fixes for changed
linker defaults on recent Linux systems, compatibility with the latest Faust2
revisions, as well as some changes for compatibility with C++0x/11. Please
check the ChangeLog for more details.
The big news is that Pure has moved to Bitbucket. GoogleCode has served us
well over the past few years, but it's time to move on and take advantage of
the improved collaboration features offered by modern source code hosting
services. Here are the new URLs so that you can update your bookmarks:
Project website: http://purelang.bitbucket.org/
Mercurial repository: https://bitbucket.org/purelang/pure-lang
Bug reports: https://bitbucket.org/purelang/pure-lang/issues
We kindly ask you to use the issue tracker on Bitbucket for future bug
reports. Old releases will continue to be available at GoogleCode, and the
mailing list remains at http://groups.google.com/group/pure-lang.
** Pure 0.56 2012-11-19
User-visible changes in this release:
* The first thing you'll notice is the spiffy new logo for the sign-on message
which was created with figlet. (If you prefer the old sign-on message then
you can disable this with the --plain option or by setting the PURE_PLAIN
environment variable.)
* Comprehensions involving patterns now filter out unmatched generator values
automatically, as it is in Haskell. The previous behaviour of raising an
exception in such cases offered no real benefits and was in fact very
inconvenient in most situations.
* The constants 'true' and 'false' are now declared as nonfix symbols in the
prelude, so that they can be used in patterns. Also, a 'bool' type was added
along with a corresponding 'boolp' predicate for the normalized truth
values, as well as a 'bool' function to convert machine integers to
normalized truth values.
* There's a new (and experimental) enum.pure module which provides some
convenience functions to facilitate the creation of enumerated types.
* Other new stuff in the language and the library includes the namespace
brackets (based on an idea by Yann Orlarey), the non-splicing vector
brackets (contributed by Jim Pryor), as well as a '__gensym__' macro for
generating fresh variable symbols on the fly. Also, there are two new
builtins '__break__' and '__trace__' to trigger debugging programmatically
from a Pure script. Details can be found in the documentation.
* Due to popular demand there's a new escaped command syntax which can be
enabled with the --escape option or the PURE_ESCAPE environment variable.
Details can be found in the Interactive Usage section of the manual. (Thanks
to Jim Pryor for his help with this one.)
* Thanks to a request by Kurt Pagani, fairly complete TeXmacs support is now
available via the new --texmacs option of the interpreter. Installation
instructions can be found in the "TeXmacs Mode" section of the INSTALL file,
and you should also take a look at the TeXmacs configuration files and
sample documents in the texmacs/plugin/pure directory.
* Readline support has been improved. Specifically, interpreter and debugger
now have separate command histories, and the latest incarnation of the
pure-readline module uses its own command history as well. (pure-readline
now also provides a more complete implementation of the readline interface
for Pure programmers.)
* Emacs Pure mode has been improved. There's a new pure-send-yank command
which allows text from the kill ring to be sent directly as command input to
the interpreter. Also, command prompts and result lines can now
automatically be stripped from the command input, which is convenient to
send sample transcripts as can be found, e.g., in the online documentation.
* Also, there's a new a2ps style sheet (etc/pure.ssh) contributed by Eddie
Rucker and a couple of new examples, including a merge sort example
(msort.pure) by Jim Pryor, an imperative-style LU decomposition example
(lu.pure) by Johannes Engels, a matrix inversion example ported by Jiri
Spitz (matrix_inversion.pure, original Q script by Rob Hubbard), and the
unit conversion example (units.pure) which was discussed on the mailing
list.
In addition, this release also sports compatibility with the forthcoming LLVM
3.2 release and quite a few bugfixes, please see the ChangeLog for details.
Thanks to all who contributed patches and reported bugs!
** Pure 0.55 2012-06-04
A maintenance release featuring some improvements in the build system, Debian
packaging and a bugfix in the compiler. Please see the ChangeLog for details.
The most noticeable user-visible change is that we now use clang and gfortran
(with dragonegg) as the default compilers for inlined C/C++ and Fortran code,
because llvm-gcc isn't supported in the LLVM 3.x series any more.
** Pure 0.54 2012-05-19
This release sports LLVM 3.1 compatibility and some minor bugfixes in the
interpreter and the library, please check the ChangeLog for details.
** Pure 0.53 2012-03-22
Another maintenance release. This one fixes a bug in the debugger and adds
some convenient options to the trace command. Moreover, the sort standard
library function was optimized so that it now runs much faster with standard
comparison predicates and basic POD types (int, bigint, double, string).
** Pure 0.52 2012-03-08
This is another bugfix and maintenance release.
The most noticable change is that the autoquoting of macro arguments is now
disabled by default. Instead, there's a new --quoteargs pragma which lets you
turn this on for macros which need it. Also, autoquoting now really defers
macro substitution in the quoted argument. This fixes the most annoying cases
of accidental name capture in macro calls embedded in quoted block constructs.
Also, the Faust support in the runtime was updated after Grame's recent
cleanup of the Faust UI data structures, so that Pure's built-in Faust
interface is ready to work with the current git revision of the Faust2 branch
(http://www.grame.fr/~letz/faust_llvm.html).
** Pure 0.51 2011-12-21
This is mostly a bugfix release.
- Fixed a bug in macro substitution, eliminating fallout from changes
concerning optimized 'when' clauses introduced in Pure 0.50.
- Fixed a bug in the Faust bitcode loader which would cause segfaults in case
of an existing but invalid Faust bitcode file.
- Added a runtime function which allows client modules such as pure-faust to
load a Faust dsp in bitcode format. This doesn't rely on eval and so also
works in batch-compiled scripts.
** Pure 0.50 2011-12-18
Most user-visible changes are in the type rules (please see the documentation
for details):
- Recursive types. Simple recursive type predicates can now do tail call
elimination during matching, so that the "naive" definition of, e.g., the
rlist type finally works in constant stack space.
- Interface types. This is a first stab at implementing a Go-like interface
type system in Pure. It has some limitations due to Pure's dynamically typed
nature, but seems to work fairly well already. Please check the new
"Interface Types" section in the manual for details.
- Type aliases are now resolved at compile time as far as possible, so that
the best possible matching code can be generated.
- Non-binding 'when' clauses of the form 'when x end' where x is just an
expression are now completely eliminated, yielding faster and much more
compact code which is equivalent to that of the '$$' operator.
- New --rewarn pragma which resets the warning option to its default. New
--ifdef / --ifndef pragmas to check whether a source option is defined or
not. Also, compilation options are now explicitly set to a default value by
means of a PURE_OPTION environment variable; please check the documentation
for details.
- Some cosmetic changes in Pure Emacs mode concerning the treatment of
shebangs (which are now considered as preprocessor rather than comment
lines) and multiline comments in cursor movement and send-line/send-defun
commands. All editor modes: add support for the new 'interface' keyword.
- Several bugfixes, please check the ChangeLog for details. Thanks are due to
Peter Summerland and Stephan Rudlof for reporting some particularly annoying
bugs in the hash() and force() functions. Also, the Faust bitcode loader got
a complete overhaul, so that it works with the batch compiler now and
properly cleans up old code when a Faust module gets reloaded in interactive
mode. In addition, the Faust interface now finally implements full support
for dsp and controller metadata.
** Pure 0.49 2011-11-16
This release has been tested and is known to work with the latest LLVM 3.0
release candidates (and continues to work with older LLVM releases >= 2.5).
