2009-09-24, version 1.3.1
=========================

Implementation changes
----------------------

* Fixed a crash in double_approx(cl_RA) on 64-bit platforms.

* Add basic support for Renesas SH (sh4).


2009-06-30, version 1.3.0
=========================

Other changes
-------------

* Use the GNU autotools as build system.

* Implemented a more portable module dependency mechanism.


2008-04-05, version 1.2.2
=========================

Implementation changes
----------------------

* Re-establish CLN-1.2.0 ABI and fix ARM build, both inadvertently broken in
  the previous release.


2008-03-24, version 1.2.1
=========================

Implementation changes
----------------------

* Fixed some bugs in the output of numbers larger than 2^32 decimal digits.

Other changes
-------------

* Modifying C/C++ operators like +=, ++, etc. are now enabled by default.


2008-01-19, version 1.2.0
=========================

Algorithmic changes
-------------------

* Save big amounts of memory in computation of some functions and constants by:
    - Avoiding pre-computation of series terms and instead computing them in a
      streamed way.
    - Avoiding computation with excess precision in binary splitting algorithm
      by coercion of intermediate integer values to floating-point as early as
      possible.

Implementation changes
----------------------

* Added support for huge numbers:
    - intC used for all counter loops.
    - intE is now a 64-bit type on all 64-bit platforms and even on selected
      32-bit platforms.

* CLN now uses C++ exceptions for error handling. The cl_abort() hook is not
  supported any more. Please refer to the documentation to learn about existing
  exception types.

* Fixed a bug on i386 where comparing Z/2Z ring zeros returnd random results.

* Removed cl_boolean. Use built-in C++ bool instead.

Other changes
-------------

* Dropped the cln-config script. Please use pkg-config instead.

* Updated infrastructure to that of libtool-1.5.24.

* Changed encoding of all files to UTF-8.

* Fix compilation issues with GCC-4.3.

* Fix linking issues on platforms that don't feature weak symbols (like win32).


2006-08-08, version 1.1.13
==========================

* Compilation fixes for 64-bit brokenness introduced in last release.


2006-08-06, version 1.1.12
==========================

Implementation changes
----------------------

* Fix rare assertion when printing quite large floats.

Other changes
-------------

* Compilation fixes for several platforms: *BSD, Intel Mac, and MinGW.


2005-11-23, version 1.1.11
==========================

Algorithmic changes
-------------------

* Considerably improved performance of number input.


2005-10-22, version 1.1.10
==========================

Implementation changes
----------------------

* Removed the vptr of modular integer and univariate polynomial
  classes in order to fix some crashes in that sector of CLN.  Code
  using this needs to be recompiled.

* Many more operator signatures make the integration of CLN types with
  float/double easier.

Other changes
-------------

* Several compilation fixes.

* Made it possible to cross-compile CLN.


2004-11-03, version 1.1.9
=========================

Algorithmic changes
-------------------

* Input of numbers in bases 2, 4, 8, 16 and 32 is now done in linear bit
  complexity as opposed to O(N^2). Useful for all kinds of persistency.

Implementation changes
----------------------

* Fixed several bugs in the integer input and output routines that could be
  blamed for random crashes in the following cases: output in base 32 for
  quite large numbers, input in base 2 for fixnums and input in base 3 for
  fixnums on a 64 bit target.

* Fixed crash when radix specifiers were used in input streams.

* Speed up on x86_64 and ia64 by adding some inline assembly.

Other changes
-------------

* Fixes for compilation on MacOS X and little endian Mips.


2004-07-01, version 1.1.8
=========================

Implementation changes
----------------------

* Fix bug in log(a,b) when b is 1/n.

* Fix crash in shared library on AMD64.

Other changes
-------------

* Fixes for some language conformance issues.


2004-05-02, version 1.1.7
=========================

Implementation changes
----------------------

* Fixed two corner case bugs in cl_LF multiplication and sqrt.

* Workaround GCC 3.[123] compiler bug.

* Parallel build should work reliably now.


2004-01-01, version 1.1.6
=========================

Functional changes
------------------

* New function
  `sintL ldegree(const cl_UP& x)'.
  (Suggested by Munagala Ramanath.)

