amiga-clib2/library/changes

c.lib 1.212 (27.11.2016)

- Unused slabs which get recycled are no longer reinitialized from
  scratch if their chunk size matches what the allocator needed.
  If the chunk size matches, the list of available chunks is
  left unchanged, and just the various counters are reset.

- Added __get_slab_stats() function.

- Added support for global __slab_purge_threshold tuning variable.


c.lib 1.211 (23.11.2016)

- Added more consistency checking to the slab allocator, which is
  built if DEBUG is defined in "stdlib_slab.c".

- Memory allocations are no longer guaranteed to be aligned to
  64 bit word boundaries. In fact, this has not even worked
  reliably in the past 10 years.

- Memory allocation request sizes are now rounded to multiples of
  32 bit words (the size of an address pointer) instead to the
  size of a 64 bit word.

- Reduced the memory footprint of the memory allocation management
  data structures by reusing the most significant bit of the
  memory allocation size. This allows many more allocations to fit
  into the 32 byte chunk slabs, but limits the maximum memory
  allocation size to a little less than 2 GBytes.

- Added integer overflow checks to the memory management code.

- Reduced the memory management overhead further. This cuts an
  additional 8 bytes per allocation, unless neither the slab
  allocator nor memory pools are available. With this reduction
  the slab allocator is able to use 16 byte chunks, which cover
  memory allocation requests of 1..8 bytes.

- Fixed a bug caused by returning an allocation back to a slab
  which passed the wrong pointer.


c.lib 1.210 (22.11.2016)

- Added __get_slab_allocations() function which will report information
  about each memory allocation made by the slab allocator which does
  not come from a slab.

- If the first slab in the list of slabs which share the same chunk
  size has no more room, it means that all other slabs following
  it have no room either. This speeds up the test to find a slab with
  free space, which can now abort and directly proceed to allocate
  memory for a new slab.

- If an empty slab's decay count hits zero, it is moved to the front
  of the empty slab list to be reclaimed more quickly.

- Allocations made from the slab now carry a pointer back to the
  slab which they are a part of. This speeds up deallocation but
  has the downside of making the smallest usable slab chunk size
  64 bytes, which is double what used to be the minimum before.


c.lib 1.209 (21.11.2016)

- The maximum slab size is now 2^17 bytes (= 131072). If you request
  a slab size larger than this, you will get slab sizes of 131072
  bytes instead.

- Enabling the memory management debugging code no longer produces
  compiler errors.


c.lib 1.208 (19.11.2016)

- Updated <stdlib.h> with new functions and data structures for
  use with the slab allocator.

- Added __get_slab_usage() function which can be used to query
  the slab allocator memory usage at runtime.


c.lib 1.207 (18.11.2016)

- Added a slab allocator which replaces the use of memory pools or the
  plain AllocMem() operations, respectively. In order to activate the
  slab allocator, choose a slab size (e.g. 2048 bytes or 4096 bytes)
  and declare a global variable like this:

     ULONG __slab_max_size = 2048;

  Memory allocations smaller than the slab size will be made from
  "slabs", i.e. large chunks of memory of the given size. Larger
  allocations will be managed separately.


m.lib 1.206 (24.4.2015)

- The fscanf() family failed to parse and convert %f parameters correctly
  if the respective number did not begin with a digit, but a decimal
  point. Hence ".7" would not be processed, but "0.7" would.

c.lib 1.206 (24.4.2015)

- Reworked the __putc() and putc() macros to reference the 'c' input
  parameter only once, and to be free of side-effects when tinkering
  with the buffer position.

- isatty() had the __fd_lock() call in the wrong place, which could have
  led to cleanup problems later.

- The close action in the stdio, socket and termios hook code now
  also zaps the fd pointer itself after cleaning up the file descriptor
  table entry.

- Removed the remains of all the stack extension and stack overflow/underflow
  checking code. It never actually worked. The bit that does work is the stack
  usage measurement code, plus the bit that sets up the the custom stack
  according to local setting or by calling a query function.


c.lib 1.205 (21.8.2010)

- Added dlclose(), dlerror(), dlopen() and dlsym() functions, which are
  available only under OS4. There is a variant of dlopen() in libunix.a
  which will perform a path name conversion. Note that these functions
  will not work in the thread-safe variant of the library because it
  would be unwise to tinker with the currently running program's binary.

- Added support for ELF shared objects and libraries. This is implemented through
  constructor/destructor functions, which means that you can use this
  functionality even in Amiga Exec style shared libraries, with the proper
  library initialization code to invoke the constructor/destructor functions
  for you.

- Updated uname() to recognize AmigaOS 4.1.

- The translation from Unix to Amiga path names did not properly process
  multiple occurences of "/./" in the path name. Thanks go to Steven Solie
  for finding the issue.

- The detection of "/./" and "/../" patterns in Unix path names to be
  translated into Amiga path names did not test if it was overrunning
  the end of the string.

- If strcmp(), strncmp() and memcmp() detect a pair of different
  characters, then the function result must be calculated as if the
  characters were of type "unsigned char". This is a requirement
  according to the ISO 'C' (1994) standard. Thanks go to Georg Steger
  for finding the issue.

- The definitions for INT_MIN, INT_MAX and UINT_MAX in <limits.h> no
  longer use long integer types, as prompted by Steven Solie.


c.lib 1.204 (11.11.2008)

- The memory allocated by malloc() and friends is now of type MEMF_PRIVATE
  under OS4 and beyond. The AmigaOS 2.x/3.x compatible code will still
  use MEMF_ANY in the same situation, though. Other uses of MEMF_ANY have
  been replaced as well where MEMF_PRIVATE would have made better sense.

- I/O buffers allocated are now aligned according to the CPU cache line size,
  if the operating system can supply such detailed information.

- unsetenv() now returns a status value.

- Corrected the function prototype for wcspbrk().

- Added function prototypes for mbrtowc_l(), wcscoll_l(), wcscspn() and wcsrchr().


c.lib 1.203 (28.4.2008)

- Added 68k stubs to amiga_rexxvars.c for the OS4 build to use. The new code now
  works just about exactly like the amiga.lib RVI code used to do, which means
  that there is no artificial length limit for the number of characters a string
  retrieved may have, and the code is largely reentrant.

- To work around a bug in the Roadshow TCP/IP stack (since fixed), the waitselect()
  function now substitutes a 10 microsecond timeout for a zero length timeout.

- isatty() no longer crashes if passed a socket rather than a file.

- Rewrote the GetRexxVar()/SetRexxVar() functions (mostly) in 'C', using available
  code as a reference. The 68k stubs currently only work for the 68k version of
  the library, and a solution for the OS4 build still needs to be found. The code
  is currently untested, but it should be complete.

- The printf() family now ignores argument specifications, as in "%2$d %1$d",
  which are used for localization on some platforms. This is a non-standard
  feature and the way clib2 treats it for now is just intended to avoid
  trouble while performing the conversion.

- [tboeckel]: when compiling amiga_rexxvars.c for m68k/OS3 the GetRexxVarFromMsg()
  and SetRexxVarFromMsg() can no longer cause linkage errors if the header files
  predate the SDK for OS 3.5.

- [jlangner]: the log() and log10() functions always returned -inf() even if the
  arguments was within the valid range for a log() function. Using DBL_EPSILON as
  the threshold was not correct as EPSILON is 2.2204460492503131E-16 whereas
  values like 1E-200 are still valid double values for a log().


c.lib 1.202 (16.1.2007)

- Added llrint() function contributed by Henning Nielsen Lund. Thank you
  very much!

- <unistd.h> now also include <stdio.h>, so that the SEEK_SET, etc. macros
  are defined for lseek() to use.

- Added a wrapper function which handles the thread-safe stdio stream
  resolution.