It sports the following changes:
- New built-in __list__ macro. This expands a tuple to a list literal,
preserving embedded tuples in the same way that list values are parsed.
(See the "Inspection" section in the Pure Library Manual for details.)
- New conditional compilation pragmas --if, --ifnot, --else, --endif,
--enable, --disable; the latter two are also available as command line
options. (See the "Conditional Compilation" section in the Pure Manual for
details.)
- The prelude now has some experimental optimization rules for vectors of int
and double ranges as well as list and matrix slices of contiguous ranges,
which make these operations really fly in the most common cases, since the
construction of intermediate list values is avoided. (See the "Slicing" and
"Matrix Construction and Conversions" sections in the Pure Library Manual
for details.)
- Some bugfixes and improvements in the comparison operations of the
dictionary and set data types.
- The runtime now has support for interpreter-local storage and a pointer
equality callback, so that external modules can hook into same() and the
null check for tagged pointer types.
As usual, all the gory details can be found in the ChangeLog.
** Pure 0.48 2011-10-18
This release has been tested and is known to work with LLVM 3.0 (and continues
to work with older LLVM releases >= 2.5).
Please note that beginning with this release, in addition to GMP you also need
to have the GNU MPFR library (http://www.mpfr.org/) installed to compile this
package. Also, there have been some changes in the internal runtime API which
required a bump of the library version number, so you'll have to recompile the
addon modules that you need.
Here's a quick summary of the other user-visible changes. As usual, please
check the ChangeLog for details.
- LLVM 3.0 compatibility. This has been tested with the LLVM 3.0 release
branch (at the time of this writing, LLVM 3.0 hasn't been released yet).
- Experimental DragonEgg support. This has been tested with the DragonEgg 3.0
release branch. Specifically, the C/C++/Fortran inlining facility now has
some special support for DragonEgg so that using gcc, g++ and gfortran with
the -fplugin=dragonegg option as your PURE_CC/PURE_CXX/PURE_FC compiler
should work automagically.
- New --defined/--nodefined and --warn/--nowarn pragmas, please check the
documentation for details.
- New __func__ and __namespace__ builtins which expand to the current function
and namespace at the call site. Also, __locals__ has been changed to a
built-in macro (rather than substituting it away at code generation time) so
that proper lambda lifting is performed on the local function references.
- Library: Regex functions and pointer arithmetic have been moved into
separate regex and pointers modules. This means that you'll have to add an
explicit import of these modules to use these facilities. Also, there's a
new lasterrpos function which gives more detailed error information about a
failed val/eval/evalcmd operation, and the system module offers a new and
improved printf/scanf implementation which supports GMP and MPFR (%Zd, %Rg)
conversions.
- Emacs Pure mode: Changed the keybindings for the block traversal commands to
use the Meta prefix. (This used to be Ctrl+Shift, but this interferes with
the standard binding for the shifted word movement commands for selecting
ranges of text in cua mode.) Also, pure-mark-defun (C-M-h) and
pure-send-defun (C-c C-f) now allow you to mark or send an inline code
section (%< ... %>), if point is at or inside such a section.
** Pure 0.47 2011-03-29
This release has been tested and is known to work with LLVM 2.9 (and continues
to work with older LLVM releases >= 2.5).
* Build System
pkg-config support was added to the sources, which simplifies compiling Pure
modules and linking them against the Pure runtime library (see
examples/hellomod for an example).
* Language
Support for extern declarations of varargs functions was added. This makes it
possible to call functions like printf directly from Pure. Please check the
manual (section "Variadic C Functions") for details.
Inline code sections are now denoted '%<...%>' instead of the previous
'%{...%}' syntax. This makes it possible to properly support them in Pure mode
(see below), and also avoids syntactic ambiguities (note that '%{' is legal
Pure syntax; it could denote a % operator followed by a matrix value).
Following an extensive discussion on the mailing list, the support for
user-defined type tags was completely revamped and support for matrix patterns
was added. Please check the corresponding manual sections ("Rule Syntax" in
the Pure Manual and "Prelude Types" in the Pure Library Manual) for details.
The reflection capabilities have been improved as well. In particular, the new
get_fundef/get_typedef/get_macdef and add_fundef/add_typedef/add_macdef
functions allow direct inspection and manipulation of the rewriting rules of
the running program. This uses a new built-in quoted representation of
rewriting rules and special expressions (conditionals, lambda, case/when/with)
which can be manipulated freely in functions and macros and then be evaluated
when needed. Details can be found in the "Built-in Macros and Special
Expressions" and "Reflection" sections of the Pure Manual.
* Interpreter
The user can now define interactive commands by placing suitable function
definitions in the special __cmd__ namespace. Please check the "Interactive
Usage" section in the manual for details.
* Pure Mode
Emacs Pure mode went through a comprehensive overhaul, offering a bunch of
bugfixes, improvements and new features, see the ChangeLog for details. The
most important user-visible changes are listed below:
- At long last, Pure mode now offers indentation of Pure code (TAB) and
filling of comment paragraphs (M-q). Also, "electric" characters (';', '=',
parentheses etc.) have been added which do automatic indentation and
alignment of right-hand sides of equations.
- As suggested by Jason E. Aten, Pure mode now provides commands for feeding
Pure code into a running interpreter instance.
- Some new cursor movements commands (pure-prev/next/backward/forward-defun)
are available for traversing the block hierarchy of a Pure program. There's
also a new pure-mark-defun command to mark Pure definitions.
- There's a new convenience command pure-scratchpad to create a *pure-scratch*
buffer and put it in Pure mode, and a pure-quit command to quit the Pure
interpreter and kill its buffer.
- Debugging mode is now controlled with the pure-debug-mode flag, a toggle for
this can be found in the Pure menu. Also, a new customization variable,
pure-stacksize, allows you to set the value of the PURE_STACK environment
variable in Emacs.
- The pure-help command now properly looks up symbols in the global help
index, and completion of Pure symbols was added to Pure mode and a number of
commands which, like pure-help, take a symbol as input in the minibuffer.
Also, Pure mode gets byte-compiled and installed now by default, if Emacs can
be detected at configure time.
** Pure 0.46 2010-12-05
The biggest change in this release is that the Pure documentation is now
formatted using Sphinx (http://sphinx.pocoo.org/). This implies some changes
in the way the docs are organized, generated and accessed. The benefits are:
- More complete and prettier docs in a variety of formats (including html,
htmlhelp, pdf and epub). You can take a look at the docs here:
http://docs.pure-lang.googlecode.com/hg/index.html
- Global index and cross-referencing between different manuals. This
encompasses all available manuals, including the addon documentation.
Searching the global index is also supported by the interpreter's 'help'
command.
- Specific RST markup for Pure-related items such as functions, macros, etc.,
syntax highlighting of Pure code (using Pygments), inline latex math and all
the other goodies that Sphinx offers.
To make this possible, the documentation is now in a separate package which
must be installed in addition to the interpreter. (If you have a working
internet connection and the wget program, just run 'sudo make install-docs'
while installing the interpreter. Sphinx is not required to do this. Please
see the INSTALL file for details.)
Other than that, there are two new configure options to make it possible to
force static linking against LLVM, and to use MPIR instead of GMP for the
bigint arithmetic. Please check the ChangeLog for details.
