mercury/NEWS

NEWS since Mercury 14.01.x
--------------------------

Changes that may break compatibility:

* When the Mercury compiler looks at code using Mercury keywords (such as
  "func" and "pred") and Mercury operators (such as clause neck operator ":-")
  but which cannot be parsed as the Mercury constructs that those Mercury
  keywords and operators are part of, it now generates a specific error
  message for each discrepancy. In the past, when it found such code,
  the compiler considered such code to construct terms and/or to call
  functions and predicates. Usually, the terms did not belong to any
  declared type and the called functions and predicates did not exist,
  leading to confusing error messages from the compiler.

  The compiler's new approach generates considerably better diagnostics,
  but it does mean that code that uses Mercury's keywords and/or operators
  in ways that are inconsistent with Mercury's own uses of those keywords
  and operators won't compile anymore. Such code will need to be changed,
  typically by changing the names of some function symbols, functions
  or predicates.

* We have enabled stricter checking of non-ground final insts to reject more
  mode-incorrect code.  Due to compiler limitations, some code that should be
  accepted will now be rejected.  They will require modifications to appease
  the compiler.

* We have enabled stricter checking of the requirement that a type, inst
  or a mode that is *not* exported from a module may not be used in the
  declarations of entities (such as predicates and typeclasses) that *is*
  exported from that module. This may require the explicit export of e.g.
  some types that previously were not exported.

* We have enabled stricter checking of type declarations and definitions:
  all declarations and definitions of a type must agree on whether the type
  is a solver type or not.

* References to everything imported via `:- use_module' declarations
  must now be fully module qualified.

* `for' is now an operator.

* It is now an error for a program to redefine a builtin type.  The affected
  type names are:

      int
      int{8,16,32}
      uint
      uint{8,16,32}
      float
      character
      string
      {}
      =
      pred
      func
      pure
      semipure
      impure
      ''

* It is now an error for a program to redefine a builtin inst.  The affected
  inst names are:

      =<
      any
      bound
      bound_unique
      clobbered
      clobbered_any
      free
      ground
      is
      mostly_clobbered
      mostly_unique
      mostly_unique_any
      not_reached
      unique
      unique_any

* It is now an error for a program to redefine a builtin mode.  The affected
  mode names are:

      =
      >>
      any_func
      any_pred
      func
      is
      pred

* We have deleted the builtin inst synonyms `old' and `new': their uses
  should be replaced with `any' and `free' respectively.

* We have deleted the builtin modes `no' and `oo': their uses should be
  replaced with `oa' and `ia' respectively.

* The minimum version of the Java platform required by Mercury's Java
  backend is now Java SE 6.

* The representation of integer constants in the standard library's term and
  lexer modules has been generalised.  The base, signedness and size of each
  integer constant is now recorded.  Furthermore, these modules now use
  arbitrary-precision integers to represent the values of integer constants.

  Code that relies on the old representation of integer constants used by
  the term or lexer modules may use the `old_term_parser' library in the
  extras instead.

* We have changed the semantics of int.(<<) and int.(>>) so that they throw
  an exception if their second operand is not in [0, bits_per_int).  For now,
  the old behaviour of these operations is provided by the functions
  int.legacy_left_shift/2 and int.legacy_right_shift/2.  These functions
  will be deleted in a future release.

Changes to the Mercury language:

* We have added a new primitive type, uint, which is an unsigned integer type
  of the same size as Mercury's int type.  Literals of this type must have the
  suffix `u', for example:

       999u
       0b1111100111u
       0o1747u
       0x3e7u

  Basic operations on uint values are provided by the new standard library
  module `uint'.  (See below for details.)

* We have added a new kind of scope to the language: determinism checks
  for switch arms. These scopes are introduced by any of the new keywords

      require_switch_arms_det
      require_switch_arms_semidet
      require_switch_arms_multi
      require_switch_arms_nondet
      require_switch_arms_cc_multi
      require_switch_arms_cc_nondet
      require_switch_arms_erroneous
      require_switch_arms_failure

  `require_switch_arms_<determinism> [Var] Goal' tells the compiler to require
  `Goal' to be a switch on `Var' in which all the switch arms have determinisms
  at least as tight as <determinism>, and to generate error messages
  for any violations of this requirement.

