Changes

/*
 * Femto OS - Copyright (C) 2008-2014  Ruud Vlaming
 *
 * This file is part of the Femto OS distribution.
 */

plans for 0.93 => 0.94
(1)  Feature to determine tasks that run 100% load.
(2)  Simple heapmanager
(3)  Port of shell application to other avr hardware
(4)  cross platform terminal en trace facility.

plans for 0.922 => 0.93
(1)  xmega support

changes 0.92 => 0.922 [23.06.2011]
(1) Corrected two minor bugs in femto os core
(2) Upgrade to AVR Studio 5

changes 0.912 => 0.92 [18.02.2010]
(1)  Added taskQueuRequestOnName() method for convineance
(2)  Added cfgUseSuspendOnDelay to make it easier to handle suspend requests.
(3)  Added genQueuUnread() and genQueuUnwrite to undo queue operations,
     handy in failing communications.
(4)  minor fixes


changes 0.91 => 0.912 [26.11.2009]
(1)  fixed bug for several demo's (FlashLeds broken)
(2)  Support for WinAVR and AVR Studio (with installer)
(3)  minor fixes

changes 0.89 => 0.91 [15.10.2009]
(1)  taskKillTask,taskRestart become taskTerminate, taskRecreate
(2)  Rewrote switchblock handling, now handles watchdog, blocking,
     delays, file system operations etc in a senseable manner.
(3)  Accuracy of timemeasurement of taskmanager improved (a little)
(4)  Introduction of the shared stack, task that not need to run
     simultaneaously may share the stack. Shared task run cooperative
     among each other, but preemptive with the other tasks.
(5)  Suspension of tasks now required cooperation of the task to be
     suspended. In prior implementations the result of suspend was
     unclear. Blocked tasks where ignored, whereas suspendig tasks
     holding locks where also blocking other tasks. 
(6)  Introduced to extra tests on stackusage, on of which explictly
     checks a predefined savety region of the stackspace.
(7)  Extended the trace possibilities, userwords can now be traced
     as well as taskloads, os timing, watermarks and registeruse.
(8)  Corrected a few typo's in the code and auxiliary scripts
(9)  Introduced hierarchical round robin prioritizing. This makes
     sure no lower priority tasks are starved from execution in
     case of regular alternating task switches.
(10) Introduced prioritized release to force the unblocking in
     the highest priority when needed.
(11) Solved bug in load measurement of isr's.
(12) Extended error messages with function number.
(13) Redefinition of all error callId's (and thus trace ID's)

changes 0.88 => 0.89 [11.09.2009]
(1) Made possible the separation between background variables and stack os.
(2) Introduce several other internal memory protection measures against
    agressive optimization of gcc.
(3) Adjustments to the installer
(4) 13 new ports
(5) New HelloWorld example and getting started guide.
(6) Improved compile and flash script. Spits out info about internal numbering.

changes 0.87 => 0.88 [06.02.2009]
(1) Solved bug, priority was not correctly restored after lifting.
(2) Reintroduced tick interrupt blocking, but now implemented correctly.
    This was needed since without the possibility to enable/disable
    tick interrupts it was not possible to run in cooperative mode.
    Functionality regarding switch critical is kept.
(3) Solved bug, tickcounter was not increased in case all tasks where in
    the cooperative mode.
(4) Solved a bug in the genLogTask that served incorrect values for
    Task Load measurement (cumulative values were passed instead of
    totals).
(5) Improved and expanded error handling, it is now possible to mark
    all errors as fatal, that gives you more time to inspect what
    went wrong in your code. Errors inside isr are fatal per default.
(6) Orthogonalized configuration parameters somewhat more, and i
    renamed a few methodes. I am not completely satisfied yet, but
    for now it must be sufficient. Removed vfgBoolReturn which is
    now default with cfgTimeout
(7) Removed backend hack for the compiler, use of OS_task attribute
    is a suitable replacement.
(8) Improved the toolchain installation script.


changes 0.86 => 0.87 [28.11.2008]
(1)  Added the support for stacks larger as 255 bytes. The preprocessor
     determines if the extra code needs to be compiled in. Furthermore,
     if all stacks fit in the zeroth ram page, the SP_H is set permanently
     to zero. (Saves some bytes on the really small devices)
(2)  The sleeptime was calculated assuming the tickrate was fixed at approx 1000Hz.
     However, this need not to be true, this has been corrected. Furthermore
     configuration parameters have been added to set several prescalers and
     the clock frequency.
(3)  Extra preprocessor functions that make live easy.
(4)  Ports to 30 new devices, most examples run on all devices now,
     (depending on the available hardware of course).
(5)  For tracing, USI support has been removed and has been replaced by
     bitbanging. This is much more portable.
(6)  isrFireEvent introduced for those devices that do not have a GPIO register.
(7)  Solved a (non critical) bug in the way the file system handled sleeping.
(8)  Setting tick interrupts is removed. It turned out that even if tick
     interrupts where disabled, tasks could still switch (through an other
     interrupt with forced context switch. Thus i changed
     'enter/exit tick critical' into 'enter/exit switch critical'. This is
     still interrupted by a tick, but you can be sure no other tasks get
     started.
(9)  The used of the T bit has been changed. There is no more need for external
     storage upon context switch. (Not all devices have a register free). Since
     this new trick utilizes the H flag (lost upon context switch) it is still
     possible if needed.
(10) Changed from gcc-4.2.3 libc-1.61 binutils-2.18 to gcc-4.3.2 and libc-1.6.2
     binutils-2.18. (All fully pathced) A hacked backend is no longer needed due
     to the availability of the OS_task attribute. Bikini maps to that attribute.
(11) Support for three bytes pc on the stack (mega256x)

changes 0.85 => 0.86 [23.08.2008]
(1)  Added the support for a file system, which required the major
     change in the way blocking is handled inside the Femto OS.
(2)  Removed the need for volatile variables, reduced footprint
(3)  Introduced anonymous unions for the tickcounter, reduced code
(4)  Added support for named events which can be fired from every
     location (isr, tasks, barks) and which are also usable when
     the core is interruptible.
(5)  Added the support for capabilities to be defined per task. The
     gives a find grained mechanism for the preprocessor to further
     optimize then code. Saves ram and flash.
(6)  API was changed on several points to reflect the new possibilities
(7)  Solved a critical bug in the portRestore methode, registers r3-r0
     were not saved correctly in some situations, resulting in a crash.