Also note that the Pure source code is maintained in a Mercurial repository
now. You can check out the latest development sources with:
hg clone https://pure-lang.googlecode.com/hg pure-lang
** Pure 0.45 2010-10-31
This release is mostly concerned with the C interface in general and the
bitcode interface in particular, featuring many improvements and bugfixes in
this area. The most important user-visible changes are listed below.
* Improvements and bugfixes in the bitcode interface. In particular, bitcode
modules can now be imported into different namespaces and can be made
'public' or 'private' on a per-namespace basis. Also, there's now more
leeway in redeclaring external functions in different ways, which gives a
way to work around the char* versus void* issue in the bitcode interface.
Last but not least, Faust bitcode modules can now be reloaded on the fly,
dsp pointers are now fully type-checked, and some new convenience functions
as well as a little bit of runtime type information for Faust modules has
been added.
* Inlining of C, C++, Fortran and Faust code in Pure scripts. The interpreter
now compiles these to bitcode and loads the bitcode modules on the fly,
provided that you have the necessary compilers installed (llvm-gcc for
C/C++/Fortran, clang for C/C++, faust2 for Faust). See the new "Inline Code"
section in the manual for details.
* New int** and double** conversions for matrices which generate a vector of
row pointers on the fly. (This was already used in the Faust interface
previously, but is now available for all applications.) For consistency, the
related int* and double* conversions now also treat the matrices as
reference arguments, so that they can be modified in-place. Moreover, the
missing byte and long long vector conversions (int matrix to char* or
char**, as well as to int64* or int64**) were added, and there's now also
support for argv-style vector arguments (char**, void**). This is all
described in detail in the "C Interface" section of the manual which was
completely rewritten.
* Full checking of pointer types in the C interface. To these ends, pointer
values can now be tagged with type information at runtime, and the compiler
keeps track of all C pointer types used in extern declarations. void* can
still be used to pass any kind of pointer, but otherwise the pointer types
will have to match when calling an external function (or you'll have to use
a type cast; operations for that as well as runtime type information are
provided in the prelude).
There's no doubt that this change will raise lots of backward compatibility
issues. But with all the C library modules out there, checking pointers for
C type compatibility was becoming a real issue, and the workarounds were
ugly. A real solution was needed, so better we do this now than after 1.0 is
out and a lot more library modules have been written.
* Sentries can now be placed on any kind of Pure expression. This is
especially convenient with atomic non-pointer data such as file descriptors
since these can be finalized now. (Alas, this also needs 4 bytes more per
expression node on 32 bit systems. Storage size on 64 bit systems is still
the same, though.)
Besides these changes, the interpreter also sports the following new features:
* The new --eager pragma instructs the interpreter to compile a given function
eagerly. This is particularly useful to avoid hiccups in realtime functions.
* The interpreter can now do a "cold" restart (reloading all scripts) using
'run' without a script argument. You can also employ this to quickly rerun
the interpreter with debugging enabled using 'run -g'.
** Pure 0.44 2010-10-01
This release sports a bunch of bugfixes and OSX and FreeBSD compatibility
fixes. The build system was also improved, and the sources are now compatible
with the forthcoming LLVM 2.8 release. As usual, please check the ChangeLog
and the manual for details. The most important user-visible changes are listed
below.
- Experimental support for importing LLVM bitcode files ('using "bc:..."').
This also includes special support for importing Faust bitcode modules
('using "dsp:..."'). See the manual section "LLVM Bitcode Interface" for
details. Some examples can be found in examples/bitcode.
- Constants holding runtime data such as pointers are now properly handled by
the batch compiler.
- New --const, --noconst code generation options. These allow you to tell the
batch compiler whether constants should be computed at compile time or at
run time.
- New --required pragma. This allows you to tell the batch compiler about
functions which should always be linked into the output code.
- Cosmetic changes in the expression printer. Combinations of non-associative
operators on the same precedence level are now properly parenthesized. So,
e.g., '(x=>y)=>z' unparses as is now, rather than '(=>) (x=>y) z'.
- Fixed up treatment of floating point constants in the lexer so that typos
are now treated in a manner consistent with integer constants.
- New 'trace' debugging command. This enables you to trace reductions by
certain functions without actually interrupting the evaluation.
- Allow printed expression results to be piped through the program named with
the PURE_LESS environment variable (similar to what 'show' does with
PURE_MORE).
- As discussed on the mailing list, -w now also warns about implicit
declarations of symbols in the default namespace.
Standard library:
- New time-related functions, 'key' and 'val', vector functions.
- Overhaul of the precedence table so that $$ and $ are now on different
precedence levels.
- Prevent segfaults due to 'fclose' et al being run twice on the same file
pointer.
- 'tuple', 'string' and 'matrix' now act as the identity if the argument is
already of the corresponding type.
Thanks to Michael Ashton, asitdepends, John Cowan, Masahiro Miura, Eddie
Rucker and Peter Summerland for their bug reports and suggestions. Also,
special thanks are due to Roman Neuhauser for his work on improving the build
system, to Stephane Letz for the joint work on the Pure-Faust bitcode
interface, and to Yann Orlarey for his hospitality and fruitful discussions
while visiting Grame.
** Pure 0.43 2010-03-17
More polishing, bugfixes and general improvements on the road to Pure 1.0.
- Integer literals can now be specified in base 2 (0b or 0B prefix), as
suggested by Geert Janssen.
- The debugger now provides a basic post mortem debugging facility. To make
this work, you have to invoke the interpreter in debugging mode (-g). The
new 'bt' command then prints a backtrace of the call sequence after an
evaluation died with an unhandled exception. As with 'show', the output of
'bt' can be piped through PURE_MORE or captured in a string with 'evalcmd'.
- The interpreter can now be run with the --ctags or --etags option to create
a tags file. This file can be used with vi, emacs and other editors which
support this feature, in order to quickly locate global symbol declarations
and definitions in Pure scripts. Emacs Pure mode now also supports this
with a new 'Make Tags' command in the Pure menu.
- The 'using' clause now features a new 'sys:' tag (e.g., 'using "sys:math";')
to bypass the search of the current script directory, which works analogous
to the C preprocessor's '#include <...>' construct.
- Better programming-in-the-large support. There's a new "scoped" namespace
construct ('namespace xyz with ... end') and the rules for creating new
symbols in namespaces have been revised. Basically, new global symbols are
now always created in the current rather than the default namespace, which
is what most people expect. (You can use the new -w option to check existing
scripts for backward compatibility issues, where symbols are placed into a
different namespace following the new namespace rules.) Also, there's now a
restricted form of the 'using namespace' clause which gives full control
over which symbols from a namespace should be visible as unqualified
symbols. Please see the section on namespaces in the manual for details.
- Complete overhaul of the dictionary and set data types. We now also support
multidicts as well as hashed sets and bags. Moreover, set-like operations
(union, difference, intersection and subdict comparisons) are now available
for dictionaries, and different kinds of dictionaries or sets/bags can now
be mixed in infix operations (+, -, *, <, <= etc.). See the "Dictionaries"
and "Sets and Bags" sections in the Pure library manual for details.
- An example showing how to interface to GNU Fortran has been added. See
fortran.{pure,f90} in the examples directory.
- LLVM 2.7 compatibility fixes. Thanks to Jeffrey Yasskin, the TCO-related JIT
breakage on 64 bit systems present in recent LLVM svn revisions seems to be
fixed now, so that we expect Pure >= 0.43 to work with the forthcoming LLVM
2.7 release.