Implementation changes
----------------------

* Fixed a bug in polynomial subtraction.
  (Reported by Munagala Ramanath.)

Other changes
-------------

* Provide a package control file cln.pc.

* Updated infrastructure to autoconf-2.57.

* Fixed compilation issues with GCC-3.4 prereleases and on Mac OSX.


2002-05-28, version 1.1.5
=========================

Other changes
-------------

* Fix compilation issues with GCC-3.1.


2002-01-04, version 1.1.4
=========================

Other changes
-------------

* Fix compilation issues with GMP-4.0 and on IBM s390.

* Updated infrastructure to autoconf-2.52.


2001-11-05, version 1.1.3
=========================

Implementation changes
----------------------

* Fix a computation error of sin and sinh for small arguments.
  (Reported by Christopher Kennedy.)

* Module ordering works with gcc-3.0 -fuse-cxa-atexit now.

* -DNO_ASM works properly on Sparc.


2001-07-25, version 1.1.2

Implementation changes
----------------------

* Minor cleanups and portability issues.
  (No need to upgrade if you have been happy so far.)


2001-05-31, version 1.1.1
=========================

Implementation changes
----------------------

* Minor cleanups for GCC 3.0 and libstdc++-v3.

* Fixes several little bugs and portability issues.


2000-12-14, version 1.1.0
=========================

Functional changes
------------------

* ISO/IEC 14882 fashion adjustments:
  Put everything into namespace cln.  All fundamental data types still
  carry their old names.  Other non-macro identifiers are now written as 
  cln::foo instead of cl_foo, except where the cl_ comes from a data type
  (as in cl_I_to_int()).  Headers are installed into a separate
  directory, #include <cln/foo.h> instead of <cl_foo.h>.  Applications
  must be manually ported to the new style.  We apologize for the
  inconvenience.

Implementation changes
----------------------

* Removed internal copy of mpn subdirectory tree from libgmp-2.0.2.
  Configuring with --with-gmp now relies on an installed libgmp library
  version 3 or above.  We recommend libgmp-3.1 or newer.

* Adjusted several break-even points to make better use of a faster libgmp and
  better match present-day CPUs.

* Fix several errors for architectures with 64-bit wide words.
  (Others than Alpha, which worked already.)

* Fix compilation errors with current GCC-3.0 snapshots.

Other changes
-------------

* Added package tools: script cln-config and autoconf macro AC_PATH_CLN
  (in file cln.m4).


2000-01-13, version 1.0.3
=========================

Functional changes
------------------

* New function
  `cl_I doublefactorial (uintL n)'.

Implementation changes
----------------------

* Fix several little configuration errors.

* Fix some compilation errors with gcc-2.96 prereleases.


1999-09-07, version 1.0.2
=========================

Functional changes
------------------

* New functions
  `cl_I numerator (const cl_RA&)'
  `cl_I denominator (const cl_RA&)'.
  Suggested by Richard Kreckel and Sylvain Pion.

* New function `cl_equal_hashcode' for the classes cl_N, cl_R, cl_RA, cl_I,
  cl_F, cl_SF, cl_FF, cl_DF, cl_LF.
  Suggested by Richard Kreckel.

Implementation changes
----------------------

* Fix an endless loop when either of the functions
  `cl_boolean rootp (const cl_RA&, uintL, cl_RA*)',
  `cl_boolean rootp (const cl_RA&, const cl_I&, cl_RA*)'
  was called with an integer argument. Thanks to Richard Kreckel.

* Fix a bug in the addition and subtraction of rational numbers which could
  cause unnormalized rational numbers like 3/1 to be created.


1999-06-09, version 1.0.1
=========================

Algorithmic changes
-------------------

* Speed up the functions `logand', `lognand', `logandc2'.

Implementation changes
----------------------

* Fix compilation errors with gcc-2.7.2, egcs-1.1.2 and gcc-2.95.

* Fix compilation errors on HPPA, MIPS, some versions of DEC Alpha, OpenBSD,
  and SPARC with LiDIA.

* Fix compilation errors on Win32. Thanks to Mumit Khan.