- In tcflush() a break signal can no longer cause the read flush loop
  to be quit with two semaphores still locked.

- In __obtain_daemon_message() the test to verify if the bsdsocket.library API
  would support the server API functionality checked the wrong feature. Fixed.

- Switched over the fd->fd_DefaultFile references to fd->fd_Socket where
  sockets are used rather than file handles.

- Added functions which modify the callback function and the userdata pointer
  stored in a low level unbuffered file/socket data structure. These function
  perform the proper locking and are thus safe to use in a thread-safe environment.

- The low level unbuffered file/socket now has a public equivalent, which
  is defined (along with the typedefs and flags) in <sys/clib2_io.h>. Functions
  for tinkering with it are still to come.

- The math kernel code no longer uses its own private scalbn() function.

- Added a function prototype for the _exit() function. Note that _exit() is
  not an ISO 'C' function.

- Corrected the getopt() function prototype, as prompted by Henning Nielsen Lund.

- The printf() family no longer adds a 0 or 0x prefix if the alternate
  conversion modifier is present for the %o and %x conversions and the
  value to be converted is 0 already. Put another way, printf("%#x %#o",0,0);
  now prints "0 0". This required another change so that %p always includes
  the 0x prefix even if the pointer involved is a NULL pointer.

- readlink() no longer sort-of-works for files and directories. It now only
  works for soft linked objects and returns an error for everything else.
  This is based upon a fix by Peter Bengtsson. Thank you very much!

- Moved the lstat() local Lock() function into its own separate file.

- uname() now returns correct and robust information for OS version
  numbers > 36. This integrates a fix by Peter Bengtsson. Thank you
  very much!

- Moved the crtbegin.o/crtend.o files out of the link libraries. Moving
  them in was intended to work as a fix for the shared library build, but
  now it seems that this has to be done at the link stage through the
  GCC specs file...

- Integrated a fix for __rem_pio2() which affects sin(), tan() and cos(),
  contributed by Steven Solie. Thank you very much!

- The internal 'struct fd' file descriptor table entry data structure
  now has a user data field entry.

- Rearranged the contents of the 'struct fd' file descriptor table entry
  data structure in preparation for making it public. Also added a version
  field so that user code can handle changes to it gracefully. The default
  file is no longer a BCPL pointer to a file handle by default, but
  both a BPTR and a socket identifier, wrapped into a union.

- Added experimental tilde expansion in Unix path names. This still needs
  some more work.

- __get_default_file() called __fd_unlock() without having called
  __fd_lock() first. Ouch.

- Removed an unnecessary pair of __fd_lock()..__fd_unlock() calls from
  ttyname_r().

- The libunix.a unlink() function is now reentrant, or at least thread-safe.

- You can now make unlink() stop after a failed deletion attempt which
  failed because the object to be deleted was reported as being "in use".
  The libunix.a variant defaults to report the deletion to have succeeded
  under these circumstances and later tries to delete the files marked
  for deletion. See <dos.h> for a brief documentation of how to change
  the behaviour.

- basename() and dirname() can no longer return NULL. They truncate the
  resulting path name instead. This is done so because some code that
  calls basename() or dirname() does not check if the function's return
  value is NULL.

- The SetOwner() fall-back code for Kickstart 2.04 was passing the wrong
  parameters to the file system. The first (dp_Arg1) should have been
  zero. Ouch.

- basename() is not supposed to modify the string it is passed and should
  return a pointer to a string which can be modified. Now it does. Same
  thing for dirname().

- asctime_r() now returns NULL if the buffer is too short to hold even a single
  byte of data.

- ttyname() now calls ttyname_r(). Also, the libunix.a version of ttyname_r()
  will produce "/CONSOLE" rather than "CONSOLE:".


c.lib 1.201 (21.9.2006)

- If defined, the local environment variable "DISABLE_COMMANDLINE_WILDCARD_EXPANSION"
  will disable expansion of wildcard patterns passed on the command line.
  Note that if the variable is not set then the global variable
  '__expand_wildcard_args' will provide the defaults for the switch that
  controls whether the wildcard expansion takes place. And after the
  environment variable has been checked, the '__expand_wildcard_args_check'
  function pointer can still be used to override the switch.

- fstat() now works with "NIL:" and "/dev/null", respectively. Not that
  it returns much useful information, though.

- The _PC_MAX_INPUT query for file handles now returns the default
  buffer size.

- Integrated the new OS4 build makefile collection, as contributed by
  Steven Solie. Thank you very much!

- Added the missing strerror_r() function.

- fpathconf() should work with the stdio streams, even in the thread-safe
  library version, again.

- Updated m68k specs file in /documentation to contain an own __CLIB2__
  define so that existing m68k compilers also have this define. In addition,
  the common "-noixemul" option can now also be specified but will do a NOP
  so that Makefiles sharing multiple runtime lib setups work without
  a warning.

- Added a new callback function which can be used in programs which want
  to avoid that the command line wildcard expansion takes place.  

- __get_default_file() now dynamically fills in file handles for the
  stdin/stdout/stderr streams if it's part of the thread-safe library.

- fpathconf() now checks if the file descriptor is really referring to a file.

- The termios hook entry code could file descriptor's embedded file handle
  rather than what the thread safe library had dynamically bound to the
  stdin/stdout/stderr streams.

- execve() now finds commands in the current directory again, even if you
  omit the leading "./" path name.

- The execve() code that looks for the command/script file now begins by
  checking if the file name includes path separators. If it does not,
  then the search for the command begins with the resident command list;
  otherwise the local directories are checked. The new code also properly
  cleans up after itself (a FreeDeviceProc() was missing) and the "PROGDIR:"
  lock now always refers to the directory in which the command/script
  file is found.

- Added the missing fdim() and fdimf() functions to the build
  makefiles. The OS4 build makefile was missing lrint() and
  lrintf(). Ouch.

- Added the POSIX exec() family functions, based upon code contributed
  by Henning Nielsen Lund. Thank you very much!

- atoll() no longer sets 'errno' directly but now calls __set_errno()
  instead.

- Added pathconf(), fpathconf(), sysinfo(), ftime(), ulimit(), getrlimit()
  and setrlimit() as contributed by Peter Bengtsson. Thank you very much!

- Added the missing S_ISSOCKET() macro to <sys/stat.h>. Note that this
  is not actually a POSIX feature.

- Added fmin(), fminf(), fmax() and fmaxf() to the build makefiles.
  Somehow I must have forgotten about them :-(

- select() and waitselect() can now be called without any file descriptor
  sets to work with, as some software does which uses select() in place
  of sleep().

- The <sys/ioctl.h> header file now includes both <sys/select.h> and
  <sys/filio.h> for better compatibility with the TCP/IP stack header
  files.

- Added the global variable __expand_wildcard_args which can be used
  to disable wildcard pattern expansion of command line parameters when
  linked against "libunix.a". Note that this has no effect on the "regular"
  libc.a behaviour.


c.lib 1.200 (17.4.2006)

- The default break signal mask (SIGBREAKF_CTRL_C) is no longer
  hard-coded. You can override it at link time with a different
  variable value for __break_signal_mask. This may have to be
  augmented by an API for changing the value.

- Added a waitselect() function which works very much like the
  bsdsocket.library/WaitSelect() function.

- Added <sys/time.h> include to <sys/socket.h> due to latest "TimeVal" change
  in the OS4 SDK. Otherwise "struct timeval" will not be defined at the time
  the <net/if.h> of the netincludes will be included by <proto/bsdsocket.h>.

- The strftime() hook function had the locale and character parameters
  switched, which made 'setlocale(LC_ALL,""); strftime(..);' unusable.
  Fixed.