Many thanks to all who reported bugs and participated in the mailing list
discussions leading up to this release!
** Pure 0.42 2010-01-29
This release adds some facilities to inspect memory usage in the interpreter,
and fixes some performance bugs in the library.
* The new 'mem' command prints current memory usage (in terms of expression
cells) in the interpreter, and the 'stats' command has a new '-m' option
which enables printing the expression memory used in evaluations. Example:
> mem
413 cells (0 free)
> stats -m
> let xs = scanl (+) 1 (1..100);
0s, 306 cells
> mem
716 cells (306 free)
* Some performance issues with 'split' and the regex operations (reported by
Jiri Spitz) were fixed. Also, the regexg, regexgg and regsplit operation now
have lazy alternatives which return a stream instead of an eager list, which
makes processing of big strings more efficient. Examples can be found in the
library manual.
** Pure 0.41 2010-01-21
This release fixes some of the remaining warts of the language and the
implementation, specifically:
* Indirect tail calls are now fully optimized, so that any remaining
limitations concerning the use of tail recursion in indirect function calls
and mutually recursive globals have been removed. (This required some slight
adjustments in the runtime ABI, so the runtime library version was bumped as
well.)
* Due to popular demand, the semantics of the short-circuit logical operators
(&& and ||) were changed so that these builtins are both extensible and
tail-recursive now. The new implementation behaves as if && and || were
defined by the following equations (except that the second operand y is only
evaluated if necessary):
x::int && y = if x then x else y;
x::int || y = if x then y else x;
* A new syntax for declaring infix/prefix/postfix operators with a precedence
given by an existing operator has been added. So you can write declarations
like the following,
infixl (+) ++;
which declares a new infixl operator ++ at the same precedence level as +.
* The runtime and the prelude now offer some special support for symbolic
vectors of the form {key=>value,...}, which provide a convenient
representation for record-like data structures.
* Users of the batch compiler (pure -c) should note that stripping of unused
functions (-s) is now the default in batch compilation. To get a full,
unstripped executable, you have to use the -u option. (The -s option is
still supported for backward compatibility, but doesn't have any effect,
except that it overrides a previous -u option.)
Detailed information about these changes can be found in the Pure manual.
Also, some bugs were squashed, a few more functions were added to the library,
and the Mersenne Twister ('random' function) was updated to the 2002 version
which fixes some issues with certain kinds of seed values.
Last but not least, Hoigaard/autotelicum contributed syntax highlighting for
BBEdit and TextWrangler (OSX editors), this is now included in the etc
subdirectory (please see pure.plist).
As usual, all the gory details can be found in the ChangeLog.
** Pure 0.40 2009-12-23
This release implements proper multi-argument lambdas and support for
non-linear rules such as 'a*b+a*c = a*(b+c)', as discussed on the mailing
list. Please see the ChangeLog and the documentation for details.
Also, a few more examples have been added. (Special thanks to Peter
Bernschneider for his Sudoku solver example.)
** Pure 0.39 2009-12-06
Besides some bugfixes, this release features a new code generation option to
perform constant folding in the compiler frontend, and a pragma syntax to
specify code generation options in the source on a per-rule basis. Please see
the ChangeLog for details.
** Pure 0.38 2009-12-01
This release sports some bugfixes (see the ChangeLog for details), as well as
improvements in the handling of const aggregates and signal exceptions. Here
is a list of the most important user-visible changes:
- Only constant scalars are still inlined, constant aggregates are now cached
in a read-only global variable instead. As discussed on the mailing list,
this makes the efficient handling of big constant values much easier for the
programmer.
- Some extra checks were added to the prolog of Pure functions, so that stack
overflows and pending signals are now caught reliably. This is enabled by
default, but as it incurs a (typically small) runtime cost, generation of
the extra checks can be controlled with the PURE_NOCHECKS environment
variable as well as the new --nochecks and --checks command line options of
the interpreter.
- The PURE_NOTC environment variable and the --notc/--tc command line switches
can now be used to control the use of tail call optimization at runtime.
** Pure 0.37 2009-11-21
Pure goes LGPL (GNU Lesser General Public license), in order to enable
commercial Pure programs and extensions. More precisely:
* The Pure runtime and standard library are now under the LGPLv3+.
* pure.cc remains under the GPLv3+ (readline dependency). However, an
alternate, readline-free version (pure_norl.cc) is now available under a
BSD-style license.
* The examples are now under a BSD-style license as well.
Please see the README (or 'help copying') for details.
Other user-visible changes (please see the ChangeLog for details):
* Support for BSD editline/libedit as a drop-in replacement for GNU readline.
Also, you can now optionally disable readline and/or editline support at
configure time (--without-readline, --without-editline). Neither is required
to build Pure any more.
* Removed all dependencies on the GSL library. Numeric Pure matrices now work
without GSL.
* XML entity escapes in strings were updated to the latest W3C entity
definitions (2009-11-19).
* The prelude defines a new 'sort' function which can be used to sort lists
and matrices using a given order predicate.
* Bugfixes in the operator precedence parser.
Moreover, Michel Salim has kindly provided Pure RPM packages for all recent
Fedora releases. Thanks Michel! You can find these packages here:
http://koji.fedoraproject.org/koji/packageinfo?packageID=9272
** Pure 0.36 2009-10-06
This is mainly a bugfix release, see the ChangeLog for details. Also, there
are some improvements in the runtime library, and the 'val' operation can be
used to parse simple expressions now, as suggested by Vili Aapro.
Please note that the changes in the runtime required a bump of the runtime
library version, so you'll have to recompile your addon modules as well.
** Pure 0.35 2009-09-14
This release sports various PowerPC compatibility fixes, please see the
ChangeLog and the SYSTEM NOTES section in the INSTALL file for details.
Moreover, support for online help was added to Emacs Pure mode (C-c C-h in
Pure and Pure-Eval mode, also available in the Pure menu). To get the most out
of this, make sure that you have emacs-w3m (http://emacs-w3m.namazu.org/)
installed.
** Pure 0.34 2009-09-03
This release sports some additional fixes for various compilation quirks
reported by Roman Neuhauser, as well as some changes and improvements as noted
below.
- The pure_interp_compile() routine in the runtime library was beefed up to
better support realtime modules such as pd-pure. (This required a bump of
the runtime library version, so you'll have to recompile addon modules once
again, sorry for the hassle.)
- The batch compiler was overhauled to add LLVM bitcode support and remove the
dependency on llvmc and llvm-gcc (standard gcc is used to handle assembly
and linkage now). Please see the ChangeLog for details.
- At long last, the 'nullary' keyword, which has been deprecated since Pure
0.26, was removed, so it isn't a special keyword any more. If you haven't
updated your Pure sources yet, you'll have to do so now.
** Pure 0.33 2009-08-28
Another bugfix release. This one fixes some more incompatibilities with latest
LLVM revisions and some failed LLVM assertions hit by the overhauled version
of the batch compiler if LLVM was compiled with --enable-assertions. Moreover,
the linkage options were massaged a bit, reportedly they work better with
FreeBSD and FC12 now.
Emacs Pure mode now provides sensible defaults for the location of the Pure
interpreter executable and the library directory if PATH and/or the PURELIB
environment variable are set accordingly. This makes it easier to install Pure
mode on Windows, as you don't have to edit the file any more if you want to
use it with the Pure MSI package.
