path: root/externals/grill/flext/changes.txt

flext - C++ layer for Max/MSP and pd (pure data) externals

Copyright (c) 2001-2004 Thomas Grill (xovo@gmx.net)
For information on usage and redistribution, and for a DISCLAIMER OF ALL
WARRANTIES, see the file, "license.txt," in this distribution.  

This package seeks to encourage the development of open source software
for the pd and Max/MSP platforms.

Donations for further development of the package are highly appreciated.

----------------------------------------------------------------------------

Version history:

0.4.8:
- fixes for 64 bit builds (size_t is integer type of pointer size)
- finally added a autoconf system (many many thanks to Tim Blechmann)
- more character escapes for attribute editor

0.4.7:
- added flext::GetBool (just because flext::GetInt has been there for a while)
- added bool to usable types for creation arguments
- protection for SIMD routines with count=0
- support exceptions for setup functions, object creation and destruction, method handling
- small fixes for MSVC6 incompatibilites
- fixes for Max 4.5 headers
- removed Mach-O/CFM glue stuff again and added makefiles for Max/MachO
- in flext::StopThread don't wait for thread to have stopped!
- support PD tooltips (only in devel branch at the moment)
- attribute editor: added individual apply buttons
- attribute editor: added escaping for , and ; 
- attribute editor: zoomed editor accepts , as newline separator
- attribute editor: close editor window on object destruction
- fixed $0-arguments in attribute saving
- introducing ring buffer for message queue (thanks to Tom Schouten)
- printing of atoms into buffer with better error checking

0.4.6:
- added a text edit window for list attributes
- finally use global allocator operators (MFC doesn't like it but who cares)
- fixed entry of local variables with attribute editor
- help window for attribute editor
- appended lacking ~ for tilde object help patches
- uniform STL container for all kinds of attribute, method, class handling
- support for MinGW
- added flext_obj::CanvasArgs
- fixed small problem with buffer->Valid() on startup
- fixed buffer overrun problems with flext::post, flext::error... (but still to improve)
- improved handling of patcher arguments
- also initialize flext::sym_int as MakeSymbol("int") for PD (since the variable is there anyhow)
- fixed bug with attribute dumping
- default arguments for object construction (types float0,int0,t_symtype0 for FLEXT_NEW*)
- added flext::sym__ for empty symbol ""

0.4.5:
- added some more SIMD functions
- PD attributes: added a TCL/TK editor dialog ("properties") and save attributes to patcher
- fixed wrong returned result of flext::buffer::set function
- fix for linux static exported function name-clash (individual linker namings for exported flext base classes)
- made message queue stuff global (static) for all flext objects
- added flext_base::DumpAttrib to send an attribute value to the attribute outlet
- added "getmethods" message (for attribute-enabled externals) to list methods for a specified inlet (default = 0)
- "getattributes" or "getmethods" output lists are now alphabethically sorted
- flext::StopThread function to terminate running threads (started with LaunchThread)
- added flext::CmpAtom and AtomList::operator < methods ... useful for using lists as keys for STL
- make use of new PD thread locking (PD functions sys_lock() and sys_unlock()), queue messages only if necessary
- better FLEXT_OSAPI resolution
- ListAttributes (or the getattributes message) now lists attributes in the order they were created (first class, then object scope)
- enforcing usage of STL
- explicit boolean attributes (great for attribute editor layout!)
- flext_dsp reported wrong inlet/outlet count (CntInSig, CntOutSig functions)
- added support for patcher arguments for attributes (use # instead of $ to save them with a patch)
- moved memory functions and console output from flext to super-base-class flext_root and derived all helper classes from that
- changed FLEXT_THREAD_X: wrappers doesn't delete the passed data pointer anymore -> the called method is responsible for that (and it knows the type....)
- enabled binding of more than one function to a symbol and added flext_base::GetBoundMethod
- native support for Windows threading model

0.4.4:
- fixed deadly bug for Max/MSP method-to-symbol-binding proxies
- some fixes for CodeWarrior Mach-O compilation 
- fixed destruction bug for hashed "item arrays"
- added flext_base::ToSelf* methods for queued (deferred) self messaging
- deleted flext::buffer console outputs for undefined arrays/buffer~s: should be done by externals
- a few changes for the flext shared library
- tiny update for PD 0.37 header file usage
- fixed a bug associated to attribute argument lists
- added functions for version number and string (flext::Version and flext::VerStr, mainly for usage of a shared flext library)
- flext_base::UnbindMethod returns user data
- fixed compilation problem with t_symbol-type attributes
- added methods for sending boolean values (ToOutBool, ToQueueBool, ToSelfBool)

0.4.3:
- added forgotten flext_base::ToQueueString method
- added timer functions and flext::Timer class
- added functions for SIMD support
- fixed race condition when using LaunchThread in a setup function (now waiting for thread helper to initialize)
- added flext::Forward function to send messages to bound symbols
- added "zero" flag to flext::buffer resize operation
- fixed bug for Max/MSP buffer resize with preservation of contents
- fixed bug with thread initialization (caused PD@OSX to crash on startup)
- flext::setup is only run once now
- fixed creation bug with objects that have attributes but no outlets
- added FLEXT_BINDMETHOD,FLEXT_UNBINDMETHOD to bind a method to a symbol 
- fixed bug with hard thread termination (of incooperative threads)

