changes

一些常用的数据结构库

New in 1.20

    1. Bugfix in except.h. Modified non-volatile auto variables were
       being accessed after longjmp.

New in 1.19

    1. Rewrite of broken dict_free.
    2. Fixed embarassing build breakages that accidentally went into 1.18
    3. Function hash_scan_delete_free renamed to hash_scan_delfree to be
       distinct from hash_scan_delete in the first 14 characters.
    4. To resolve inconsistencies between hash_free and dict_free,
       and a difference between the actual behavior of hash_free  and
       the documented behavior, these two functions are marked obsolescent.
       The functions dict_free_nodes and hash_free_nodes are provided.
       The obsolescent functions continue to work as before, for now.
    5. Documentation of hash_free is fixed to say that it also subjects
       the hash to hash_destroy, which is what the implementation does.
    6. Documentation states what release it is for.

New in 1.18

    1. Error in assert expression in list_merge fixed.
    2. Semantics of list_merge extended to allow list to be merged
       onto itself (which is a noop).
    3. Clarified interface specification of list_transfer and list_extract;
       the source and destination list may be the same object.
    4. New functions:
       dict_init_like: create a dictionary similar to another one;
       dict_similar: determine whether two dictionaries are similar;
       dict_merge: merge contents of one dictionary to another.
    5. Dictionary test main can juggle multiple dictionaries, and test
       dict_merge. 
    6. If a hash node is inserted into some hash, it is a now a constraint
       violation to insert it again into some hash.
    7. The hash_scan_delete_free function has been implemented; it is to
       hash_scan_delete what hash_delete_free is to hash_delete.

New in 1.17

    Carl van Tast <[e-mail]>:
    1. Removed references to ``safe malloc'' from some comments.
    2. Swapped ``allowed'' and ``not allowed'' in comment to
       verify_bintree.
    3. Fixed comment to list_next: this function never returns the
       sentinel.
    4. lnode_pool_init: nodes[i].prev = nodes instead of nodes + 1. This
       saves one or two CPU cycles :-) and it gives a valid address even
       if we have a (somewhat pathological) pool with just one element.

    Kaz:
    5. Dropped extra parameter from tree rotation functions in dict.c. Should
       shave a few cycles.
    6. Fixed error in the duplicate key iteration idiom example in the
       documentation (see the section on dict_upper_bound).
    7. Forgotten #include <string.h> added to hash.c

New in 1.16

    1. Added an interface for loading the contents of a dictionary from an
       ordered sequence. This is done in O(n) time by a direct bottom-up
       construction of the red-black tree, making it much faster than
       the O(n log n) process of inserting each element.
    2. Miscellaneous cleanup: missing const qualifiers were added
       to key pointer parameters, some incorrect comments fixed;
       spelling errors corrected in documentation.

New in 1.15

    1. Another potential exception handling memory leak fixed. This one
       has to do with throwing an exception from within a try-catch region
       in which an exception was just caught. The new exception replaces
       the old without the old's dynamic memory being disposed of.
    2. Restrictions added on except_rethrow.
    3. Exception module must now be explicitly initialized with except_init.
    4. Structure members in exception header renamed to adhere to documented
       namespace.
    5. The exwrap.[ch] source files are gone. There is support for memory
       allocation with exception handling in except.c, which supports user
       defined allocators.
    6. Three bugfixes to sfx parser. First, unary operators take a cast
       expression, not a unary expression. Secondly, sizeof doesn't throw a syntax
       error anymore on things that look like casts, but maybe are not.
       Thirdly, empty parentheses weren't handled right in treatment of
       ambiguous expressions, e.g. (a)() was declared a syntax error.
    7. Changed the representation of hash table chains. They are now
       singly linked lists, which means that the overhead of managing 
       back pointers is gone. Only deletion is slightly more complicated
       now because it has to search from the beginning of the chain.
       [Rationale: this is okay, since chains are supposed to be short
       in a hash table!]
    8. Rewritten test main() in list.c. It's now more like the others
       with a menu. Previously it was essentially a file sorting program.
    9. New function: list_find. Exhaustively searches the list for a
       matching entry, returns pointer to node if found.

New in 1.14

    1. Got rid of some overbearing copyright restrictions. There is no need for
       executables to contain copyright notices. In fact, there are no
       restrictions on the use, or distribution in executable form.
    2. Tiny tweak in red-black fixup code of dict_insert.
    3. Keys in hash and dict are declared const void * now in all functions
       rather than plain void *.  This means that casts are no longer
       necessary when calling insert or lookup functions with const
       data as the key. But casts of the return value of hnode_getkey
       or dnode_getkey may be required.
    4. Fixed compile breakage of except.c when posix thread support enabled.
    5. Side effect assertion interface now performs caching, to avoid
       parsing the same expressions over and over again.  Thus debugging with
       KAZLIB_SIDEEFFECT_DEBUG incurs a smaller performance hit.
    6. Major bugfix to sfx expression parser. The function dealing with 
       disambiguating casts had to be rewritten to do more sophisticated
       lookahead and backtracking. It all started with Mark Brady discovered
       that (a++)+b was being incorrectly diagnosed as a syntax error.
    7. Added documentation. more examples for uses of dictionaries, and
       exception handling. Some documentation about the internals
       of exception handling added. Changed document format for narrower
       margins, reducing page count and increasing readability.
    8. Bugfix in except_rethrow. It was freeing the dynamic data of the
       exception even though it's not handled yet.

New in 1.13

    1. Fixed some potential memory leaks in except.c.
    2. Finished all interface documentation. All that is left now
       is to flesh out the implementation notes.
    3. Fixed a bug in POSIX threaded variant of except.c. Null
       function pointer dereference in unhandled exception case.
    4. Macros beginning with E[A-Z] have been renamed to stay out
       of space reserved for <errno.h>.
    5. Identifiers in exwrap.[ch] have been renamed from having 
       ex_ prefixed to having exwrap_ prefixes.

New in 1.12

    1. COOL! New module for detecting side effects in C expressions.
    2. Serious bugfix in hash_init().  The computation of the initial hash
       mask was completely botched up. Historically this code has seen little
       testing because hashing over a user supplied table is not extendible. 
       Users of hash_create() are not affected.
    3. Tried to make computation of hash_val_t_bit more threadsafe. It should
       be okay if writes to int objects are atomic, and concurrent writes of
       the same int value to a given object are safe.
    4. Makefile renamed to Makefile.gcc. Makevile.vc added. The rename
       is retroactive to all prior releases.
    5. OPAQUE_DEBUG becomes KAZLIB_OPAQUE_DEBUG and TEST_MAIN becomes