changes 0.841 => 0.85 [18.06.2008]
(1)  genKill changed into taskKill due to the fact that we must be
     able to clean the slotstack en release the blocking tasks that
     were hold by the killed task.
(2)  Corrected a (very rare) situation where a task could be barked
     at by the watchdog, while in a blocked state and return blocked
     without any locks stored. Now, a recreated task always will
     be non blocked and non delayed.
(3)  Suspend is not longer possible on blocked tasks for we have no
     mechanism to release the blocks after resume. It is still possible
     when it holds locks and is running, but it would be unwise to
     suspend such a task.
(4)  Several internal changes.
Note : version 0.85 was never released as a seperate version.

changes 0.84 => 0.841 [10.06.2008]
(1)  Updated installation toolchain.

changes 0.83 => 0.84 [02.06.2008]
(1)  The synchronization mechanism was rewritten. The main improvement
     is that it is now possible to simultaneously lock on two queu's
     or mutexes. So if you need two resources there is no need to
     spinlock. If you need more as two, this is still needed, but those
     cases are much more rare. In the same proces the footprint for
     using single locks has been reduced.
(2)  Only tasks really needing locks by them selfes consume room for
     the slotstack now. This reduces the need for ram, depening on the
     task.
(3)  Mutexes are no longer called 'SyncLock', but simply 'mutex'.
     Rendez vous is always multiple now, if you need waiting only for
     one other task, supply 2 as parameter, or use the forOne macro.
     Futhermore Mutexes are now treated as a special kind of queue.
     These steps reduced code duplication.
(4)  Delay information is only stored in the tcb if needed, may reduce
     ram in particular applications. The parameter cfgUseDely was introduced.
(5)  In any (external) call that expects an tasknumber you may now supply
     defCurrentTaskNumber. This is only allowed when you are inside
     a real task (code that has been reached from Loop_XXX). From init,
     bark or isr it is strictly forbidden. You can also supply
     defCurrentTaskName in a XXXOnName Call, this has the same effect.
(6)  There have been several changes in the api to make naming
     more consistent and clear.
 
changes 0.82 => 0.83 [29.04.2008]
(1) There was a problem with functions declared naked, and the gcc
    needing a framepointer on those functions. Although there may
    not be a need for a framepointer in a strict sense (for example
    there are still registers free) gcc may decide to use one anyway,
    Even if you compile with -fomit-frame-pointer. If one is used
    while not being setup we have a problem. To resolve several
    solutions are possible.
    - make all functions standard (noretrun will not do, most
      functions do return albeit by a 'virtual' return). This cost
      quite a lot of code. It is made possible by and extra
      configuration parameter.
    - make all functions naked and let a script check for frame-
      pointers and warn about it, and install a counter measures
      framepointer option. This is possible also via an extra
      configuration parameter.
    - hack the avc.c backend and introduce the bikini attribute,
      which only sets up the frame pointer, but does not save any
      registers. This is done too.
    Per default all functions are defined bikini which falls back
    to naked on standard gcc. None of the solutions is ideal, it
    would be a lot better if the compiler would really obey the
    -fomit-framepointer compiler option, which is almost always
    possible, and certainly in the situations i saw gcc using
    a framepointer.
Note: version 0.83 was never released as a seperate version.

changes 0.80 => 0.82 [31.03.2008]
(1)  Changes to the comments
(2)  Slots are no longer default with every task when used (saves bytes)
(3)  Tasks can hold more than one slot, to be determined by the
     SlotSize_Task parameter. Maximum set to 6 simultaneous slots.
(4)  Tasks are numbered 0..M for those who use slots, M+1..N for the rest,
     i.e. the tasks with slots have the lowest tasknumber. The number of
     tasks with slots is cfgNumberOfTasksWithSlot, so that always:
     cfgNumberOfTasksWithSlot <= cfgNumberOfTasks
(5)  With the property above, all loops over tasks with slots could be
     reduced to loop only over those tasks possibly having slots, thus
     reducing overhead.
(6)  Slots should now be used for one type only, i.e. the type is coupled
     staticly to the slot number. This reduces the flexibilty somewhat
     but makes confusion less likely and checking easier. Queus have the
     lowest slot numbers, followed by the mutexes followed by the
     rendez vous (wait locks)
(7)  TaskControlBlock has become of variable length holding the
     slots of a task in a dummy array. Note reading a non existing slot
     may result in a full crash, check!
(8)  Fields uiGlobTick and uiSemaDog (i never liked the names) have been
     replaced by uiTaskMonitor and uiTaskSlot. Also, some bitfields changed
     sides. This lead to increase of some of the examples.
(9)  Internal changes to types and code
(10) Extra example that makes use of multiple slots. FemtoOS_Passon
(11) Removed cfgCheckSynchronization
(12) Adjusted Shell task and communication program to cope with
     new slots. Note you cannot use the new communication program
     on the old Shell task and vice versa.