- fchown() and chown() now accept owner and group ID values of -1, which
  indicate that the respective information should not be changed.

- The OS4 library build now includes the crtbegin.o and crtend.o object
  files in the libc.a library, which solves a problem with the thread-safe
  shared library support code.


c.lib 1.199 (6.3.2006)

- In <stdio.h> MAXPATHLEN is now equivalent to PATH_MAX (from <limits.h>),
  as suggested by Henning Nielsen Lund.

- Added lockf() support, as supplied by Henning Nielsen Lund.
  Thank you very much!

- symlink() now also works for absolute and relative links, and is
  part of "libunix.a". This patch was supplied by Henning Nielsen Lund.
  Thank you very much!

- The __main() stub function expected by the 68k GCC build is no
  longer part of "stdlib_main.c". This helps the C++ support, since the
  __main symbol definition no longer clashes with the definition in
  "libgcc.a".

- The printf() family stripped trailing zeroes from the integer part
  of %g output. Fixed.

- Moved an allocation size roundup operation in realloc().


c.lib 1.198 (11.12.2005)

- Added <fenv.h> for C99, and the (yet unimplemented) functions
  feclearexcept(), fegetenv(), fegetexceptflag(), fegetround(),
  feholdexcept(), feraiseexcept(), fesetenv(), fesetexceptflag(),
  fetestexcept(), fetestround() and feupdateenv().

- Replaced the old pow() implementation. However, powf() may need to
  be changed to set a domain error.

- In libunix.a malloc(), calloc() and realloc() no longer treat a
  request to allocate 0 bytes as an error, returning NULL. They all
  return a pointer sized memory chunk (= four bytes) initialized to
  NULL (= 0) instead.

- The alloca() implementation which allocates memory from the system
  rather than the local stack frame is thread-safe now. It also
  interacts with the realloc(), calloc(), free() and malloc() functions
  in that the alloca() cleanup routine is called once alloca() has
  done its job. If all the memory allocated through alloca() has been
  released no further calls to the cleanup function will be made.

- In the thread-safe library, realloc() permitted two different overlapping
  calls to succeed in trying to reallocate the same chunk of memory due to
  a race condition. Fixed.

- Added a new function __get_mem_stats() (the prototype is in <dos.h>) which
  can be used to query the current and maximum memory usage, counting the
  allocations performed through malloc(), free() and all other functions which
  use them.

- Added another function called __reset_max_mem_stats() which will reset the
  counters for "maximum amount of memory used" and "maximum number of chunks
  allocated" to the current figures for these values.

- Fixed the alloca() declaration in <stdlib.h> so that software which keys
  off the fact whether or not the alloca preprocessor symbol is defined
  will do the right thing.

- Added an optional call-back function which can be called if alloca()
  is about to return NULL, which some software does not consider.
  Rather than letting such software drop into an illegal memory access
  or worse, that call-back function will be invoked instead, which can
  print an error message and eventually call abort().


c.lib 1.197 (4.11.2005)

- Updated math_hypot.c to a newer (e_hypot.c 1.3 95/01/18) version from
  fdlibm which uses macros for manipulating the high and low words of a
  double, like the rest of fdlibm functions in clib2. The previous version
  would give bogus results when compiled with -O3 in clib2 which lead to
  "interesting" results (and lots of fun while searching for the problem)
  in Ghostscript. Wish I managed to track this down yesterday for 1.196
  release... <aantonijevic>

- Implemented atanh() and atanhf() which were not listed in the TODO
  file but were still unimplemented up until now.

- Replaced ldexp() and modf().

- Added __set_h_errno() and __get_h_errno() functions.

- The thread-safe library now tries to enable bsdsocket.library base
  sharing and attempts to make the TCP/IP stack call the local
  __set_errno() and __set_h_errno(), so that any Process calling
  the library functions will get a chance to see proper error
  codes. Whether any of these features could be enabled can be tested
  through two global variables __can_share_socket_library_base and
  __thread_safe_errno_h_errno which are defined in <dos.h>. Note that
  for both features to work you will need the Roadshow bsdsocket.library
  version 4.275 or higher.

- In <stdbool.h> the preprocessor definition should read
  "__bool_true_false_are_defined" rather than "__bool_true_and_false_are_defined".
  Thanks go to Peter Bengtsson for finding and fixing it.

- The network startup code no longer checks for the presence of a possible
  daemon startup message by default. You now have to enable this feature
  by declaring a global variable called "__check_daemon_startup" which is
  described in <dos.h>.

- Moved the code which rebinds the standard I/O streams to the server
  socket into a separate function which can be overridden by user code.

- Updated the new __obtain_daemon_message() function to call a
  bsdsocket.library API function to determine if what appears to be
  a valid daemon startup message is sound.

- Replaced the OS4 specific build makefile with an updated version
  prepared by Steven Solie. Thank you very much!


c.lib 1.196 (11.10.2005)

- Removed the various workarounds associated with <unistd.h>, required
  for building code that references the networking API. It is now possible,
  provided you build your code with an existing TCP/IP API header file
  set (such as from the "Roadshow" SDK) without running into data type
  or function prototype declaration conflicts. I have added local versions
  of <arpa/inet.h>, <netinet/in.h>, <netdb.h>, <sys/filio.h>, <sys/ioccom.h>,
  <sys/ioctl.h>, <sys/select.h> and <sys/socket.h> and updated <unistd.h>,
  <pwd.h> and <grp.h> accordingly. Note that some of the function prototypes
  will conflict with the bsdsocket.library/usergroup.library API, which is
  why you can disable the declarations through the __NO_NET_API preprocessor
  symbol: if defined, none of the function prototypes will be defined.

- Implemented lrintf(), lrint(), lroundf(), lround(), nearbyintf(),
  nearbyint(), remquof(), remquo(), roundf(), round(), tgammaf(),
  tgamma(), truncf(), trunc(). Sort of implemented fmaf() and fma(),
  which really ought to be done in "SIMD" fashion.

  This completes the "real" floating point math library (ignoring
  for a moment that the floating point environment code is still
  not implemented).

- accept() now calls the bsdsocket.library accept() function first
  and then hooks up the socket with the clib2 data structures. This
  makes it possible to have several Processes calling the accept()
  stub at the same time.

- When the main() function returns directly or indirectly by way of
  calling exit(), the IoErr() value set at this time will now be
  preserved until the program really returns control to the shell.
  This turned out to be easier to implement than I had originally
  thought, thanks to the fact that the atexit() functions are all
  invoked straight after the exit() call.

- strtoimax(), strtoumax(), strtod(), strtof(), strtol(), strtoll(),
  strtoul() and strtoull() now count how many digits were converted.
  If that number is zero, then the conversion is considered to have
  failed: 0 is returned and the 'ptr' parameter reference is initialized
  with a pointer to the string to be converted. This has two consequences:

     1) A floating point number that has no digits in the significand
        is no longer converted to zero. For example ".e10" can no longer
        result in a successful conversion.

     2) You can detect if a conversion is successful by comparing pointers
        after the conversion function has returned:

           char * str;
           char * ptr;
           long n;

           n = strtol(str,&ptr,10);
           if(n == 0 && str == ptr)
              printf("conversion has failed");

- Following detection of a stack overflow, the startup code eventually
  called _exit(). It should have called exit() instead.

- tmpnam() wrote more than L_tmpnam bytes to the name buffer. Also, the
  TMP_MAX value was off by one.


c.lib 1.195 (3.9.2005)

- In __time_delay() the CheckIO() test was wrong and should have
  tested for a request still in progress. Fixed.

- Modified select() to support plain files, too. The new code now
  compares the current file position against the current file size,
  if that file is on a file system, or simply checks if there is
  any data in the file (which works for the default PIPE: device).