Thanks are due to Roman Neuhauser for helping to sort out the issues
preventing Pure to work on FreeBSD.
** Pure 0.32 2009-08-27
This release fixes the performance issues with the batch compiler on the
latest LLVM versions in svn, and the source was updated once more to make it
compile with the latest LLVM revision. Also, 'pure --version' now prints the
LLVM version the interpreter was compiled with. The test system was revised so
that 'make check' returns a proper error code now. Emacs Pure mode was
overhauled as well. The new mode is derived from fundamental mode and the
broken auto-indentation stuff was removed.
** Pure 0.31 2009-08-24
This release brings support for LLVM 2.6+ and some further improvements in the
parser and the batch compiler, please see the ChangeLog for details.
** Pure 0.30 2009-08-22
This release sports a new -s option for the batch compiler which strips unused
functions from the generated code. This considerably reduces both compilation
times and code size (especially if the program imports huge libraries such as
pure-gtk). However, note that the batch compiler only does a static analysis
of which functions can be reached from the initialization code, so you have to
be careful if your program uses 'eval', 'evalcmd' or 'val'.
** Pure 0.29 2009-08-20
Another bugfix release. This one fixes some Windows-related issues (see the
ChangeLog for details) and a bug in the 'clear ans' command which prevented
the 'ans' value from being garbage-collected. Moreover, sentries can now
be placed also on symbols and function objects.
** Pure 0.28 2009-08-16
This release fixes an annoying bug in the new operator precedence parser.
Also, namespace settings now stick when running a script interactively or
sourcing the interactive initialization files, as suggested by Max Wolf.
In addition, various useful process and I/O-related functions were added to
the system module and the following new primitives are now available in the
prelude (see the library manual for details):
- The 'reduce' macro enables you to implement local rewriting rules. See
examples/rules.pure for an example.
- The 'blob' and 'val' functions provide for safe expression serialization.
See examples/fork.pure for an example.
** Pure 0.27 2009-08-09
This is a bugfix release which cleans up some of the mess in qualid/type tag
resolution left behind by the changes in the namespace system and the addition
of user-defined type tags in the 0.26 release. In particular, it fixes a
critical bug in the lexer which caused any global qualid of the form ::foo to
be mistaken for a type tag. Also, it is now also possible to specify a qualid
as a type tag.
** Pure 0.26 2009-08-07
This started out as a mere bugfix release, but then it became much more. A
bunch of bugs were fixed, some implementation restrictions were removed, and
some interesting new features were implemented. (This also required some
changes in the internal runtime API, so existing modules have to be
recompiled.)
You can find the most important fixes and improvements in this release
below. As usual, please see the ChangeLog for details.
- As suggested by John Cowan, the 'nullary' keyword has been replaced with
'nonfix', which is more in line with the other fixity keywords. Existing
Pure scripts should be edited accordingly. For the time being, 'nullary'
still works as a synonym for 'nonfix', but the compiler warns that it is
deprecated.
- The expression parser went through a major overhaul so that it essentially
supports an unlimited number of precedence levels now. (The implementation
still imposes a limit, but it's very large.) The standard operator system
has been overhauled as well, so that it becomes much easier to sneak in
additional levels. There are now 20 standard precedence levels which are
numbered from 1000 to 2900, see prelude.pure for details. The relative
priorities aren't changed, so most existing scripts should be unaffected,
but of course you may have to adjust the precedences of your own operator
declarations accordingly.
- The code generated for numeric and string list/tuple/matrix constants has
been improved a lot. This speeds up compilation and enables the interpreter
to handle much bigger constant aggregates of these types than before.
- Values used in constant definitions and 'eval' calls may now contain
arbitrary run time data. Thus you can now write stuff like 'const x = malloc
10;' or 'const y = \x->x*x;'. (This only works in interpreted scripts, the
batch compiler still forbids run time data in constants.)
- Tail call elimination now works in batch-compiled scripts.
In addition, the following new features have been implemented (please see the
manual for details):
- Added support for Miranda/Haskell-style operator sections. Thus you can now
write '(x+)' for '(+) x' and '(+x)' for 'flip (+) x', respectively.
- Added support for 'outfix' operators which let you declare unary operators
in the form of pairs of custom bracket symbols, such as: 'outfix BEGIN END;'.
These have been adopted from Wm Leler's Bertrand language.
- Added support for hierarchical namespaces (suggested by Max Wolf). Also,
qualified module names are now permitted in 'using' clauses and will be
translated to corresponding pathnames, as suggested by Eddie Rucker.
- Added the notation 'x::bar', where 'bar' must be a known function identifier
or nonfix symbol, as a shorthand for the "as" pattern 'x@(bar _)'. This is
convenient with special data types (like the dictionaries and sets in the
standard library) which are represented using their own unary data
constructors.
- Added readdir function to the system module (constributed by Jiri Spitz).
Many thanks to all who reported bugs and suggested improvements!
** Pure 0.25 2009-06-19
The biggest change in this release is that in the C interface, 'long' now
denotes a real C long int (whose size usually depends on the architecture),
instead of always assuming 64 bit integers. This makes it easier to interface
to C functions involving such values. Real 64 bit integers are now denoted
'int64', and 'int8/int16/int32' are provided as synonyms for the
'char/short/int' types. Affected bits and pieces in the standard library were
fixed to accommodate these changes, see the ChangeLog for details.
Unfortunately, this also required changes which break backward compatibility
in the runtime library, so existing addon modules must be recompiled as
well. Moreover, the following addons have been updated, so you should make
sure that you run the latest releases of these (required versions are
indicated): pure-ffi (0.6), pure-gen (0.6), pure-gl (0.6), pure-gtk (0.3),
pure-odbc (0.3).
In addition, there is a new 'ans' function to retrieve the most recent result
printed on the interactive command line. Some minor bugs that crept into the
0.23/0.24 releases were fixed as well. Please see the ChangeLog for details.
** Pure 0.24 2009-05-30
This is a maintenance release featuring some additions and cosmetic changes in
the runtime API (as requested by John Cowan), and an overhaul of arithmetic
sequences of double values (these now round the upper bound to the nearest
grid point, in order to prevent rounding artifacts, as reported by Eddie
Rucker).
** Pure 0.23 2009-05-27
This release features various bugfixes and optimizations in interpreter and
library, see the ChangeLog for details. Also, support for quoted expressions
was improved in various ways. In particular, please note the following
user-visible changes:
- Matrix transposition is now a function, transpose x. (The single quote is
now used for quoting expressions, see below.)
- 'x is provided as a synonym for quote x. Moreover, macro and constant
substitution is now inhibited in quoted subterms. (OTOH, special constructs
and local variables inside quoted expressions are still evaluated as usual.
Please see the new subsection "The Quote" of the "Caveats and Notes" section
in the manual for details.)
- For a symbolic matrix x, 'x now inhibits the implicit splicing of nested
submatrices. Symbolic matrices with nested submatrices are now printed in
quoted form, so that they can be reconstructed from their textual
representation.
- Two new (and much improved) alternative implementations of Lisp-like
quasiquote are available. These are now provided as separate library
modules. Please note that these are still considered experimental right now.
Also, as suggested by John Cowan, the BROWSER environment variable is now used
by the interpreter to determine the help browser if PURE_HELP is not set. See
http://www.catb.org/~esr/BROWSER/.
Finally note that the runtime library number was bumped to 1.0, so addon
modules need to be recompiled for the new version.