* We have changed the meaning of require_complete_switch scopes slightly:
  the compiler now generates an error if the goal inside the scope
  is not a switch on the named variable.

* We have added a new kind of scope to the language for disabling warnings
  within the scope. A goal such as

      disable_warnings [singleton_vars] (
        Goal
      )

  is equivalent to Goal, with the exception that the compiler will not generate
  warnings about singleton variables inside Goal.

* We have added an extension to include external files
  in pragma foreign_decl and pragma foreign_code declarations.

* We have added a foreign type assertion `word_aligned_pointer' that
  allows the Mercury implementation to avoid boxing values of the foreign type
  that the assertion is for when the type appears as the sole argument
  of a data constructor.

* We have added a new pragma named `consider_used'. This pragma tells
  the compiler to consider the predicate or function it names to be used,
  preventing it from generating unused procedure warnings either for
  any of its procedures, or for any of the procedures they call,
  directly or indirectly.

* The Java backend now supports defining foreign types as primitive Java
  types.

* Digits in numeric literals may now be separated by underscores in order
  to improve readability.

Changes to the Mercury standard library:

* We have added a new module, uint, that exports basic operations on the new
  uint type.

  The following predicates and functions have been added to other standard
  library modules in order to support the uint type:

  - integer.to_uint/2
  - integer.det_to_uint/1
  - integer.from_uint/1
  - string.uint_to_string/1
  - io.write_uint/3, io.write_uint/4
  - stream.string_writer.put_int/4

* We have added variants of the process_options predicates to the getopt
  and getopt_io modules that represent errors using a type instead of strings.
  A new function, option_error_to_string/1, can be used to convert values
  of the new error type into strings.

* We have added the print_line and write_line family of predicates to the
  io module.  These behave like the print and write predicates, but also
  write a terminating newline.

* We have added the predicates write_array/5 and write_array/6 to the io
  module; these predicates write arrays using a user-specified procedure
  to write the elements and separating the elements with a user-specified
  separator string.

* io.print and string_writer.print now print arbitrary precision integers
  in their decimal form instead of printing their underlying representation.

* We have added temp_directory/3, make_temp_directory/3 and
  make_temp_directory/5 predicates to the io module.

* We have added a variant of io.set_environment_var that returns an io.res
  value rather than throwing an exception. We have added the predicate
  io.have_set_environment_var/0 that tests for the ability to set
  environment variables on the current platform.

* We have removed the io.poly_type equivalence type from the io module.
  string.poly_type should be used directly.

* We have made these predicates exception safe: the current input or output
  stream is restored to the original stream if an exception is thrown during
  their execution:

    - io.read/4
    - io.write_list/6
    - io.write_array/6
    - io.read_binary/4
    - io.write_binary/4

* We have added predicates to the lexer and parser modules that read from
  an explicitly specified input stream, not from the current input stream.

* We have added a module for discrete interval encoding trees, which are a
  highly efficient set implementation for fat sets.  This module is a
  contribution from Yes Logic Pty. Ltd.

+ We have added two new modules for concurrent programming: thread.barrier
  and thread.future.  The barrier module provides a barrier type which can
  be used to control progress in concurrent code.  The future module
  provides future and future_io data types which can be used to compute
  values in parallel using other threads.  These modules were contributed by
  Mission Critical IT.

* We have added a new module, ranges, that represents sets of integers using
  a sorted list of ranges.  The new module exports operations that make it
  suitable for implementing domains in finite domain constraint solvers.

* We have added thread.spawn_native/4 to dedicate an OS thread to a Mercury
  thread. thread.spawn/4 was added as well.

* We have added the function from_reverse_list/1 to the version_array module.

* We have added thread.num_processors/3 which returns the number of
  processors available for parallel work.

