New in 3.1:

* The generated C code is now in ANSI-C by default. If you want to support

  pre-ANSI-C compilers, you need to provide the option --language=C on the

  command line or %language=C in the source file.

* The 'len' parameter of the hash function and of the lookup function is now

  of type 'size_t' instead of 'unsigned int'. This makes it safe to call these

  functions with strings of length > 4 GB, on 64-bit machines.

* Added option --constants-prefix.

* Added declaration %define constants-prefix.

New in 3.0.4:

* gperf is now licensed under the GPL version 3 or newer.

* The generated C code gives less warnings with gcc-4.2.x in c99 or gnu99 mode.

New in 3.0.3:

* The generated C code is compatible with gcc-4.3.x in c99 or gnu99 mode.

New in 3.0.2:

* Compiles with g++-4.0.x.

* Added option --length-table-name.

* Added declaration %define length-table-name.

* Fixed  #line directives for filenames containing backslashes.

New in 3.0.1:

* Bug fix.

New in 3.0:

* Added option --output that allows to specify the output file name.

* Some options have been renamed:

    --hash-fn-name=NAME     -->   --hash-function-name=NAME

    --lookup-fn-name=NAME   -->   --lookup-function-name=NAME

    --compare-strlen        -->   --compare-lengths

    --global                -->   --global-table

  The older variants are still supported for backward compatibility.

* The following options can now be specified inside the input file:

    %delimiters=DELIMITER-LIST

    %struct-type

    %ignore-case

    %language=LANGUAGE-NAME

    %define slot-name NAME

    %define initializer-suffix INITIALIZERS

    %define hash-function-name NAME

    %define lookup-function-name NAME

    %define class-name NAME

    %7bit

    %compare-lengths

    %compare-strncmp

    %readonly-tables

    %enum

    %includes

    %global-table

    %pic

    %define string-pool-name NAME

    %null-strings

    %define word-array-name NAME

    %switch=COUNT

    %omit-struct-type

* When the option -k is not given, the default key positions are now

  computed depending on the set of keywords.

* If the input file is given by name, the output file will now contain

  #line directives referring to the input file.

* Some keyword sets containing permutations, like { "xy", "yx", "xz", "zx" }

  or { "abc", "acb", "bca", "cab" }, are now handled by gperf without

  requiring the option -D.

* The generated table is usually much smaller than it was with earlier

  versions of gperf.

* Added option -m/--multiple-iterations that allows to further reduce the

  size of the generated table.

* When the search for a good hash function is not immediately successful,

  the table's size will grow as needed. Earlier versions of gperf bailed

  out with an "Internal error, duplicate hash code value".

* The options -f/--fast and -o/--occurrence-sort have no effect any more.

* Added options -P/--pic and --null-strings that optimize the generated code

  for use in shared libraries. -P/--pic does a perfect optimization but may

  require some small code changes (see the documentation for details), whereas

  --null-strings does only a half-hearted optimization but works without

  needing any change to surrounding code.

* Added option --ignore-case that produces a case independent lookup function.

* Bug fixes.

New in 2.7.2:

* Keywords may now be enclosed in double quotes; this permits the use of

  '#', ',', space or NUL inside keywords.

* Bug fixes.

New in 2.7.1:

* Added option "-F" for gcc.

New in 2.7:

* gperf is now a stand-alone package, untied from libg++.

* Autoconfiguring.

* Removed the "-a" and "-g" options, extended the "-L" option instead.

* Removed the "-p" option, it is the default.

* Added long options ("--help", "--version" etc.).

* 8-bit cleanliness is now the default; use "-7" to get the old behaviour.

* Compiles with any C++ compiler.

* Numerous small improvements.