** Pure 0.22 2009-04-03
The interpreter now provides a simple symbolic debugger in interactive mode.
To enable this, it must be invoked with the new -g option. (Please note that
this will make your scripts run *much* slower and thus should be used only to
run the debugger.) Please see section "Debugging" in the Pure Manual for
details.
** Pure 0.21 2009-03-29
This release sports some changes in the build system to make things easier for
package maintainers, and the addition of a batch compiler which allows you to
compile Pure scripts to native executables. See below for details.
Build System
------------
In preparation for the 1.0 release, the Pure runtime library now uses a proper
major.minor versioning scheme, which is independent from the interpreter
version.
Moreover, "versioned" installs (i.e., parallel installations of different Pure
versions) are now disabled by default. We decided to do this because versioned
installs presented a real headache to package maintainers. You can still do a
versioned install by running configure with the --enable-versioned option. It
is also possible to have versioned and unversioned installations coexist if
they live under different install prefixes, please see the INSTALL file for
details.
*IMPORTANT:* Because of the different installation layout, you should
uninstall any earlier Pure version before installing this release.
Also, if you're compiling from svn sources, please make sure that you rerun
configure before building the new version. You should do that anyway when
installing a new version, but it's really essential this time. :)
Batch Compiler
--------------
The interpreter now offers an experimental batch compilation option -c to
translate scripts to native executables or object files. For instance,
pure -c foo.pure -o foo
creates the executable 'foo' from the script 'foo.pure', after executing the
script as usual. This requires the *full* LLVM toolchain (including llvmc,
llvm-gcc, etc.) to work, see the INSTALL file which now also has instructions
on how to build llvm-gcc.
You can also use a command like 'pure -c foo.pure -o foo.o' to create a native
object file which can then be linked into other C/C++ programs or libraries.
For utmost flexibility, 'pure -c foo.pure -o foo.ll' allows you to create an
LLVM assembler (.ll) file. This can then be handed over to the LLVM tools,
e.g., to apply special kinds of program transformations and optimizations,
before finally compiling and linking the native executable or library.
Note that even when batch-compiling, the compiled program is executed as
usual. This is necessary because some parts of a Pure program (in particular,
'eval' as well as 'const' definitions) may be used to modify the program at
compile time, possibly depending on previous evaluations of other expressions.
There's a new system variable 'compiling' which gives the program an
indication whether it is running under the auspices of the batch compiler, so
that it can adjust accordingly. This also enables you to play with advanced
program optimization techniques such as partial evaluation
(http://en.wikipedia.org/wiki/Partial_evaluation).
Also note that the compiled executable is essentially a static snapshot of
your program at the end of the compile time execution, which is then executed
on the "bare metal", without a hosting interpreter. Only a minimal runtime
system is provided. This considerably reduces startup times and makes it
possible to deploy the executable (along with the runtime library) on systems
without a Pure installation, but also implies some limitations, mostly
concerning the use of the built-in 'eval' function.
Please see the manual for details. In particular, have a look at
pure.html#compiling-scripts and pure.html#batch-compilation to get started
with this useful new feature.
** Pure 0.20 2009-03-19
This is a bugfix release. Most notably, two unrelated bugs in pattern matching
code were fixed, please see the ChangeLog for details.
** Pure 0.19 2009-03-14
This release brings a completely revamped online documentation system ('help'
command), which now uses documentation in html format (generated with
docutils, see http://docutils.sf.net/) and an external web browser (w3m by
default, http://w3m.sf.net/) to read the documentation. The Pure manual has
been converted, and fairly extensive library documentation is now also
available. (This was generated from the library sources, using the new
pure-doc utility which is now available as a separate package.)
The other major advance is the integration of Scott E. Dillard's optimized
list-style matrix operations (map, foldl et al), which have been living on a
separate branch for a while already, and have now been merged back into the
trunk. These are *much* faster (usually an order of magnitude or more) than
the previous implementation which employed the corresponding list functions to
realize these operations.
List and matrix comprehensions have also been optimized to fully take
advantage of the new operations. This is most noticeable with matrix
comprehensions, but list comprehensions should be faster as well. Moreover,
matrix comprehensions drawing values from other matrices now preserve the
block layout of component matrices, see the ChangeLog and the manual for
details.
Special thanks are due to Scott Dillard for his substantial contributions
which, besides the new matrix operations, also include the OpenGL bindings and
his OpenGL bindings generator which became the new pure-gen utility.
** Pure 0.18 2009-02-22
Besides some bugfixes, fixes in the build system and some refactoring of the
standard library (e.g., malloc and friends are now in primitives.pure), this
release adds a getopt module (ported from Q) and marshalling of matrices to
short*, int*, float*, double* and void* arguments in the C interface.
Moreover, the language now permits 'public' and 'private' in an 'extern'
declaration. As usual, please check the ChangeLog for details.
** Pure 0.17 2009-01-30
This release features several minor bugfixes and a few changes and additions
in the library, please see the ChangeLog for details. Also, a severe
performance issue with the creation of symbolic matrices (reported by Scott
E. Dillard) was fixed.
** Pure 0.16 2009-01-12
Yet another bugfix release. Fixes a critical bug reported by Vili Aapro which
caused generation of wrong pattern-matching tables in some cases. Also adds
some missing and corrects some existing definitions in the prelude, and fixes
some quirks with the Makefile. Last but not least, the sources now compile
cleanly with Bison 2.4 and the forthcoming LLVM 2.5 (svn). As usual, the gory
details can be found in the ChangeLog.
** Pure 0.15 2008-12-16
Another bugfix release. Moreover, local definitions (when, with) may now span
the right-hand side and the guard of a rule. Thus local definitions like in
'foo x = bar y if y>0 when y = baz x end;' now apply to both the rhs and the
guard. Clauses can still be made local to the guard by placing parentheses.
Please see the manual for details.
** Pure 0.14 2008-11-21
This release features support for the recently released LLVM 2.4, as well as
some minor changes in the Unicode support and the C interface, see the
ChangeLog for details. The biggest change in this release is the new support
for namespaces, which is based on John Cowan's proposal on the mailing
list. This encompasses the following language changes:
- New 'public' keyword to explicitly declare public symbols. (This is still
the default for all special symbol declarations, such as 'nullary' or
'infix'.)
- New 'namespace' keyword which serves to declare and use namespaces. Please
refer to the manual for details. Briefly, you create a new namespace with
a declaration of the form 'namespace <identifier>;' and then declare the
symbols which should belong to the namespace. For instance:
namespace foo;
public foo bar; // 'foo' and 'bar' are public symbols in namespace 'foo'
foo x = x-1;
bar x = x+1;
// ...
namespace; // switches back to the default namespace
The 'namespace' declaration affects *all* symbols in subsequent symbol
declarations, including 'extern' declarations. For instance:
namespace foo;
extern double sin(double);
foo::sin(0.1);
- New qualified symbol syntax: foo::bar. This gives you access to the (public)
symbols in a namespace, and works with both ordinary identifiers and special
(nullary and operator) symbols. The namespace must first be introduced with
a 'namespace' or 'using namespace' declaration (see above for the former,
and below for the latter).