- The raw file descriptor structure member fd_Position is now
  always updated, regardless of whether the file supports random
  access or not.

- The wildcard escape character used in __wildcard_expand_init() was
  wrong. It should have been "'" but it was "`". Fixed.

- The startup code now references the linker symbol generated for
  the user-supplied main() function.

- log(+/-0), logb(+/-0), log10(+/-0) now return -infinity.

- getcwd() now considers a buffer size of 0 an error, and it sets
  the errno code to ERANGE if the buffer is too small.

- With Unix path semantics enabled, rename() now fails if either
  the old or the new name are empty strings. Same thing for
  opendir(), utime(), unlink(), chown(), link(), readlink(),
  realpath(), chdir(), access(), stat(), rmdir(), lstat(),
  chmod(), statfs() and open()/fopen().

- Fixed several issues with the formatted output of strftime()
  for the "C" locale: "%p" could return " PM", "%c" is now
  equivalent to "%a %b %e %T %Y", "%x" is equivalent to "%m/%d/%y"
  and "%X" is now equivalent to "%T".

- The 68020/030/040/060 versions of the library now include the
  64 bit integer conversion code. The plain 68000 version does not.

- The 64 bit integer conversion code used in by the scanf() family
  was broken due to a misplaced overflow check, which had the cruel
  effect of triggering for 32 bit integers as well. Fixed.

- DoTimer() did not allocate the local MsgPort correctly. Fixed.


c.lib 1.194 (15.7.2005)

- Added getopt_long to libunix.a <tfrieden>

- Added new libprofile.a which implements gprof gmon.out output <tfrieden>

- Added the function profil() to libc.a <tfrieden>

- Added ftello() and fseeko() which both merely call the equivalent (at
  least for clib2) ftell() and fseek() functions, respectively.

- statfs() never stated that a disk was mounted for read-only access
  because the wrong InfoData structure member was tested. Fixed.

- lstat would overwrite the name parameter via ReadLink(..., name, ...)
  instead of ReadLink(..., new_name, ...) <tfrieden>

- Implemented exp2()/exp2f() and log2()/log2f() as suggested by
  Henning Nielsen Lund. Thank you very much!

- The pattern matching code which expands command line arguments, as part
  of libunix.a, now translates the "*" wildcard into "#?" unless you prefix
  it with a backtick ("'"), which is the wildcard pattern escape character
  used on AmigaOS.

- Repaired the pattern matching code which expands command line arguments:
  it no longer requires that any parameters are quoted and it is automatically
  enabled if you link against libunix.a, without libc.a standing a chance to
  accidentally override it.

- Fixed a bug in vfprintf that would surface when the buffer was enlarged
  if the pattern was too large to fit the static internal buffer <tfrieden>

- Added file system names to statfs <tfrieden>

- Moved all the constructor code out of "stdlib_init_exit.c" and into the
  files which initialize global data, such as the new "stdlib_program_name.c"
  and "stdlib_malloc.c".

- Until I can find a way to invoke it from within the library, the
  __machine_test() function is no longer invoked.

- Tagged global data with NOCOMMON attributes.

- Added the __lib_init() and __lib_exit() functions, which are part of the
  thread-safe library and which can be used to hook up clib2 with standard
  Amiga shared library/device code. Some documentation on how to use them
  can be found in the <dos.h> header file.

- Added a complete shared library skeleton example to demonstrate how the
  clib2 shared library initialization/cleanup functions should be used.
  This is a dual-build library which will both work with the plain 68k
  AmigaOS and the PowerPC native AmigaOS4. The example source code can be
  found in the "skeleton_library" subdirectory.

- The thread-safe version of isatty() should now work for stdio
  file descriptors, too.

- Retrofitted thread-safety into the termios code.

- The socket/usergroup API function and data type definitions that used to
  be defined in <unistd.h> have been moved into a separate header file
  <sys/clib2_net.h> which, for backwards compatibility, can be include
  from within <unistd.h> if the preprocessor symbol __USE_CLIB2_NETLIB
  is defined early on. Eventually, <sys/clib2_net.h> should be removed
  from the general header files and become a local library build header
  file altogether. Note that in this case you would have to build your
  own software using common TCP/IP SDK header files.

- <sys/types.h> now also includes <time.h> and <stddef.h>.


c.lib 1.193 (4.6.2005)

- fpclassify() now returns FP_ZERO both for 0 and -0.

- nan() and nanf() now return quiet NaNs.

- Added internal __inf() and __inff() functions.

- strtof() now calls nanf() and __inff(), respectively, to produce
  the special floating point values for nan/inf/infinity.

- strtod() now calls nan() and __inf(), respectively, to produce
  the special floating point values for nan/inf/infinity.

- The scanf() family now calls nan() and __inf(), respectively, to
  produce the special floating point values for nan/inf/infinity.

- Added %D, %e, %F, %g, %G and %h for strftime(); repaired %I.

- Documented __strip_double_slash() and plugged in a memmove()
  in place of the copying loop.

- Modified __translate_unix_to_amiga_path_name() and
  __translate_amiga_to_unix_path_name() to not to call strlen()
  on the results of __strip_double_slash().

- For %C strtime() should return the century number, not the
  year number mod 100.

- The record locking code in fcntl() is no longer part of the
  regular libc.a, but only of libunix.a.

- Changed the definition of the D_S() macro to cast the pointer
  address to an unsigned 32 bit integer.

- Modified the locale-aware isalnum(), isalpha(), iscntrl(),
  isdigit(), isgraph(), islower(), isprint(), ispunct(), isspace(),
  isupper(), isxdigit(), tolower() and toupper() functions
  to clamp the input parameter to the "unsigned char" range before
  it is submitted to the respective locale.library function. This
  should be in sync with what the specs demand, which state that
  if the input parameter is neither EOF nor in the range of an
  "unsigned char" variable, then the results are undefined.

- ungetc() now returns the input character, clamped to an unsigned
  char value, upon success. Previously, if the input parameter was
  negative and not EOF, the result was identical to the input, which
  could have had negative side-effects.

-<2D>Added NAN, INFINITY, FLT_EVAL_METHOD, float_t and double_t to
  <math.h> (C99).

- Added skeleton code for acosf(), acosh(), acoshf(), asinf(), asinh(),
  asinhf(), atan2f(), atanf(), atanh(), atanhf(), cbrt(), cbrtf(),
  ceilf(), cosf(), coshf(), erf(), erfc(), erfcf(), erff(),
  exp2(), exp2f(), expf(), expm1(), expm1f(), fdim(), fdimf(),
  floorf(), fma(), fmaf(), fmax(), fmaxf(), fmin(), fminf(),
  fmodf(), frexpf(), hypotf(), ilogb(), ilogbf(), ldexpf(), lgamma(),
  lgammaf(), log10f(), log1p(), log1pf(), log2(), log2f(), logbf(),
  logf(), lrint(), lrintf(), lround(), lroundf(), modff(), nearbyint(),
  nearbyintf(), powf(), remainder(), remainderf(), remquo(), remquof(), round(),
  roundf(), scalbn(), scalbnf(), sinf(), sinhf(), sqrtf(), tanf(),
  tanhf(), tgamma(), tgammaf(), trunc() and truncf(), to be filled in
  later...

- Implemented fmin()/fminf(), fmax()/fmaxf(), fdim()/fdimf() for C99.

- Ported acosf(), asinf(), atan2f(), atanf(), ceilf(), expf(), floorf(),
  fmodf(), frexpf(), ldexpf(), log10f(), logbf(), logf(), modff(), powf(), sqrtf(),
  scalbn() and scalbnf() for C99.

- Ported cbrt(), cbrtf(), erf(), erff(), erfc(), erfcf(), expm1(),
  expm1f(), ilogb(), ilogbf(), log1p() and log1pf() for C99.

