mercury/NEWS

NEWS for Mercury 20.06.1
========================

This is a bug-fix release.

* [Mantis bug #519]. We have deleted an old workaround from the Tcl/Tk
  binding in the extras that causes linking errors with recent versions
  of Tcl/Tk.

* [Mantis bug #514]. We have fixed a bug that caused linking errors
  in the deep profiling version of the Mercury runtime library when using
  GCC with the `-fno-common` option.

  Users of GCC 10 should note that `-fno-common` is now the default.

* [Mantis bug #513]. We have fixed a bug that caused the compiler to
  generate code that incorrectly reads variable values from the stack
  in disjunctions.

* [Github issue #90]. We have fixed a bug that the caused the compiler
  to abort on code that used type classes in lambda expressions.

* We have fixed a bug that caused the low-level C backend to incorrectly
  place the contents of `foreign_code` pragmas after the contents of
  `foreign_proc` pragmas in the generated C code.

* We have fixed a bug with the debugger's `level` command where it would
  report an error when invoked without any arguments instead of resetting
  the current ancestor level to the level of the current event.


NEWS for Mercury 20.06
======================

Changes that may break compatibility
------------------------------------

* We have deleted the `one_or_more` type, and the operations on it, from
  the `list` module of the Mercury standard library, and moved them to
  a new module named `one_or_more`. This will break compatibility for
  code that used the moved functionality, but such breaks can be fixed
  simply by importing `one_or_more` as well as (or in rare cases, instead of)
  the `list` module.

* If module A contains a `:- import_module` or a `:- use_module` declaration
  for module B but does not refer to anything defined in module B, we used
  to generate a warning for the declaration only if module A had no submodules,
  because it is possible that A's submodules refer to entities defined in B.
  We now generate a warning for unused `:- import_module` and `:- use_module`
  declaration in this case as well, which will stop the program from compiling
  if `--halt-at-warn` is set. The fix is simple: replace the declaration
  being warned about in the parent module with declarations in the child
  modules that need it. In the fairly frequent case that not all child modules
  need module B, this avoids the need to recompile those child modules
  when the interface of module B changes.

* We have moved the type `domain_error` from the `math` module of the Mercury
  standard library to the `exception` module. Any module qualified references
  to the name of either the type or of its one function symbol should have
  the module qualification either updated or deleted. The reason for the move
  is that it reduces the cost of some new optimizations.

* We have added an additional constructor to the `poly_type` type from
  the `string` module. Existing code that uses that type may need to be
  updated.

Changes to the Mercury standard library
---------------------------------------

### New module: `kv_list`

* This module, whose name is short for key-value list, contains the same
  functionality as the existing `assoc_list` module, but stores lists of
  key-value pairs in a more space-efficient manner. Specifically, it uses
  one three-word cell on the heap to represent each pair, while the
  `assoc_list` module uses two two-word cells, which means that `kv_list`
  will allocate fewer heap cells. The tradeoff is that unlike assoc_lists,
  kv_lists are not standard lists, and thus cannot be manipulated
  using the functions and predicates of the `list` module.

### New module: `one_or_more`

* This module defines the `one_or_more` type, which represents nonempty lists,
  and provides operations on the values of this type. For every operation
  in the `list` module that makes sense for nonempty lists, there is a
  corresponding operation in `one_or_more`.

### New module: `one_or_more_map`

* This module implements `one_or_more_map`, a map from keys to one or more
  values. Its functionality is very similar to that of the `multi_map` module
  (in fact, the two modules define the exact same set of operations),
  but unlike `multi_map`, it uses the type system to enforce the invariant
  that every key in the map must have at least one associated value.

### New module: `thread.closeable_channel`

* This module implements closeable unbounded channels, similar to the unbounded
  channels provided by the `thread.channel` module, but with the addition of a
  close operation. Once a channel is closed, no more items can be added to it,
  and reading from a closed channel will not block indefinitely.

### Changes to the `assoc_list` module

* The following predicates and functions have been added:

    - pred `lookup/3`
    - pred `update/4`
    - pred `foldl/4`
    - pred `foldl2/6`
    - pred `foldl3/8`
    - func `foldl_keys/3`
    - func `foldl_values/3`

### Changes to the `char` module

* The following function and predicate have been added:

    - func `hash/1`
    - pred `hash/2`

### Changes to the `float` module

* The following predicate has been added:

    - pred `hash/2`

### Changes to the `hash_table` module

* The following predicates have been deprecated and will be removed in a future
  release:

    - pred `int_hash/2`
    - pred `uint_hash/2`
    - pred `float_hash/2`
    - pred `char_hash/2`
    - pred `string_hash/2`
    - pred `generic_hash/2`

* The following obsolete functions have been removed:

    - func `new/3`
    - func `new_default/1`

### Changes to the `int` module

* The following function and predicate have been added:

    - func `hash/1`
    - pred `hash/2`

### Changes to the `integer` module

* The following functions have been added:

   - func `eight/0`
   - func `sixteen/0`

### Changes to the `io` module

* The predicates `write_many/3` and `write_many/4` now work directly
  with values of type `uint`.

* The predicates `format/4` and `format/5` now work directly with values
  of type `uint`. (See the changes to the `string` module for further
  details.)