1999-01-12, version 1.0
=======================

Functional changes
------------------

* New include files, containing I/O functions:
    <cl_integer_io.h>
    <cl_rational_io.h>
    <cl_float_io.h>
    <cl_sfloat_io.h> <cl_ffloat_io.h> <cl_dfloat_io.h> <cl_lfloat_io.h>
    <cl_real_io.h>
    <cl_complex_io.h>
  The file <cl_io.h> now contains only I/O functions for characters and
  C integers.

* To access the base ring of a univariate polynomial ring, now use
  `R->basering()' instead of `R->basering'.

* Implement `plusp', `max', `min' for the classes cl_F, cl_SF, cl_FF, cl_DF,
  cl_LF, cl_RA, cl_I.

* Implement `abs' for the class cl_RA.

* Replaced `read_number' with specialized functions `read_complex',
  `read_real', `read_float', `read_rational', `read_integer'.

* Replaced the functions
  `void fprint (cl_ostream stream, int x)'
  `void fprint (cl_ostream stream, unsigned int x)'
  `void fprint (cl_ostream stream, long x)'
  `void fprint (cl_ostream stream, unsigned long x)'
  with
  `void fprintdecimal (cl_ostream stream, int x)'
  `void fprintdecimal (cl_ostream stream, unsigned int x)'
  `void fprintdecimal (cl_ostream stream, long x)'
  `void fprintdecimal (cl_ostream stream, unsigned long x)'

Algorithmic changes
-------------------

* The function `cl_I xgcd (const cl_I& a, const cl_I& b, cl_I* u, cl_I* v)'
  now normalizes the resulting u and v to be of minimum absolute value.
  (Suggested by the LiDIA group.)

* The conversion from string to cl_F, cl_R, cl_N now chooses the format of
  the resulting float depending on the number of significant digits present
  in the input string.

* Speed up the functions
  `cl_R operator* (const cl_RA&, const cl_LF&)'
  `cl_R operator* (const cl_LF&, const cl_RA&)'
  `cl_R operator/ (const cl_RA&, const cl_LF&)'
  `cl_LF operator/ (const cl_LF&, const cl_RA&)'

Implementation changes
----------------------

* Fix for `cl_I ash (const cl_I& x, const cl_I& n)' when |n| is between
  2^29 and 2^37.
  (Reported by Thomas Papanikolaou.)

* Fixed a memory leak in the long-float truncate function.
  (Reported by Keith Briggs.)

* Fixed a memory leak in the modular integer exponentiation.

* Reduced the stack space consumption. The default stack size (typically
  8 MB on Unix) should now be sufficient in order to run all CLN programs.

* Portability fixes for m68k platforms.
  (Reported and fixed by Roman Hodek.)

Other changes
-------------

* Restructured the sources. The subpackages
    - base
    - base + integer
    - base + integer + rational
    - base + integer + rational + float
    - base + integer + rational + float + real
    - base + integer + rational + float + real + complex
    - base + integer + modinteger
  are now self-contained.

* The configuration script can be called with option `--enable-shared',
  to build CLN as a shared library.

* Updated the LiDIA interface.


1998-05-07, version 0.98
========================

Implementation changes
----------------------

* Portability fixes for 64-bit platforms like DEC Alpha.
  (Reported by John Cremona and Thomas Papanikolaou.)


1998-02-27
==========

Other changes
-------------

* Portability fixes for new ISO/ANSI C++ compilers.

* Updated the LiDIA interface for LiDIA version 1.3 and fixed a couple of
  bugs.


1997-09-06
==========

Implementation changes
----------------------

* Portability fix for i686 platforms.
  (Reported by Michael Pfeifer.)


1997-09-01
==========

Functional changes
------------------

* New include files:
  <cl_ring.h>, declaring general rings,
  <cl_GV.h>, <cl_SV.h>, defining vectors,
  <cl_univpoly.h>, defining univariate polynomial rings.

Algorithmic changes
-------------------

* Speed up the internal computation of ln(2) and ln(10) by about 20%.

* Speed up the computation of exp (for 1000 digits or more) by about 10%.