- New 'using namespace' declaration. This gives you access to the (public)
symbols in the given namespace(s) without proper qualification. For
instance:
using namespace foo; // opens the namespace 'foo' for searching
foo (bar x); // 'foo' and 'bar' resolve to 'foo::foo' and 'foo::bar' here
using namespace; // closes the search namespace(s)
foo (bar x); // 'foo' and 'bar' are symbols in the default namespace here
The interpreter first searches the current namespace (as specified with the
'namespace' declaration), then the search namespaces specified in the most
recent 'using namespace' clause, and finally the default namespace. Note
that 'private' symbols are only visible in the *current* namespace, see
below for more details on this.
- Private symbols are now limited in scope to a namespace, not a source file.
To create and use private symbols, a namespace declaration has to be in
effect. You'll have to adapt existing scripts accordingly. For instance:
namespace foo;
private baz; // 'baz' is a private symbol in namespace 'foo'
baz x = 2*x;
// ...
namespace; // 'baz' is now invisible
** Pure 0.13 2008-11-04
This release sports a lot of minor bugfixes and other cosmetic changes, see
the ChangeLog for details.
In addition, please note the following language and library changes, as
discussed on the mailing list. Special thanks are due to John Cowan, Eddie
Rucker and Max Wolf for bringing these up and helping to get them sorted out.
- Allow 'const nullary' symbols which can be matched, e.g., in 'case'
expressions just like a real 'nullary' symbol.
- Swapped '~' and 'not', so that the logical and bitwise operations are now
more consistently named ~, &&, || and not/and/or, respectively.
Consequently, '!=' and '!==' were also renamed to '~=' and '~==',
respectively. Note that previously ~ was used for bitwise and 'not' for
logical negation, which was rather confusing, albeit compatible with the
naming of the 'not' operation in Haskell and ML.
- Improved Unicode support. As proposed by John Cowan, extended characters in
operator symbols are now restricted to symbols U+00A1 through U+00BF,
U+00D7, U+00F7 and U+2100 through U+2BFF. All other Unicode characters are
now usable in ordinary identifiers just like any ASCII letter, which permits
writing Pure programs in almost any language.
Last but not least, Eddie Rucker contributed Pure syntax highlighting for GNU
nano (http://www.nano-editor.org). Thanks Eddie!
** Pure 0.12 2008-10-09
Another step on the road to Pure 1.0.
- Cosmetic syntax change: Allow the lhs to be omitted in simple rules ('when',
'let' etc.) if it is just '_'. This lets you write stuff like, e.g.,
'foo x = y when y = ...; puts y; end', which is useful for debugging
purposes. Also, 'let x;' provides an alternative to just 'x;' at the
toplevel, which suppresses printing the result of x.
- New Lisp-like 'quote' builtin (special form) for quoting expressions.
Added a Lisp-like 'quasiquote' to the prelude (see examples/quasiquote.pure
for an example).
- 'eval' can now evaluate expressions as well as strings. Also added a new
'evalcmd' primitive for executing interactive interpreter commands like
'clear' and 'show' (see examples/reflection.pure for an example).
- Overhauled the 'clear', 'dump' and 'show' commands so that they take the
same basic set of selection options and work in a consistent fashion now.
- For all practical purposes, the number of definition levels is unlimited
now (previous limit was 256).
** Pure 0.11 2008-10-06
This release sports some minor bugfixes as well as a few changes in the
language as detailed below.
- The old [x;...] list comprehension syntax has been removed (this has already
been deprecated since Pure 0.7.)
- The grammar was changed to allow arbitrary expressions as guards in
equations. This means that syntax like 'foo x = bar x if y when y = baz x
end' is now recognized (previously you had to enclose the 'y when ... end'
term in parentheses).
- There's direct support for empty lists and tuples in the grammar now, so
that these are treated in the same way as empty matrices, rather than being
defined as nullary symbols in the prelude.
- The -F (filename) option of the 'dump' command was renamed to -n, since it
was too easy to confuse -F with the -f (dump functions) option.
- As a bonus, there's also a new example (trace.pure) showing how to trace
function calls for debugging purposes.
** Pure 0.10 2008-10-05
Fixed a bug in the expression printer, and added some operations to create
(sub/super-) diagonal matrices to matrices.pure.
Also please note that the Pure project has moved to Google Code, see
http://pure-lang.googlecode.com. There's also a new mailing list at
http://groups.google.com/group/pure-lang.
** Pure 0.9 2008-10-01
Another bugfix release. This one fixes some minor issues with string slices.
Moreover, contiguous list and string slices of the form x!!(i..j) are now
optimized to make them work in linear time.
** Pure 0.8 2008-09-28
This is a maintenance release. It fixes a bug in catmap which slipped into the
0.6 release (reported by Eddie Rucker) and some minor glitches in the Makefile
and the documentation. Also, the matrix conversion operations in matrices.pure
have been overhauled (in particular, the matrix-pointer conversions are much
more complete now). Details can be found in the ChangeLog.
** Pure 0.7 2008-09-26
This release brings an important new feature: GSL (GNU Scientific Library)
matrix support. Here's a brief overview of the new stuff. For more
information on GSL please refer to http://www.gnu.org/software/gsl.
GSL double, complex and integer matrices can be created with the new {x,y;u,v}
syntax, which works more or less like Octave/MATLAB matrices, but using curly
braces instead of brackets. Also, indices are zero-based (rather than
one-based) to be consistent with Pure's list and tuple structures. Here are
some simple examples of matrices:
- {1,2,3} a row vector consisting of machine ints
- {1.0;2.0;3.0} a column vector of double values (';' separates rows)
- {1,2;3,4} a 2x2 int matrix
- {1L,y+1;foo,bar} a symbolic matrix
Symbolic matrices work like the numeric matrices, but can contain an arbitrary
mixture of Pure values, and also work if GSL support isn't available. GSL
matrices are always homogeneous, i.e., they only contain values from one
numeric type, which in the Pure implementation can be machine ints, double and
complex double values. If a matrix contains values of different types, or Pure
values which cannot be stored in a GSL matrix, then a symbolic matrix is
created instead (this also includes the case of bigints, which are considered
as symbolic values as far as matrix construction is concerned). If the
interpreter was built without GSL support then symbolic matrices are the only
kind of matrices supported by the interpreter.
Pure also provides so-called matrix comprehensions as a convenient means to
create matrices from a template expression (which can denote either a scalar
or a submatrix), drawing values from lists or matrices and (optionally)
filtering the elements with predicates. These work pretty much like list
comprehensions, but return matrices instead of lists. Generator clauses in
matrix comprehensions alternate between row and column generation so that
customary mathematical notation carries over quite easily. Here's a simple
example showing how to define a function which returns a square identity
matrix of a given dimension:
> eye n = {i==j|i=1..n;j=1..n};
> eye 3;
{1,0,0;0,1,0;0,0,1}
The prelude provides some additional basic matrix operations like determining
the size and dimensions of a matrix, indexing and slicing, transposition,
type-checking and various conversions between the different kinds of
matrices. To these ends, various basic operations to deal with matrix objects
have been added to the runtime, including some public API operations to create
and inspect Pure matrix values from external C modules. Moreover, the
'pointer' function in the prelude can be used to convert matrices to Pure
pointer values, and marshalling of GSL matrices in the C interface is also
provided.
Here is a brief synopsis of some important operations which are implemented in
the prelude (you can find the definitions of these and a bunch of other matrix
operations in matrices.pure):
- #x total number of elements
- dim x number of rows and columns (as a pair)
- x' transpose of a matrix
- x!i ith matrix element in row-major order (zero-based)
- x!(i,j) jth element in ith row of the matrix (zero-based)
- x!!is, x!!(is,js) slicing (is and js are lists of machine ints)
- x==y, x!=y matrix comparisons
Adding other operations by interfacing to GSL should be a piece of cake. In
fact, we plan to provide a comprehensive Pure interface to GSL as a separate
library in the future.