### Changes to the `list` module

* The following predicate has been added:

    - pred `map_corresponding3/5`

* The following type, functions and predicates have been moved to the
  new `one_or_more` module. Note: this change can break backwards
  compatibility.

    - type `one_or_more/1`
    - func `one_or_more_cons/2`     (as `one_or_more.cons/2`)
    - func `one_or_more_to_list/1`
    - pred `list_to_one_or_more/2`
    - pred `det_list_to_one_or_more/2`

### Changes to the `map` module

* The following predicates and functions have been added:

    - pred `filter_map_values/3`
    - pred `filter_map_values_only/3`
    - pred `foldl4_values/10`
    - pred `foldl5_values/12`
    - func `keys_as_set/1`
    - pred `keys_as_set/2`

### Changes to the `multi_map` module

* The following predicates and functions have been added:

    - func `sorted_keys/1`
    - pred `sorted_keys/2`
    - func `keys_as_set/1`
    - pred `keys_as_set/2`

### Changes to the `string` module

* Formatting of `uint` values is now directly supported by `format/2` and
  `format/3`. The `poly_type/0` type has been extended with a new
  alternative, `u/1`, to allow this.

### Changes to the `stream` module

* We have added a new `reader/4` subclass, `unboxed_reader/4`. This subclass
  allows readers to provide a `get` operation that avoids boxing non-error
  outputs.

### Changes to the `stream.string_writer` module

* The predicate `format/5` now work directly with values of type `uint`.
  (See the changes to the `string` module for further details.)

### Changes to the `thread.channel` module

* The following predicate has been deprecated and will be removed in a future
  release:

    - pred `untake/4`

### Changes to the `thread.mvar` module

* The documentation of the following predicates has been clarified to match the
  actual implementation:

    - pred `try_take/4`
    - pred `try_read/4`

### Changes to the `tree234` module

* The following predicates have been added:

    - pred `filter_map_values/3`
    - pred `filter_map_values_only/3`
    - pred `foldl4_values/10`
    - pred `foldl5_values/12`

### Changes to the `uint` module

* The following function and predicate have been added:

    - func `hash/1`
    - pred `hash/2`

### Changes to the `version_array` module

* The following obsolete function has been removed:

    - func `unsafe_new/2`

### Changes to the `version_hash_table` module

* The following predicates have been deprecated and will be removed in a future
  release:

    - pred `int_hash/2`
    - pred `uint_hash/2`
    - pred `float_hash/2`
    - pred `char_hash/2`
    - pred `string_hash/2`
    - pred `generic_hash/2`

Changes to the Mercury compiler
-------------------------------

### Alternative script to invoke the compiler

* We have added an alternative script, `mercury`, to use when invoking the
  compiler on systems where the usual name, `mmc`, clashes with other
  executables.

### Changes to the treatment of unused imports

* The compiler now generates warnings for `:- import_module` and
  `:- use_module` declarations in the interface of a module even if
  that module has submodules. Previously, it generated such warnings
  only if the module had no submodules.

### Changes to code model options

* `--trail-segments`

  Grades that support trailing now always use trail segments. This means
  that the `--trail-segments` option now has no effect, and it is therefore
  deprecated.

* `--high-level-data`

  We have deleted the `--high-level-data` option. Its effects are now
  implied by the target language.

* `--tags`

  We have deleted the `--tags` option. Its effects are now implied
  by the target language.

### Changes to warnings

* [Github issue #85]. The compiler will now generate a warning about
  a disjunct that cannot succeed even if the predicate or function in which
  that disjunct occurs has other modes in which that disjunct can succeed.

  The generation of such warnings can now be disabled by wrapping
  the whole disjunction, or a larger goal containing it, in a
  `disable_warnings` scope, like this:

        disable_warnings [no_solution_disjunct] (
           Goal
        )

* [Mantis bug #497]. The compiler will now generate a warning for an
  unrecognised warning name in a `disable_warnings` scope instead of
  reporting an error.

### New warning options

* `--print-errors-warnings-when-generating-interface`

  Until now, the compiler did not try to detect problems with a module
  when generating its interface files. Now it does. To preserve compatibility,
  by default it still ignores any problems it finds then, but if this option
  is enabled, and if it does in fact find any problems, the compiler will
  report them, and if they are severe enough, it will not generate the
  interface file at all. In the usual use case where the compiler is asked
  to generate the interface file as part of rebuilding the whole executable,
  this behavior has the advantage that the rebuild, which would fail anyway,
  fails sooner.

### New auxiliary output options

* `--show-definition-extents`

  When this option is specified, the compiler will write out a list
  of the predicates and functions defined in the module, together with
  the approximate line numbers of their first and last lines,
  to `module.defn_extents`. The list will be ordered on the starting
  line numbers of the predicates and functions.

Changes to the Mercury implementation
-------------------------------------

* Grades that support trailing now always use trail segments; it is no longer
  possible to use a fixed size trail.

  One consequence of this is that the `trseg` grade component now acts
  as a synonym for the `tr` component. Since `tr` is shorter, `trseg`
  is deprecated in its favor.

  Another consequence is that the `--trail-size` and `--trail-size-kwords`
  runtime options no longer have any effect, and are deprecated.

* We have upgraded the bundled Boehm GC to v7.6.12 and libatomic_ops to v7.6.8.

For news about earlier versions, see the HISTORY file.