Implementation changes
----------------------

* Portability fix for 64-bit platforms like DEC Alpha:
  Fixed a bug: endless loop in `sqrt' for large precision.
  (Reported by Blake Jones.)

* Portability fix for RS/6000 platforms.


1997-02-21
==========

Algorithmic changes
-------------------

* Speed up the computation of the elementary transcendental functions:
  exp, sinh, cosh, asinh, atanh (for 100 digits) by about 30%,
  log (for 100 digits) by about 25%,
  sin, cos (for 1000 or 10000 digits) by about 20%,
  asin, acos, atan (for 1000 digits) by about 25%.

Implementation changes
----------------------

* Portability fix for SunOS 4 and other BSD platforms.


1997-01-31
==========

Algorithmic changes
-------------------

* Fixed a bug: endless recursion in `cl_boolean sqrtp (const cl_RA&, cl_RA*)'.

* Fixed a bug: Only the first 85% of the digits returned by `cl_eulerconst'
  for moderate precision (< 11370 digits) were correct.
  (Reported by Thomas Papanikolaou.)

Implementation changes
----------------------

* Fix static initialization order bug. Requires g++ 2.7.0 or newer
  (g++ 2.8.0 or newer on HPPA).

* New method `debug_print()', for pretty printing of CLN objects, intended
  to be called from the debugger.


1997-01-07
==========

Functional changes
------------------

* New functions `float cl_float_approx (const cl_R&)',
                `double cl_double_approx (const cl_R&)'
  for converting a real number to a C `float' or `double'.

Algorithmic changes
-------------------

* Speed up `cl_zeta' using Cohen-Villegas-Zagier convergence acceleration.

Implementation changes
----------------------

* New function `cl_print', for pretty printing of CLN objects, intended to be
  called from the debugger.

* Portability fixes for 64-bit platforms like DEC Alpha.


1996-11-29
==========

Functional changes
------------------

* New include files:
  <cl_string.h>, defining strings,
  <cl_symbol.h>, defining symbols,
  <cl_proplist.h>, defining property lists,
  <cl_condition.h>, defining conditions/exceptions.

* New functions `cl_F cl_catalanconst ()',
                `cl_F cl_catalanconst (const cl_F&)',
                `cl_F cl_catalanconst (cl_float_format_t)'
  which return Catalan's constant.
  (Suggested by Thomas Papanikolaou.)

* New functions `cl_F cl_zeta (int s)',
                `cl_F cl_zeta (int s, const cl_F&)',
                `cl_F cl_zeta (int s, cl_float_format_t)'
  which return the Riemann zeta function at an integral point s>1.

* New functions `cl_F cl_exp1 ()',
                `cl_F cl_exp1 (const cl_F&)',
                `cl_F cl_exp1 (cl_float_format_t)'
  which return e = exp(1).

* New function `cl_I binomial (uintL n, uintL k)' which returns the
  binomial coefficient (n choose k).

* New functions `int cl_I_to_int (const cl_I&)',
       `unsigned int cl_I_to_uint (const cl_I&)',
                `long cl_I_to_long (const cl_I&)',
       `unsigned long cl_I_to_ulong (const cl_I&)'
  for converting an integer to a C `int' or `long'.

* New functions `float cl_float_approx (const cl_I&)',
                `float cl_float_approx (const cl_RA&)',
                `double cl_double_approx (const cl_I&)',
                `double cl_double_approx (const cl_RA&)'
  for converting a rational number to a C `float' or `double'.

Implementation changes
----------------------

* Moved the sources to subdirectories.
  (Suggested by Jörg Arndt.)

Other changes
-------------

* New benchmark for better comparing LiDIA, Pari and CLN.

* Added an interface to LiDIA, allows using CLN as a replacement of libI.
  (Suggested by Thomas Papanikolaou.)

* Added an ILOG Talk interface, for interactive use of the library.


1996-10-13
==========

Functional changes
------------------

* New include file <cl_modinteger.h>, defining modular integer rings.

* New functions `cl_F cl_eulerconst ()', `cl_F cl_eulerconst (const cl_F&)',
  `cl_F cl_eulerconst (cl_float_format_t)' which return Euler's constant.