- Ported cosf(), coshf(), sinf(), sinhf(), tanf(), tanhf()
  and hypotf() for C99.

- Ported acosh(), acoshf(), asinh(), asinhf(), lgamma(), lgammaf(),
  remainder() and remainderf() for C99.

- The scanf() family now supports character ranges for the %[
  conversion. Note that this is a non-standard feature!

- Integrated Peter Bengtsson's termios code. Thank you very much! I chose
  to add it to libunix.a rather than keeping it in a separate libtermios.a
  library.


c.lib 1.192 (12.5.2005)

- Changed how errors are detected, as returned by Write(), Read() and
  Seek(). Seek() is particularly challenging because the value it
  returns might be a valid file position and not an error.

- Replaced numeric function return codes of 0 and -1 with macros OK,
  SEEK_ERROR/ERROR to clarify the respective purposes.

- Changed how ftell() and fseek() are used, double-checking the return
  value and the errno code.

- The record locking semaphore could wind up getting added to the public
  list twice. Fixed.

- Fixed two linker errors which were caused by duplicate symbol definitions.

- Added code to the startup routine which allows you to monitor where a command
  was started from and which parameters it was invoked with.

- If fread()/fwrite() fail to read/write any data because either the number
  of records or the size of each record is zero, both now call clearerr() to
  avoid giving the caller the wrong impression that an EOF or error occured.

- The libunix.a flavour of system() no longer attempts to translate the name
  of a command unless it contains path separator characters.

- Added strtof(), llabs(), lldiv(), vsscanf() and vscanf() for C99.

- strftime() now supports %C, %n, %r, %R, %t, %T, %u, %V, and %z for C99.
  And it ignores the E and O modifiers.

- The printf() family now supports the %hh, %j, %t and %z modifiers and the
  %a/%A conversions for C99. The %j is treated like %ll; %t and %z are treated
  like %l. Also, the "infinity"/"not a number" signals now come out as the
  strings "inf" and "nan".

- Added HUGE_VALF to <math.h>.

- For the printf() "%a" conversion the exponent now comes out as a binary
  number rather than a decimal one. Now how odd is that?

- strtod() and strtof() now support "inf"/"infinity"/"nan"/"nan(..)" and
  hexadecimal floating point numbers, for C99.

- Added the fpclassify(), isfinite(), isnormal() and signbit() macros for C99.

- Reimplemented isnan() and isinf() as macros for C99. The corresponding
  functions will be dropped from the library. Note that the isinf() macro
  does not return -1, 0 or 1 like the old function did, but only 0 or 1
  depending upon whether the parameter represents +/- infinity or not.

- Added fabsf() for C99.

- The scanf() family now supports the %hh, %j, %t and %z modifiers and the
  %a/%A conversions for C99. The %j is treated like %ll; %t and %z are treated
  like %l. Also, the "inf"/"infinity"/"nan"/"nan()" keywords are processed.

- The strftime() %z conversion now prints the time zone difference as a
  "decimal" number. That is, if the difference is 5 hours and 30 minutes,
  then %z will now print "530" rather than "330".

- mktime() now handles one leap second gracefully.

- Added isblank().

- Added isunordered(), isgreater(), isgreaterequal(), isless(),
  islessequal() and islessgreater() to <math.h> for C99.

- The wchar_t type is now an 'unsigned short' integer (16 bits wide).

- Added PTRDIFF_MIN/PTRDIFF_MAX, WCHAR_MIN/WCHAR_MAX and
  WINT_MIN/WINT_MAX to <stdint.h> for C99.

- Added imaxdiv() and imaxabs() for C99.

- Added strtoimax() and strtoumax() for C99.

- Added nextafter() and nextafterf() for C99.

- Added copysign() and copysignf() for C99.

- Unless I missed something, clib2 should now be functionally complete
  with regard to C99, except for the floating point operations covered.
  These are a major challenge all by themselves, and I wonder both
  whether they are worth the effort and how one could implement them
  correctly.

- fflush() now consistently supports a NULL parameter, causing all
  streams to be flushed for which this behaviour is defined.

- The printf() family can now produce output for floating point
  numbers with more than about 77 characters, provided sufficient
  memory is available. C99 calls for a minimum of 4095 characters,
  but we're trying to allocate the space required dynamically.


c.lib 1.191 (9.4.2005)

- The name of the public record locking semaphore has to be preallocated
  for OS4 if the AllocSysObject() function is used to create it.

- Moved the signal block mask and the signal function table out of
  signal_data.c and into signal_raise.c where they are actually used.
  This makes it possible to override the default definition of the
  __check_abort_enabled variable in your own programs.

- raise() no longer resets the signal handler before it invokes the one
  currently configured. It merely blocks the delivery of the respective
  signal to prevent recursion.

- raise() now drops into abort() if a signal handler is set to
  SIG_DFL. The exception is in SIGINT delivery, which has the
  effect of printing a different termination message but otherwise
  program flow takes the same path as abort().

- Moved the __UtilityBase/__IUtility variable declarations into
  a separate file.

- Simplified the library/open close code in "stdlib_main.c".

- File descriptors produced by dup() or dup2() now work exactly like
  the original file descriptors they are duplicates of. I modified the
  function which maps file descriptor numbers to file descriptor
  table entries to return the table entries of the original files.

- In the thread-safe library, duplicated stdin/stdout/stderr
  descriptors now work like the "real" ones. This is true even if
  the "real" ones were closed and only their former aliases remain.

- Invoking fstat() on what maps to a con-handler stream now produces
  information identifying it as a character special file.

- Added more code and changes contributed by Peter Bengtsson, thank you
  very much! This includes the following:

  - Added SSIZE_MAX to <limits.h>.

  - Added <sys/uio.h>, readv() and writev().

  - Cut back the soft link resolution code in lstat().

  - In <fcntl.h> O_NDELAY is now an alias for O_NONBLOCK.

  - Added <complex.h>, carg(), cargf(), cargl(), cimag(), cimagf(),
    cimagl(), conj(), conjf(), conjl(), creal(), crealf() and creall().

  Note that the C99 support for the complex floating point data
  types is limited to GCC 3.x for now.

- Added va_copy() to <stdarg.h>.

- Added _Exit() to <stdlib.h>.

- Added <stdbool.h>.

- Added vfscanf() to <stdio.h>.

- The stdio locking in fcntl() wasn't working correctly. Fixed.

- Made the clearerr(), feof() and ferror() macros thread-safe.


c.lib 1.190 (25.3.2005)

- DoTimer() now calls AllocSysObject() rather than making up
  a MsgPort locally.

- The record locking semaphore code now builds a semaphore to add
  before it tries to find the public one in memory. That way, the
  code can spend less time in Forbid() state and, heaven forbid,
  refrain from allocating memory while in that state.

- Split the general stdio initialization/cleanup code from the
  initialization of the stdin/stdout/stderr streams.

- Moved the Workbench console stream initialization into the
  initialization code for the stdin/stdout/stderr streams and
  out of the program parameter setup.

- The current program name is now set up in the stdlib
  initialization function.

- Simplified the machine test code; moved the FPU check into
  the math initialization code.

- Added more safety checks to verify that file descriptor
  file handles are valid.

- Made the file descriptor checks in the fsync() and fdatasync()
  functions more robust.

- Cleaned up the 68k build makefile, so that the CPU and FPU
  tests and the error message display can run safely even
  on plain 68000 machines. This won't work for the 32 bit small
  data model, which implies 68020 code, but so there...

- Moved the CPU/FPU type tests into the respective linker
  libraries.

- Moved the data declarations out of math_data.c, stat_data.c,
  socket_data.c, dirent_data.c and stdio_data.c and into the
  code that initializes them.

