news

早期freebsd实现

        One minor possible incompatibility exists (ostream<< char)
        that can be eliminated via #define NO_OUTPUT_CHAR at the
        top of stream.h. This is probably necessary in order to
        compile et++. See libg++.texinfo for more details.

        The new stream implementation is not particulary pretty.
        Its main virtue is that it avoids some of the worst things 
        about both AT&T streams and old libg++ streams. A much
        superior AT&T 2.0-iostream superset is in the works
        for future release.

*** changes from libg++-1.25.0 to libg++-1.32.0

    * Nearly all data and methods previously declared as `private'
        are now declared as `protected' to make subclassing from
        library classes easier.

    * Most classes now contain a method `OK()' that checks to
        make sure that an objects internal data is in a valid state.
        See the documentation for further details

    * Mosts tests in ./tests now contain various kinds of assert
        statements. If tests execute without assertion failures,
        and without any other errors that cause aborted execution,
        you may consider them successful. The test files now include
        nearly all of my internal tests, which in turn have been
        made more extensive.

    * Plex classes are available. 
        A bit of propaganda about these:
        Plexes are very attractive replacements for arrays
        in many contexts. Try them!

    * `shrink' and `contains' have been added to Obstack

    * Files and streams have not yet been revamped (a complete
        reworking awaits the AT&T 2.0 stream specification), but
        have been adjusted to provide slightly faster IO in some 
        cases, and now support the use of a user-supplied string 
        to read or write from instead of an actual IO source.

    * Rationals are now normalized when input via `>>'.

    * atoI is fixed.

    * variable-length representations in String, Integer, etc., are
        now completely different. See the documentation for details.

    * New String functions:
       * readline -- read in a line from an istream as a String
       * prepend -- prepend stuff to a String
       * through  -- make a SubString from beginning to match point
       * from     -- make a SubString from match point to end
       * s[i]     -- now returns the char by reference

    * BitVec's are no longer supported, since their capabilities
        are now incorporated in the revised BitString class.

    * stdarg.h and regex.h are revised to support the Sun4

    * Several interesting and/or useful examples of libg++
        class use are in ./etc, mainly courtesy of Doug Schmidt.

    * .cc file names are now all less than 15 characters. Some
        .h file names are longer, but this should not present
        problems on SYSV systems.

    * pseudo-generic `proto' classes have been thoroughly revised:
        * The prototype file names now end in `P', not `.proto'
            and are in the g++-include directory
        * A single, simple-to-use collection traversal mechanism
            via pseudo-indices (`Pix') is used instead of particular
            traversal friend classes.
        * `Bag' prototypes are included
        * Base classes are now supplied, so that all implementations
            of Sets, Bags, are derived, allowing programmers to
            mix and match implementations.
        * They are now simpler to create: all comparison operators
            and the like are now defined as macros in a `defs' file
            which uses some reasonable defaults.
        * The `Dictionary' versions of Set prototypes are not now
            supported. Revised versions of such classes are forthcoming.
        * Class prototypes previously labelled as `Assoc' are now
            called `Maps', with slightly different capabilities.

    * Splay tree prototypes are available.
        A bit more propaganda:  Consider using Splay tree
        based containers (Set, Bag, Map, PQ). They are often
        the most efficient structures when performing mixtures
        of operations (adds, deletes, searches...)

    * Fixed precision reals are available, courtesy of Kurt Baudendistel

    * An ordered hash Set prototype (VOHSet) is available,
        courtesy of Doug Schmidt.

    * MLCG now allows access and modifications to the seeds.

    * The Normal random generator is fixed.

    * SampleStatistic now allows any prob value for confidence
        intervals.

    * some simple timer routines are in builtin.cc, courtesy of Doug Schmidt

    * While the Vec class prototypes are still available, they are
        currently undergoing revision in order to correspond to
        the forthcoming Matrix package (which should be available
        in the next libg++ release).

    * A C++ version of GPERF, a perfect hash function generator
       program is also available in ./etc, courtesy of Doug Schmidt.
 