* The following predicates and functions have been added to the thread.mvar
  module:

    - mvar.init/1
    - mvar.impure_init/1
    - mvar.try_read/4

+ We have deprecated the impure init/1 function in thread.semaphore.

* In C grades, the math module now provides the fused multiply-add operation
  on platforms that support it.

* Procedures in the store module no longer acquire the global lock.

* The following predicates and functions have been added to the char module:

    - is_ascii/1
    - is_decimal_digit/1
    - is_base_digit/2
    - int_to_binary_digit/2, det_int_to_binary_digit/1
    - int_to_octal_digit/2, det_int_to_octal_digit/1
    - int_to_decimal_digit/2, det_int_to_decimal_digit/1
    - int_to_hex_digit/2, det_int_to_hex_digit/1
    - base_int_to_digit/3, det_base_int_to_digit/2
    - binary_digit_to_int/2, det_binary_digit_to_int/1
    - octal_digit_to_int/2, det_octal_digit_to_int/1
    - decimal_digit_to_int/2, det_decimal_digit_to_int/1
    - hex_digit_to_int/2, det_hex_digit_to_int/1
    - base_digit_to_int/3, det_base_digit_to_int/2
    - is_leading_surrogate/1, is_trailing_surrogate/1
    - is_control/1, is_space_separator/1, is_paragraph_separator/1
    - is_line_separator/1, is_private_use/1

  The following predicates in the char module have been deprecated and will
  either be removed or have their semantics changed in a future release:

    - is_hex_digit/2
    - int_to_hex_char/2
    - digit_to_int/2
    - int_to_digit/2
    - det_int_to_digit/1, det_int_to_digit/2

  NOTE: existing code that calls char.digit_to_int/2 assuming that it will
  only succeed for decimal digits (0-9) may be broken.

* Float special values, NaNs and Infinities, are now converted to strings in
  a way that is backend- and grade-independent.  (Bug #348)

* string.base_digit_to_int/3 and string.det_base_digit_to_int/2 now check
  for overflow and underflow in all bases, not only base 10.

* We have reduced the memory allocated by string.to_lower and string.to_upper.

* The following classification predicates have been added to the float module:

   - is_finite/1
   - is_zero/1
   - is_infinite/1        (synonym for the existing is_inf/1 predicate)
   - is_nan_or_infinite/1 (synonym for the existing is_nan_or_inf/1 predicate)

  The following function has been added to the float module:

   - infinity/0

* The following predicates and functions have been added to the integer module:

   - from_string/2
   - from_base_string/3
   - to_int/2
   - det_to_int/1
   - to_base_string/2
   - negative_one/0, two/0, ten/0
   - is_zero/1

  The following functions in the integer module have been deprecated:

   - from_string/1
   - from_base_string/2
   - int/1

* The following predicates and functions have been added to the require module:

   - error/2
   - func_error/2

* The following predicates have been added to the string module:

   - is_all_alnum/1
   - is_empty/1
   - to_utf8_code_unit_list/2
   - to_utf16_code_unit_list/2
   - from_utf8_code_unit_list/2
   - from_utf16_code_unit_list/2
   - det_remove_prefix/3
   - compare_ignore_case_ascii/3

* The following predicates have been added to the map module:

   - foldl5/12
   - foldr5/12

* The following predicate has been added to the set modules:

   - intersection_and_differences

* We have a added a new type, maybe_errors, to the maybe module.

  The following predicate and function have been added to the maybe module:

   - fold2_maybe/6
   - maybe_default/2
   - pred_to_maybe/1
   - func_to_maybe/1

* The following predicates and functions have been added to the calendar
  module:

   - int_to_month/2
   - det_int_to_month/1
   - int0_to_month/2
   - det_int0_to_month/1
   - month_to_int/1
   - month_to_int0/1
   - same_date/1

* We have added a new module, psqueue, that implements a priority search
  queue ADT.  This is a blend between a priority queue and a map.  This was
  contributed by Matthias Güdemann.