- Moved a few __delete_semaphore() calls into conditional compilation
  sections where they should have been in the first place.

- Thanks to J<>rg Strohmayer, the GCC library build now manages to
  invoke the library's constructor/destructor functions in a
  very particular order. That way, you can use constructor/destructor
  functions in your own code and not have them clash with the library's
  own functions.

- Reimplemented the constructor/destructor invocation code for GCC.
  Both the 68k and PowerPC platform now invoke them in the same order
  and the 68k code uses the designated invocation priorities. The
  PowerPC destructor function now sets up the exit() jmp_buf before
  the destructor functions are called.

- Added S_IREAD, S_IWRITE and S_IEXEC aliases to <sys/stat.h>.

- Moved data out of stdlib_data.c and into the code that references
  or initializes it.

- The stdlib constructor now performs the CPU/FPU compatibility test.

- Introduced new constructor types and changed the overall priority
  order.

- Switched over the startup code and the library itself to use
  constructor/destructor functions for initialization and cleanup
  purposes.

- The destructor function invocation code no longer calls
  setjmp(). This is now done within stdlib_main.c prior to
  calling the destructor function invocation code.

- Fixed the SAS/C destructor function prototypes and verified
  that the constructors/destructors are called in the proper
  order, too.

- Aliases of file descriptors are now using the signal semaphore
  of the original file descriptor.

- close() did not return 0 if the file descriptor in question
  was really just an alias. Fixed.

- Added a feature which makes it possible to have several clients
  use the standard I/O streams (stdin/stdout/stderr) and have these
  referring to their process' Input()/Output()/ErrorOutput()
  streams. This is intended to support the upcoming shared
  library feature.

- Turns out that the 68k GCC port does not sort constructor and
  destructor functions in any way at all. I reimplemented the
  entire library constructor/destructor functionality to use the
  same approach as libnix.

- ftruncate() ended up changing the current file position, contrary
  to what it is supposed to do. Fixed.

- fcntl() did not work for sockets. Fixed.

- Reorganized the local header files, removing redundancies.

- Dropped unused stub code which is now redundant because of
  the constructor/destructor mechanism.

- Moved the call chain printing out of stdlib_main.c and into
  separate files.

- Removed some more redundant data from stdlib_main.c.

- Added the first "real" C99 function: _Exit() ;-)

- assertion failures early on during program initialization
  should no longer spell big trouble on account of the stdio
  data structures possibly not being in a well-defined and
  initialized state.

- Turns out that the constructor/destructor calling sequence
  was still wrongish for the OS4 library. So I had to change
  it *again*. Which probably means that the 68k library will
  need further changes...

- Moved stdlib_main.o into the regular libc.a, at least for
  the 68k build. The PowerPC build may follow later, provided
  I manage to get the specs file fixed. Actually, stdlib_main.o
  is in the libc.a library already. Now about that specs file...

- Moved the check for the presence of an FPU into the
  math_init.c code. I am far from certain whether this will
  have the desired effect, though. Due to how the GNU ld linker
  works, libraries are scanned once only. And the FPU check will
  be pulled in only if something references the HUGE_VAL
  constant.

- Activated the dormant thread-safe standard input/output/error
  handling code.

- Small fixes to fcntl() and select() to cover the thread-safe
  fd->fd_DefaultFile == ZERO case.


c.lib 1.189 (5.3.2005)

- Rewrote the __translate_unix_to_amiga_path_name() function to
  translate patterns such as "foo/bar/../../baz" properly, and to
  use strlen() a lot less.

- Major, major changes! Moved most of the monolithic code out of
  the file descriptor hook and into the respective functions,
  such as dup2(), fchmod(), fchown(), fcntl(), fdatasync(), fstatfs(),
  fsync(), ftruncate() and lseek(). Code which is not strictly
  required will no longer find its way into your programs if you
  link with the updated library.

  NOTE: these changes require that the entire library is rebuilt!

- The buffered and unbuffered file hook code is now invoked through
  function pointers alone. The utility.library/CallHookPkt mechanism
  is no longer required.

- Moved the entire lseek() code relevant for files into the hook
  function.

- Simplified the close() function which now just calls into the
  hook code to perform whatever is necessary. The hook code is
  responsible for cleaning up after aliases, etc. This change in
  turn made it possible to greatly simplify the hook code for
  buffered files which now bypasses close/read/write/lseek and
  directly invokes the hook code for unbuffered files.

- Added various floating point constants to <math.h>, courtesy
  of Henning Nielsen Lund. Thank you very much!

- When using the wildcard expansion code for command line
  parameters (which is by default linked in with libunix.a),
  regular expressions can no longer prompt dos.library requesters
  to appear. However, to be on the safe side, if you are expecting
  to pass regular expressions on the command line, do not use
  the wildcard expansion code such as by overriding the library
  symbols with dummy functions such as are used in the file
  "stdlib_wildcard_expand.c".

- Added a new variable '__open_locale'<27>which can be used to
  restrict all library functions to use the "C" language locale
  rather than the current system locale settings. In addition
  to that, two new functions __locale_exit() and __locale_init()
  can be used to close and (re-)open the system locale at a
  later time.

- Local ("static") functions are now identified by the STATIC
  qualifier. This was done in preparation for changes that will
  deal with global and local data and the issue of thread safety.

- Added stdio thread locking functions flockfile(), funlockfile(),
  and ftrylockfile().

- Modified the internal FILE structure to allow for thread locking.
  Note that this again requires that the library is rebuilt!

- Added or modified macros for getc_unlocked(), getchar_unlocked(),
  putc_unlocked() and putchar_unlocked().

- Added rand_r().

- Added flockfile()/funlockfile() wrappers around all stdio
  functions.

- Added more semaphore locking around the basic stdio, memory, locale
  and dirent data operations. That should do it! While the library is
  not reentrant (this is not ixemul.library) it should be thread-safe
  now. Thread-safe in the sense of POSIX 1003.1c-1995.

- The thread-safety code is now subject to conditional compilation.
  Both the library and the user code need to be rebuilt with the
  preprocessor symbol __THREAD_SAFE defined to get thread-safe
  code.

- Extended the thread-safety locking to the file descriptors.

  NOTE: these changes require that the entire library is rebuilt!

- The translation of Unix to Amiga path names now silently accepts
  absolute Amiga path names passed to it and will use them without
  changing them.

- Added atoll(), ffs(), ftw(), nftw(), lstat() and uname() code
  contributed by Peter Bengtsson. Thank you very much!

- Reworked the code that handles quoting for the wildcard expansion
  routine. We no longer allocate memory and then modify it, but
  call a function for each quoted parameter which does whatever is
  necessary.

- The shell command parameter parser now considers the non-breaking
  space character (ISO code 160) to be a blank space character, too.

- Moved the signal semaphore allocation/initialization/deallocation
  into a dedicated module. This also has the advantage that it's
  harder to break code by accidentally forgetting to call
  InitSemaphore() after having allocated the memory for it.

- Rewrote the code that allocates the file descriptor and file
  buffer tables so that all the memory allocations are in one
  place and it's possible to specify exactly how many table
  entries are required at a time.

- Creation and initialization of semaphores now uses the AmigaOS4
  specific functions for this purpose, if available.

- In the thread-safe variant, the library now tries to allow
  multiple concurrent callers to use the socket functions. Note
  that this works only with the Roadshow TCP/IP stack, and the
  results with other TCP/IP stacks are rather unpredictable.


c.lib 1.188 (7.2.2005)

- Folded duplicate code in "stdio_init_exit.c" into a common function.

- Simplified the code in "time_asctime_r.c" which builds the time
  string. It gracefully handles buffer sizes which are too short
  by returning an empty string.