Other user-visible changes prompted by the introduction of the matrix syntax
are listed below:
- Changes in the comprehension syntax: '|' is now used to separate the
template expression and the generator and filter clauses. This change was
necessary to accommodate the matrix syntax which uses ';' to separate
different rows in a matrix. For consistency, this applies to both list and
matrix comprehensions. The old [x;...] syntax is still supported in list
comprehensions, but is flagged with a "deprecated" warning by the compiler.
- Arithmetic sequences with a stepsize different from 1 are now written x:y..z
instead of x,y..z. This makes it possible to give the .. operator a higher
precedence than the ',' operator, which makes writing matrix slices like
x!!(i..j,k..l) much more convenient.
** Pure 0.6 2008-09-12
New stuff in this release (please see the ChangeLog and the manual for
details):
- Macros: These are implemented as rewriting rules which are applied at
compile time, and can be used for all usual preprocessing purposes, including
the optimization and inlining of function calls at the source level, and the
programming of user-defined special forms.
- Thunks a.k.a. "futures": These are realized as anonymous parameterless
closures (a la Alice ML), and are used to implement call-by-need and lazy data
structures. In particular, the prelude now implements lazy lists a.k.a.
"streams".
- Private namespaces. You can now declare symbols as private in a module, to
hide away internal helper functions and constructor symbols, and to keep the
global public namespace clean.
- Sentries (object finalizers) and references (mutable expression pointers).
File objects in the system module now employ sentries in order to close
themselves when they're garbage-collected.
- New interactive startup files (.purerc). These are just normal Pure scripts
with additional definitions for interactive usage.
- The 'list' command was renamed to 'show' (to avoid a clash with the prelude
function 'list'), and a new 'dump' command was added. The latter is similar to
'show', but saves a snapshot of your current interactive environment in a
file.
- User-defined hook for the expression printer (__show__). This allows you to
define your own custom print representations for expressions at runtime.
- Syntax highlighting for gedit (contributed by Eddie Rucker, thanks!).
- "PurePad", a little graphical frontend for the Pure interpreter on Windows.
** Pure 0.5 2008-08-24
This release sports LLVM 2.3 support and a bunch of bug fixes and improvements
in the language, the standard library and the code generator. As usual, please
check the ChangeLog for details. Here is a brief rundown of the most important
changes:
- Language: Haskell-like 'as' patterns. Constant definitions. Revised
list-of-tuples and 'using' syntax, as discussed on the mailing list. New $$
sequence operator. The XML entity character escapes were updated to the latest
from W3C.
- Updated syntax highlighting modes. Kate mode now supports folding of
comments and block structure.
- Improved script and dynamic library search algorithms, as discussed on the
mailing list.
- Various improvements in the C interface. Also refactored the runtime library
to provide a semantically complete public API for module writers.
- Improvements and bugfixes in the code generator.
- Library: Jiri Spitz' port of the Q container types (array.pure, dict.pure,
heap.pure, set.pure). New math.pure module which implements additional
mathematical functions as well as complex and rational numbers. New time- and
signal-related functions in the system module.
- More examples. In particular, make sure you have a look at Libor Spacek's
cool Mayan calendar and his unbelievingly fast n-queens algorithm. :)
- Better OSX support.
- Thanks to Rooslan S. Khayrov's patches, this release now works with LLVM
2.3. Please note that LLVM 2.2 support has been dropped, as we encountered
various issues with the LLVM 2.2 JIT.
- Toni Graffy has contributed openSUSE packages (available via Packman),
Alvaro Castro Castilla a Gentoo ebuild. Ryan Schmidt continues to maintain the
MacPorts package.
A big thank you to all who reported bugs and contributed code and patches, in
particular: Alvaro Castro Castilla, Toni Graffy, Rooslan S. Khayrov, Eddie
Rucker, Ryan Schmidt, Libor Spacek and Jiri Spitz.
** Pure 0.4 2008-06-19
This release features some more bug and portability fixes, a cleanup of the
source tree and an overhaul of the build system, see the ChangeLog for
details. Building a separate runtime lib on x86-64 works now (but requires a
patched LLVM, see the INSTALL file for details). Moreover, it is now possible
to install different Pure versions in parallel.
An Emacs mode for Pure and support for executing Pure scripts using "shebangs"
has been added. Paging of the 'list' command is now implemented using the
program specified with the PURE_MORE environment variable. This allows you to
disable this option (if PURE_MORE is undefined) or choose any pager program
and options that you prefer. Define PURE_MORE=more in your shell startup files
to get back the old behaviour of piping 'list' output through 'more'.
There's also a new syntax for multiple left-hand sides in function definitions
and 'case' rules, as suggested by Jiri Spitz and discussed on the mailing
list. Please refer to the manual page for details. To accommodate this change,
the bitwise operators '&' and '|' were renamed to 'and' and 'or',
respectively.
** Pure 0.3 2008-06-06
This release sports a lot of improvements as well as bug and portability
fixes, see the ChangeLog for details. Many memory leaks have been plugged, and
tail call elimination has been improved a lot. The build system has gone
through a major overhaul, adding autoconf support. 64 bit support has been
improved as well, and Pure now builds and runs fine on MS Windows. Many
library functions have been rewritten to make them tail-recursive, and some
new functions have been added. Last but not least, the runtime support is now
implemented as a separate shared library which makes it possible to link
external modules against the runtime, and reduces the memory footprint when
multiple instances of the interpreter are run as different processes.
Special thanks to Tim Haynes, John Lunney, Eddie Rucker, Ryan Schmidt, Libor
Spacek and Jiri Spitz for contributions, suggestions and bug reports.
** Pure 0.2 2008-05-04
On the heels of Pure 0.1 comes the first bugfix release which addresses a
couple of bugs, misfeatures and Mac OSX compatibility issues, please refer to
the ChangeLog for details. I also added a more detailed INSTALL guide (thanks
are due to Eddie Rucker who wrote most of the new material in this guide) and
updated the manpage with a few minor corrections and some remarks about issues
raised on the Pure mailing list.
Please note that there are still some issues with Pure on 64 bit systems (as
well as on Ubuntu running on PowerPC) which are still on my TODO list, these
will hopefully be fixed in the next release.
Thanks to all who sent in bug reports and patches, in particular: Chris
Double, Tim Haynes, Eddie Rucker, Ryan Schmidt and Libor Spacek. (I hope I
didn't forget anyone.)
** Pure 0.1 2008-04-29
The much-awaited initial release. ;-) The interpreter is already fully
functional, but of course there's still a lot to be done (see the TODO file
for details). Please note that this is a preliminary, "beta" release, so
expect some bugs (and please report them to the author!).
The Pure project is now hosted at SourceForge, see http://pure-lang.sf.net. A
mailing list should soon be available, too.
See the INSTALLATION section in the README file to get up and running quickly.
After Pure is installed, read the Pure manual page (also available in various
formats from the Pure website) and have a look at the stuff in the examples
subdir, especially hello.pure, and review the standard library modules
(lib/*.pure).
Enjoy!
Albert Graef <Dr.Graef@t-online.de>