*** changes from libg++-1.22.2 to libg++-1.25.0

  * All reported errors from the previous release are fixed, and many
    suggested modifications have been performed. Thanks to all who
    have sent bug reports and comments, including those with mail
    addresses that I have not been able to reply to. (My mail connections
    are sometimes more than a little fragile. If you send me mail and
    I have not replied within a few days, you may want to try again. Sorry.)

  * A serious problem in the use of Obstacks by conversion functions 
    has been repaired.

  * A bug in maintaining reference counts on `find' and related operations
    in List prototype classes has been fixed.

  * Strings now support self case manipulation in addition to the
    functional versions.

  * Some new functions have been added to builtin.[h.cc]. All are now
    documented in libg++.texinfo

  * Overload declarations are now performed in std.h and math.h,
    rather than builtin.h,  to eliminate some include file ordering
    problems.

  * Random, RNG, and SampleStatistic classes are available,
    thanks to Dirk Grunwald.

  * A BitVec class is available.

  * The are many new generic container class prototype files. These
    support container classes based on elementary data structures
    (and a couple of non-elementary ones). Please read the documentation.

  * Class prototype OSet has been modified and renamed OLSet.

  * The genclass utility has been modified

*** changes from libg++-1.22.1 to 1.22.2

  * Tests files are now in a separate directory.

  * Several minor errors (Complex unary -, several BitString functions)
    have been repaired.

  * pow(0,0) returns 1 for all versions of pow.

  * An experimental generic class prototyping feature is provided,
    including prototypes for lisp-style lists and ordered list-based sets.

  * several include files with the same names as those in AT&T CC
    are provided. These simply #include other files. OOPS should now
    compile using only g++-include files. See, however, the note
    about struct exception in file math.h

  * Some rearrangement of files containing char* conversion has been
    done to eliminate linking of unnecessary classes.

  * The inline-only-when-optimizing feature is still not supported.

* New Random number generator classes are not yet available. Stubs for
  these files are in this directory. They should be available for next
  release.

*** changes from libg++-1.21.1 to 1.22.0

  * All documentation is in a stand-alone texinfo file, libg++.texinfo,
    and is on its way to becoming a decent piece of documentation.

  * All reported errors from the previous release are fixed, and many
    suggested modifications have been performed. Thanks to all who
    have sent bug reports and comments, including those with mail
    addresses that I have not been able to reply to. (My mail connections
    are sometimes more than a little fragile. If you send me mail and
    I have not replied within a few days, you may want to try again. Sorry.)

  * New Complex, BitSet, and BitString classes are available. You will
    be performing a valuable service if you try these out and report
    back any bugs/comments/suggestions about these or any other classes.

  * File `values.h' has been added. This contains various system
    constants like the number of bits per long, etc. It contains
    much of the same information as sun <values.h>, although a
    few names and things differ slightly.

  * Files `builtin.h', `builtin.cc', and `convert.cc' have been added.
    `builtin' contains common inline and non-inline functions on
    builtin types (like `abs'). `convert' contains code for performing
    IO and char* conversions, mainly via Obstacks. Most of these
    functions are not new -- they have been collected from other .h
    and .cc files.

  * Files `std.h' and `math.h' now declare all libc.a C functions 
    in a way that allows any of them to be overloaded in C++.

  * Strings and Integers now perform expansion via realloc() -- 
    see libconfig.h about whether you should #define SHOULD_FREE_TO_REALLOC.

  * `eatwhite' is supported for istreams.

  * File::getline(String, ...) and get(String) have been removed in order to
    maintain greater independence of different classes.

  * Strings now provide substring matching via new versions of
    `contains' and `matches', `common_prefix' and `common_suffix'.
    Also, there is more support for case operations via `fcompare', etc.
    The versions of `decompose' that do not deal with Regexes have
    been deleted since they provide little functionality over other
    operations. Also, a few special case functions dealing with char*'s
    have been deleted since the required constructors are necessary
    anyway. Several other corrections have been made in String.cc,
    including the elimination of a few aliasing problems.

  * The implementations of Integers and Rationals are now both much
    more efficient and well-tested, while remaining machine independent.
    A few minor visible features have been added and/or changed.
        
  * The `box' command in PlotFile is now simulated, whether or not
    it is present in libplot.a

  * The inlining-only-if-optimizing feature is still not fully implemented.