- Moved the "tm->tm_wday" initialization out of the hook function
  in "time_strftime.c" since it was to be called only once anyway.

- Lost a few compiler warnings in "unistd_time_delay.c" and
  "time_gettimeofday.c".

- Folded duplicate code in "time_mktime.c"; also, errno is no longer
  modified unless the library is built with the "CHECK_FOR_NULL_POINTERS"
  option.

- We now allocate the AnchorPath used in the unistd_wildcard_expand.c
  code. Also, the contents of the AnchorPath structure are no longer
  modified between calls. MatchEnd() has to be sufficient.

- Moved redundant code out of the readdir()/opendir()/closedir()
  functions which is not required unless the code is built for
  Unix compatibility mode.

- Lost the __not_a_number and __infinity variables, including the
  code which initialized them.

- Reading/changing the errno variable is no longer done directly, but
  involves accessor functions.

- References to the HUGE_VAL quantity now involve an accessor function, too.

- Changed the manner in which the __huge_val constant is initialized by
  the __math_init() function. The new approach should be more portable.

- The library no longer sends ACTION_DISK_INFO packets to the console
  handler. The side-effects were too varied and irritating after all.

- Added the fsync() and fdatasync() functions and the <stdint.h> and
  <inttypes.h> header files contributed by Peter Bengtsson. Thank
  you very much!

- Tweaked the build makefiles to produce fewer meaningless warnings.


c.lib 1.187 (29.1.2005)

- The default console output window opened when a program is launched
  from Workbench would open and stay open. This was not intended to
  happen and is a side-effect of the new stdio initialization code which
  checks if the stdio streams are in non-blocking mode. Fixed.

- Moved the common DateStamp to time_t conversion code into a shared
  function.

- The fall-back function for converting time into a string in strftime()
  now calls itself for the "%c", "%x" and "%X" format specifiers.

- mktime() is supposed to convert the time specification, given as local
  time, into the number of seconds since January 1st, 1970, relative to
  UTC. This didn't really work up until now since the time value returned
  was given as local time.

- Plugged in a different algorithm for calculating the day of the week in
  strftime() and asctime_r(). This one isn't sensitive to the effects of
  adding/subtracting the local time zone.

- Changed the algorithm that calculates the number of days that have passed
  so far as used by the the __convert_time() function and the conversion
  code in strftime().

- Also changed the algorithm used by strftime() to produce the week numbers
  (the '%U' and '%W' format specifiers). The new method is much simpler
  than the old one.

- Made the code that converts a 'time_t' value into the 'struct DateStamp',
  as used by strftime() and utime(), into its own function. This also
  fixes a bug in the code strftime() would use which was completely
  unaware of the local time zone settings.


c.lib 1.186 (14.1.2005)

- Redirecting stderr to a file no longer has the effect of showing error
  messages and assertion failure notifications as requesters. The exception
  is in redirecting stderr to NIL: which will prompt the requester use.

- gettimeofday() now calls GetSysTime() rather than DateStamp() to obtain
  the current system time. This resolves granularity issues since the
  DateStamp() result was only accurate by 1/50 of a second.

- The "ptrdiff_t" definition in <stddef.h> now defaults to type 'int' rather
  than 'long int'.

- The "char" limits in <limits.h> are now set up according to the current
  compiler settings, which can either default to an unsigned or
  signed definition.

- Changed the rules again for the use of stderr/stdout redirection when
  printing error messages. It is always safe to redirect them now and
  no requester will appear unless you specifically set the value of the
  "__no_standard_io" variable to TRUE in your program.

- Removed a misplaced IsInteractive() from the stdio initialization
  function. Now this could have been big trouble...

- Removed tests for FileHandle->fh_Type != NULL which used to precede
  all IsInterative() tests. I verified that IsInteractive() will always
  return FALSE for NIL: type file handles.

- Dropped the special flag variable used by the abort() function that
  tracks whether or not console output is possible. We now use the
  global "__no_standard_io" instead.

- Made the <ctype.h> macros more robust.

- Removed the "NIL:" file handle tests preceding the Open("CONSOLE:",..)
  calls. As of Kickstart 2.x and beyond these are no longer a source of
  trouble.

- The V37/V40 compatibility code is no longer built for the AmigaOS4
  version of the library.

- Switched over the last use of DeviceProc() to GetDeviceProc(), etc.

- open() no longer examines a file after opening it in order to figure
  out whether read/write accesses are permitted. This decision is now
  for the file system to make.

- Whether or not stdio console streams are blocking or non-blocking
  is now determined at initialization time. The I/O mode is restored before
  the program exits. Previously, any changes to the I/O mode would persist.

- Lost some more code that is not required for AmigaOS 4.x and can be
  handled conveniently through conditional compilation.

- close() did not reset the non-blocking file property, as it should
  have. This only worked for files which were closed anyway, but not
  for the stdio streams. Fixed.

- Added a missing definition to stdio_init_exit.c which is part of
  the OS4 header files, but not of the older header file distributions.



c.lib 1.185 (2.1.2005)

- Moved the environment variable cleanup code into a destructor function.

- Fixed a typo in the hstrerror() function.

- The common error reporting function __show_error() could throw Enforcer
  hits if the program was not launched from Shell. Fixed.

- Moved the memory initialization and cleanup functions into the
  malloc/free code itself and updated the alloca code to do its
  own data management.

- Finally optimized the alloca() memory cleanup code.

- Tried to make it possible to have debug versions of the memory
  management code in the default link library along with the
  normal versions. Let's see how well this works out in the GCC
  version. SAS/C does not seem to like it...

- Changed the definition of alloca() for GCC, which now defaults
  to the built-in function.

- Updated the stdio.h, stdlib.h, string.h and unistd.h header files
  to declare function prototypes for the __MEM_DEBUG versions of the
  library functions only if that preprocessor symbol is defined.

- If a SIGINT signal is caught and processed by the default signal
  handler, the localized "*** BREAK" string will be printed rather
  than the built-in one.

- The data structure alignment (file I/O buffer) is now configurable
  at compile time. The default used to be 16 bytes, which is appropriate
  for the 68040/68060 but not for the PowerPC, which uses 32 or 128
  bytes per cache line.

- fwrite() now flushes the entire "buffer" for unbuffered files. The
  exception are "interactive" files such as console windows. For these
  line buffered output is used.

- Whether or not a file is bound to an interactive device, such as a
  console window, is now checked and remembered after a file descriptor
  has been associated with it.


c.lib 1.184 (28.11.2004)

- Added asctime_r(), ctime_r(), gmtime_r(), localtime_r() and strtok_r().

- Added stubs for the Rexx Variables Interface code that used to
  be part of amiga.lib. While comparable functionality is available
  in rexxsyslib.library V45, the new stubs might be helpful during
  porting. Care must be taken since these functions don't work
  exactly like the originals.

- Integrated strlcpy() and strlcat() which are intended to be safer
  replacements for strncpy() and strncat().

- The program's task priority is now configurable through an external
  variable '__priority'.

- The process name to be used when detaching can be configured through
  the new '__process_name' variable.

- The minimum required operating system version can be configured
  through the new '__minimum_os_lib_version' variable; a matching
  error message can be provided through the new '__minimum_os_lib_error'
  variable.

- The default console window specification can be overriden through
  the new '__stdio_window_specification' variable.

- The socket initialization code did not set up a reference to the
  'h_errno' variable correctly. This had the effect of making name
  and address resolution errors trash the 'errno' variable instead
  and leaving 'h_errno' always set to 0. Fixed.

- For sockets, ioctl() and fcntl() now interact on the FIONBIO/FIOASYNC
  requests (ioctl) and the O_NOBLOCK/O_ASYNC flags (fcntl).