* We have added the following predicate to the int module:

  - all_true_in_range/3

* We have added a predicate named is_dummy_context to the term module.

* The following predicates and functions have been added to the list module:

   - reverse_prepend/2
   - reverse_prepend/3
   - take_while/4
   - take_while/3
   - take_while/2
   - drop_while/3
   - drop_while/2

* The takewhile/4 predicate has been deprecated in the list module,
  take_while/4 can be used instead.

* The take/3 and drop/3 predicates in the list module have been modified
  so that they fail if their first argument is less than zero.

* The split_upto/4 and take_upto/3 predicates and take_upto/2 function
  in the list module have been modified so that they throw an exception
  if their first argument is negative.

* The following predicate and function in the builtin module have been
  deprecated and will be removed in a future release:

   - promise_only_solution/1
   - promise_only_solution_io/4

  Existing code that uses either of these should be replaced with code that
  uses a `promise_equivalent_solutions' goal instead.

* The following modes in the builtin module have been deprecated and will
  be removed in a future release:

   - input/0
   - output/0

  Existing code that uses these modes should replace their use with `in'
  or `out' respectively.

* The following predicates and functions have been added to the cord module:

   - to_list/1     (synonym for the existing list/1 function)
   - to_rev_list/1 (synonym for the existing rev_list/1 function)
   - cons/3
   - snoc/3
   - find_first_match/3

* The following function has been added to the array2d module

   - is_empty/1

* The following predicates have been added to the time module:

   - localtime/4
   - mktime/4

  The following functions in the time module have been deprecated:

   - localtime/1
   - mktime/1
   - ctime/1

* The following predicates have been added to the rbtree module:

   - foldl_values/4
   - foldl2_values/6

* The following predicates have been added to the hash_table module:

   - fold2/6
   - fold3/8

* We have added the predicate svremove/4 to the assoc_list module.
  This like the remove/4 predicate but with its arguments are in
  an order more conducive to the use of state variable notation.

* We have made several changes to the pretty_printer module:

  - The type formatter_limit has been renamed to func_symbol_limit,
    since this better reflects the type's purpose.
  - We have replaced the set_formatter function with a predicate
    of the same name, since this allows the use of state variable notation
    when setting up more than one type-specific formatter,
  - We have renamed the write_doc_to_stream predicate as put_doc,
    to better fit in with the names other predicates that operate on
    values of the stream.writer typeclass, and changed its interface
    to group the prettyprinter parameters together in a value of the type
    that was designed for this purpose.

* We have deprecated the following functions from the std_util module:

  - maybe_pred/3
  - maybe_func/2

* We have added the following predicates and functions to the digraph module:

  - return_vertices_in_from_to_order/2
  - return_vertices_in_to_from_order/2
  - return_sccs_in_from_to_order/1
  - return_sccs_in_to_from_order/1

* We have added the following predicates and functions to the bag module:

  - singleton/1
  - insert_duplicates/4
  - det_insert_duplicates/4
  - det_insert_duplicates/3

Changes to the Mercury compiler:

* We have added a new option --warn-dead-preds. While the existing option
  --warn-dead-procs asks the compiler to generate warnings for every
  unused procedure of the module being compiled, the new option asks
  the compiler to generate a warning for an unused procedure only if
  none of the procedures of a predicate or function is used.

* We have added a new option --warn-implicit-stream-calls, which asks
  the compiler to generate a warning for every call to a predicate p/n
  if there is also a predicate p/(n+1) which differs from p/n only in that
  it has additional argument at the front of its argument list that specifies
  an I/O stream. This is intended to generate warnings for calls to predicates
  such as io.write_string/3, which writes to the current output stream,
  to encourage programmers to call io.write_string/4 instead. If the option
  is given, the compiler will also generate warnings for calls to predicates
  such as io.see, io.seen, io.tell and io.told, which set the current input
  or output streams respectively.