Algorithmic changes
-------------------

* Speed up square root of large numbers by use of Newton's algorithm.

* Speed up multiplication and division of large long-floats by small integers.

* Speed up the computation of pi, e, exp, log, sin, cos, tan, sinh, cosh, tanh,
  asin, acos, atan, asinh, acosh, atanh.
  All of them are now computed with asymptotically fast algorithms, of
  bit complexity O(log(N)^2*M(N)) = O(N^(1+epsilon)) for N digits.

* Fixed several bugs in the transcendental functions routines. In particular,
  the `log' function went into an endless loop.

* Fixed a bug: The cl_I -> cl_DF and cl_I -> cl_LF conversions didn't work
  correctly for negative integers.


1996-09-08
==========

Functional changes
------------------

* New include file <cl_numtheory.h>, containing number theoretic functions,
  for now only the Jacobi symbol.

Algorithmic changes
-------------------

* Speed up squaring of large numbers by 30%.

* Speed up division of large numbers by use of Newton's algorithm. The
  improvement is noticeable for numbers with at least about 1000 words =
  32000 bits = 10000 decimal digits.

* Speed up the binary-to-decimal conversion of large numbers by use of
  a divide-and-conquer method. The improvement is noticeable for numbers
  of at least 1250 words = 40000 bits = 12000 decimal digits.

* Fixed a bug: The functions `cl_boolean minusp (const type&)' and
  `cl_boolean zerop (const type&)' were missing for type=cl_I and type=cl_RA.

* Fixed a bug: The function `cl_boolean logtest (const cl_I&, const cl_I&)'
  didn't work correctly if both arguments were bignums.

* Fixed a bug: The function `cl_I mod (const cl_I&, const cl_I&)', when called
  as `mod(-n*m,m)' (n>0), returned m instead of 0.

* Fixed a bug: The function `uintL power2p (const cl_I&)' returned wrong
  values for arguments x = 2^n, n >= 29.

Implementation changes
----------------------

* Speed up CLN by using the low-level addition/subtraction, shift and
  multiplication loops from GNU gmp-2.0.2.

* Fixed a severe reference counting bug in the constructors
  `type::type (const long)' and `type::type (const unsigned long)'
  for type=cl_number, type=cl_N, type=cl_RA, type=cl_I that most often
  caused a core dump.

* CLN's word sequences can be chosen to be stored big-endian in memory
  (like it was up to now) or little-endian (needed for interfacing to GMP).


1996-05-20
==========

Functional changes
------------------

* New include file <cl_timing.h>, contains timing facilities.

Algorithmic changes
-------------------

* Speed up the multiplication of large numbers by use of an FFT based
  algorithm (Schönhage-Strassen). The improvement is noticeable when both
  factors have at least 1250 words = 40000 bits = 12000 decimal digits.

* Speed up the functions `cl_I gcd (const cl_I&, const cl_I&)' and
  `cl_I xgcd (const cl_I&, const cl_I&, cl_I*, cl_I*)' by a factor
  of 1.5. Simple trick.

* Speed up the function `cl_boolean sqrtp (const cl_I&, cl_I*)' using a
  trick from Henri Cohen.

* Fixed an endless loop in the function `uint32 gcd (uint32, uint32)'
  which occurred when one of the arguments is zero.

* Fixed an endless loop in the function `cl_I factorial (uintL)'.

Implementation changes
----------------------

* <cl_io.h> now uses iostreams by default instead of stdio FILE pointers.
  (Reported by Tito Flagella.)

* Fixed linking error when compiling without -O.
  (Reported by Tito Flagella.)

* Fixed syntax error in <cl_input.h>.
  (Reported by Tito Flagella.)

* Fixed syntax errors in src/cl_DS.h and src/cl_FF_plus.h.
  (Reported by Marcus Daniels.)

* More portability fixes.

* Configuration now supports the --srcdir option.
  (Reported by Marcus Daniels.)


1996-01-03
==========

* Compiles well on Sparc, using gcc.

* Mail to beta testers.


1995-12-06
==========

* Compiles well on Linux, using gcc.