- popen() now accepts "rb" and "wb" as mode parameters. However, "r+",
  "w+" and variants thereof are still unsupported due to the
  unidirectional pipe support in the standard "PIPE:" device.


c.lib 1.183 (13.11.2004)

- Cleaned up the OS4 build makefile, losing redundant libraries,
  adding more startup object code and ultimatively making the whole
  rebuild logic work again: if code changes and dependencies are
  set up correctly, it will now get rebuilt. Previously, such
  changes went unnoticed and you had to rebuild the entire library
  from scratch.

- Added stubs for CreatePort(), DeletePort(), CreateTask(), DeleteTask()
  and NewList() which have equivalents in exec.library V50 but for which
  it might be useful if ported code didn't have to reference these
  explicitly.

- mktemp() was broken in libunix.a with Unix path semantics enabled.
  This was because the name template was translated and translated
  back again, overwriting the translation buffer. This, funny enough,
  broke Samba's printing feature. Fixed by translating the name only
  before each test for "uniqueness" is made. The new code also handles
  empty "" templates gracefully, which was a problem with both the
  "standard" and the Unix path semantics flavour.

  Why is it that I find bugs like this always after having just
  released another library update?


c.lib 1.182 (8.11.2004)

- Changed the error abort condition for the %s conversion of the
  scanf() family. It now matches the abort conditions for all other
  conversions and no longer ignores whether any other parameters were
  converted before. This was a quirk in the older implementation.

- The scanf() family now accepts %E and %G in place of %f and %X in
  place of %x.

- Simplified the common code that fopen(), freopen() and fdopen()
  share and which has to figure out by looking at a file access
  mode specification which parameters should be used.

- Dropped error detection in the scanf() family. The EOF has to be
  good enough. Also, ungetc() failure still leads to error handling.



c.lib 1.181 (26.10.2004)

- The scanf() family now always returns the number of assignments made
  unless an error occured or an EOF was hit during conversion before the
  first assignment could be made.

- strtok() did not terminate properly if the last token in the string
  did not end with a separator character but with a '\0' byte. Fixed.

- The directory scanning functions opendir/closedir did not get the global
  directory data tracking data structure initialized which later led to
  Enforcer hits and maybe trashed memory. Fixed.


c.lib 1.180 (23.10.2004)

- The printf() family now produces no output at all for %e, %f and %g if the
  floating point support code is disabled. Previously, a minimum field width
  specification could take effect, printing a series of 0 or blank space
  characters where no output should have been produced.


c.lib 1.179 (22.10.2004)

- The scanf() family no longer assumes that a leading '0' indicates that the
  following digits form an octal number if the conversion type has been
  specified as already '%x' already.


c.lib 1.178 (7.10.2004)

- The OS4 version had floating point math support code enabled in all
  libraries and not just "libm.a". Fixed.


c.lib 1.177 (29.9.2004)

- Moved the locale initialization/cleanup code into constructors
  and destructors.

- The socket cleanup function is now a destructor.

- The math cleanup function is now a destructor.

- The wildcard cleanup function is now a destructor.

- The stdio cleanup function is now a destructor.

- The stack extension cleanup function is now a destructor.

- The code that cleans up after the program's current directory
  was changed is now a destructor function.

- Moved the initialization/cleanup code for unlink() into constructors
  and destructors.

- Moved the initialization/cleanup code for usergroup.library into
  constructors and destructors.

- Added usleep(), and created wrapper code that both sleep() and
  usleep() can use.

- Added strtoll() and strtoull(), with further changes to <limits.h>
  and <stdlib.h>.

- The socket exit code now calls the common stdio function which
  flushes and shuts down all buffered and unbuffered files.

- Fixed the stack swapping function which, for reasons unknown, ceased
  to work...

- The start time used by clock() is now initialized by a constructor
  function.

- NOTE THAT ALL THE CHANGES WITH REGARD TO USE OF DESTRUCTOR AND
  CONSTRUCTOR FUNCTIONS REQUIRE A COMPLETE REBUILD OF THE LIBRARY! IF
  YOU DO NOT DO THIS, THE CONSTRUCTOR/DESTRUCTOR FUNCTIONS MAY NOT
  GET CALLED.


c.lib 1.175 (10.9.2004)

- <stdarg.h> now tries to use the compiler supplied, machine specific
  <stdarg.h> file and also includes the OS4-specifc <sys/amigaos-va.h>
  file, if necessary.

- Added another test to <sys/time.h> to check if the "timeval" definition is
  already in place. If it is, <exec/types.h> will not be included.

- <sys/time.h> can be made to define the timeval data structure locally if
  the __USE_CLIB2_TIMEVAL preprocessor symbol is defined. In this case no
  <exec/types.h> or <devices/timer.h> file will be read.

- The sprintf() family did not properly handle empty strings as format
  specifications. This would result in a necessary buffer flush action
  getting skipped, which consequently did not put the string termination
  character into the output buffer.


c.lib 1.174 (27.8.2004)

- The parameters of atan2() were swapped. Fixed.

- Merged fdlibm 5.3 changes with __kernel_tan(), __exp() and __pow()
  functions.

- Replaced the rint() and rintf() functions with the fdlibm code.


c.lib 1.173 (25.8.2004)

- The <assert.h> header file was missing the C++ 'extern "C" { .. }'
  declarations.

- Added a new function __get_default_file() which allows direct access
  to the low level file handle/socket descriptor associated with a
  file descriptor.


c.lib 1.172 (21.8.2004)

- acos() now returns 0 for a domain error.

- asin() now returns 0 for a domain error.

- atan2() now returns 0 for a domain error.

- fmod() now returns x if y == 0 and sets a
  domain error.

- sqrt() now returns 0 for a domain error.

- Added NaN and +Inf constants to the math library
  initialization code which could come in handy later.


c.lib 1.171 (16.8.2004)

- Added "math_hypot.c"

- The 68k build makefile now builds the vfprintf/vfscanf functions
  with %lld/%llu support.

- Split the release notes file into changes and actual release notes,
  which may overlap but differ in technical content.


c.lib 1.170 (14.8.2004)

- Added "math_logb.c", and it appears to work, too.


c.lib 1.169 (8.8.2004)

- Since the 'long double' data type is not really supported by the GCC
  versions we use (and not supported by SAS/C either) building the
  library with support code for it is now a configurable option. Unless
  enabled, 'long double' is now treated like 'double'.

- Reworked the HUGE_VAL definition which previously would default to
  Infinity. The new code properly defines this to the largest
  representable floating point number. The 'great' thing about the
  HUGE_VAL definition in this library is that it's a reference to
  a binary constant initialized by the library startup code. So any
  application checking for HUGE_VAL merely needs to be relinked
  rather than recompiled.


c.lib 1.168 (7.8.2004)

- Ditched __is_infinity() and __is_not_a_number(), brought them back
  as isinf() and isnan().

- Updated code and header files to state exactly which ISO 'C' version
  is meant by the "The following is not part of the ISO 'C' standard."
  warning.

- Integrated rint() and rintf().


c.lib 1.167 (29.7.2004)

- Modified __is_infinity() and __is_not_a_number() to use more portable
  methods for accessing the low level representation of the IEEE 754
  numbers they work on.

- Fixed a compiler warning in h_strerror()


c.lib 1.166 (28.7.2004)

- Added h_strerror() function to libnet.a; there's a global 'h_errno'
  variable available, too.

- <signal.h> now defines a type 'sig_t'.

- <unistd.h> now allows the 'fd_set' type to be referenced as
  'struct fd_set', too.


c.lib 1.165 (26.7.2004)

- In printf(), if the precision is 0 and the value to be printed for
  the %d, %i, %u, %o and %x conversion types is 0, no output should
  be produced. Fixed.