* We have added a new option --warn-non-contiguous-decls, which asks
  the compiler to generate a warning if the mode declaration(s) of a
  predicate or function don't immediately follow its ":- pred" or ":- func"
  declaration. This is option is turned on by default.

* We have added the new options --warn-inconsistent-pred-order-clauses
  and --warn-inconsistent-pred-order-foreign-procs. Both of these ask the
  compiler to generate a warning if, among either (a) the set of exported
  predicates and functions of the module, or (b) the set of nonexported
  predicates and functions of the module, the order of their definitions
  does not match the order of their declarations. The first option
  applies only to predicates and functions defined by Mercury clauses;
  the second applies to predicates and functions defined by either
  Mercury clauses or foreign procedures.
  
  The option --warn-inconsistent-pred-order is a shorter synonym for
  --warn-inconsistent-pred-order-clauses.

* We have added a new option --inhibit-style-warnings, which tells
  the compiler not to generate any warnings that are purely about
  programming style, and do not point out code that is reasonably likely
  to be wrong.

* We have added a new option --inline-linear-tail-rec-sccs that tells
  the compiler, when it finds a set of procedures that each contain
  one tail call to one *other* member of the set, to inline the tail calls
  to turn those mutually recursive procedures into procedures that each
  contain only *self* tail recursion.

* We have fixed a long-standing bug causing crashes in deep profiling
  grades, related to unify/compare for tuples.  (Bug #3)

* We have fixed some bugs with constrained polymorphic modes.

* We have removed legacy support for the following systems:
    - IRIX
    - OSF/1

* The asm_fast* and reg* grades now work on 64-bit Intel OS X systems when
  using GCC as the C compiler.
  See README.MacOS for further details.

* The compiler now reports an error for binary/octal/hexadecimal integer
  literals that cannot be represented in the compiler's native int type.

* Class files generated for executables in the Java grade are now automatically
  packaged up into Java archives (JARs).

* The --generate-module-order and --imports-graph options no longer imply
  --generate-dependencies.

* The compiler does not set the runtime search path when
  --mercury-linkage=static is used.

* We have added an option --no-default-runtime-library-directory to
  prevent the compiler adding any directories to the runtime search path
  automatically.

* We have renamed the --inhibit-accumulator-warnings option to
  --no-warn-accumulator-swaps.

* We have disabled intermodule optimisation of any predicates or functions
  using `try' goals. This fixes a serious issue as `try' goals are not
  properly written to .opt files, so when read back would not actually
  catch any exceptions.

* We have disabled the old --num-reserved-objects option.

* We have upgraded to the bundled Boehm GC to 7.6.0 and libatomic_ops to
  7.4.4.

Change to the Mercury debugger:

* Interactive queries are now supported on OS X.

* The "break" command can now auto-complete on the filename:linenumber pairs
  of events, provided the Mercury system can access the readline library.

* Interactive queries have been improved:
    - Queries can use variable bindings from the current environment.
    - Exceptions thrown from queries are properly handled.
    - Non-canonical output bindings are now printed in solutions.
    - Underscore variables are no longer printed in solutions.

* We have added a "browse --web" command to view terms in a web browser.

Changes to the extras distribution:

* We have added support for Unicode and other enhancements to the lex and
  regex libraries.  Thanks to Sebastian Godelet.


NEWS for Mercury 14.01.2
------------------------

This is a bug-fix release.

* Fix array.sort, which has been buggy since 2001.  You may wish to
  reference array.sort_fix_2014 to ensure that you using the fixed version.
* Fix the handling of nondet code by the auto-parallelisation analysis in
  mdprof_create_feedback.  (Bug #364)
* Fix string.between_codepoints so that the clamping of start/end points
  works as documented.


NEWS for Mercury 14.01.1
------------------------

This is a bug-fix release.

* The function string.string/1 and related functions now handle version
  arrays properly.
* Fix resource leaks in dir fold predicates.
* The mfilterjavac program is now generated with the correct file extension
  on Windows.
* A problem that caused compilation of the Boehm GC to fail on 64-bit
  openSUSE 13.1 systems has been fixed.  (Github issue #14)
* The documentation now builds correctly on Cygwin systems.
* The script configure_mingw_cross now supports 64-bit Windows targets.
* We have added workarounds for problems with (arguably broken)
  system headers on MinGW and MinGW64 systems.
* The MinGW port now builds in the absence of POSIX threads library.
* Low-level C parallel grades now work on Windows instead of crashing
  at startup.  (Bug #338)
* We now use thread-safe alternatives to strerror().  (Bug #340)
* We have added the configure option --enable-gc-mmap.
* We configure Boehm GC to use mmap in threaded grades on Linux to avoid
  conflicts with glibc malloc leading to memory corruption.
* A problem that caused string.format/[23] to sometimes return incorrect
  results when formatting floats with the 'g' conversion specifier has
  been fixed.  This bug only affected the non-C backends.  (Bug #342)
* string.format now handles special float values (i.e. nan, inf,  and -inf)
  correctly with the non-C backends.
* A bug that caused io.write_float/[34] to append ".0" to float special values
  has been fixed.  This bug affected the C and C# backends.
* In the C# and Java grades, the predicate string.from_char_list now
  implements the documented behaviour for input lists containing null
  characters (i.e. it throws an exception).
  Likewise, for string.from_reverse_char_list in the C# grade.
* We have fixed a problem that caused `mmc --make' to attempt to install
  libraries in non-existent grades.

Changes to the Mercury compiler:

* The compiler now supports stripping of executables in a separate
  post-link step.  The new options, --strip-executable-command,
  --strip-executable-shared-flags and --strip-executable-static-flags
  are used to control this.
  (This is now the default on Mac OS X systems.)


NEWS for Mercury 14.01
----------------------

Changes to the Mercury language:

* Repeated type variables may now occur in the heads of type class instances.
  For example, instance declarations like the following are now allowed:

      :- instance foo(list(T), map(T, T)).

Changes to the Mercury standard library:

* We have added the function cord.condense/1.

* The following functions in the standard library's cord module now use
  constant stack space: foldl/3, foldl_pred/4.

* We have added the following predicates to the array and version_array
  modules: is_empty/1, all_true/2 and all_false/2.

* We have added the following predicates and functions to the map module:
  det_min_key/1, det_max_key/1, foldl2_values/6 and foldl3_values/8.

* We have added the following predicates to the list module: foldr2/6,
  foldr3/8, det_take/3 and map_foldr/5.

* We have added the following predicates to the bag module:
  foldl/4, foldl2/6, and to_list_only_duplicates/2. The predicates
  old union/3, intersect/3, least_upper_bound/3 and subtract/3 all had
  complexities that depended strongly on the size of their second argument,
  and much more weakly on the size of their first argument. We have renamed
  these to union_small/3, intersect_small/3, least_upper_bound_small/3 and
  subtract_small/3 respectively, and replaced them with new implementations
  of the original four predicates whose complexity is proportional to
  the total size of the two input arguments.

* We have added the following predicates to the assoc_list module:
  foldl2_values/6 and foldl3_values/8.

* We have added the following predicates and functions to the pqueue module:
  is_empty/1, peek/3, peek_key/2, peek_value/2, det_peek/3, merge/3,
  det_peek_key/1 and det_peek_value/1.

* We have added the predicate bimap.equal/2.

* We have added the following predicates to the int module: fold_up3/9 and
  fold_down3/9.

Changes to the Mercury compiler:

* On Mac OS X systems the compiler is now configured use the version of the
  host system as the default value for the deployment target.

  A new configuration option, `--with-macosx-deployment-target', allows
  an alternative value to be selected at configuration time.

Portability improvements:

* We have made the implementation compatible with GCC 4.8 and Visual Studio
  2013.

* We have made the implementation compatible with OS X 10.9.

Changes to the extras distribution:

* We've added a library that provides support for accessing the function
  trail from Mercury code.


For news about earlier versions, see the HISTORY file.