0.4.2:
- started port for jMax
- moved CLASS_MAINSIGNALIN to class scope (fixed "float method overwritten" warning)
- unix makefiles: CXX should be commented out if standard (to enable environmental settings)
- introduced default returns for the different flext::GetA* functions
- pragma for Metrowerks compilers: enumsalwaysint on, bool on
- MaxMSP: added assist string for attribute outlets
- added new POSIX for ThrCond::TimedWait (but still have to find out when to enable it....)
- added CopySamples and ZeroSamples
- fixed bug for DSP object with zero signal inlets in PD
- added validity check for sample buffers

0.4.1:
- full port for Max@OSX
- completely redesigned message and attribute handling: now hashed and much more efficient
- greatly enhanced object creation and destruction (esp. for library objects)
- class setup functions now take t_classid type arg... this is BACKWARDS-INCOMPATIBLE for Max/MSP
- added some prerequisites for usage of flext as a shared library
- completed Max/MSPs inlet/outlet assist description functionality
- Max/MSP signal objects: fixed bug of reporting wrong number of inlets
- put overloaded new/delete into flext support class
- introduced "Finalize" virtual method, called after all "Init" methods
- fixed crash issue in flext_dsp when there are NO signal inlets or outlets defined
	(this is possibly only a problem for the signal2 tutorial example)
- added flext::GetType(t_atom &), flext::ZeroMem
- put the clock_delay for the message queue inside the thread-locked area
	Ok for the actual object, but PD has to be thread-safe here as well
- BACKWARDS-INCOMPATIBLE CHANGE: flext_base::m_methodmain and flext_base::m_method_ have got 
	const modifiers.... these virtual methods are used rarely (except for example in py/pyext)
- now MUCH cleaner platform identification... FLEXT_SYS,FLEXT_OS,FLEXT_CPU definitions
- also FLEXT_DEBUG for debug build
- SndObjs: virtual FreeObject routine is now called at destruction (by parent class), derived class doesn't need to call it!
- SndObjs: fixed typo (numbers of output slots was wrong) and init bug
- STK: added support for STK object classes (flstk.{h,cpp})
- introduced a helper thread for launching flext threads in the background
- threads are now handled on class (as opposed to object) level
- threads can now be terminated separately
- put more flext functions into flext static class
- replaced ChangePriority by RefPriority
- fixed setting of priorities...  problems were caused by a compiler bug (MSVC 6)
- made a portable threading interface with support for threading libraries other than pthreads (FLEXT_THREADS defined as FLEXT_THR_*)
- implemented threading support with the MacOS MP thread library
- stripped the ThrMutex and ThrCond classes of non-portable (and irrelevant) functionality
- simplified "ToQueue*" and threaded "ToOut*" message queue mechanism for Max/MSP 
- deprecated FLEXT_ADDMETHOD_V and FLEXT_ADDMETHOD_A definitions which only lead to confusion
- BACKWARDS-INCOMPATIBLE CHANGE: flext_sndobj::NewObjs must return a boolean!!

0.4.0:
- the use of the const keyword is enforced (e.g. the preferred type for symbols is now "const t_symbol *")
	- there _might_ be some problems with sensitive compilers
- introduced Max/Jitter-like attribute functionality ("@value" command line, "getvalue" get and "value" set functions)
- introduced a flext static class for general flext functions (to clean up the flext_base class)
- creation argument handling is now done by flext
	no more weird PD re-ordering of arguments 
- no more support for the Apple MPW compiler - MacOS9 is dying anyway...
- calling SetupInOut() has become obsolete 
	- flext creates all inlets/outlets by itself after the constructor has finished
	- this implies that CntIn(),CntOut() and the outlet pointers are not valid in the constructor
	- there is a virtual bool Init() function that may be used for such initialization
- completely redesigned FLEXT_NEW macros, usage of dynamic classes (in fllib.cpp)
- added ToQueue* functions - like ToOut* but messages or not directly sent (well suited for deadlock situations)
- introduced additional per-class methods and attributes (just like it ever was in Max and PD)
- fixed potentially dangerous typo in flext.cpp - (was: FLEXT_THREAD instead of FLEXT_THREADS)
- added OSX/darwin support (originally done by Adam T. Lindsay)
- SndObj interface now also available for cygwin and BCC 
- added prepend and append functions to AtomList class
- added IsNothing, SetNothing, CanbeBool and GetABool functions
- eliminated the remaining Get*Flint and Set*Flint functions 
- added/completed Is/Canbe/Get/Set for pointer atoms
- added print/scan functions for atoms
- fixed anything outlets for Max/MSP... for some strange reason this severe bug has not had severe consequences....

0.3.3:
- PD: fixed bug for DSP objects having no signal inlets
	this also enables floats into a non-signal leftmost inlet
- revisited priority stuff for detached threads
- Bind/unbind functions for flext classes (in MaxMSP only one object can be bound)
- made "t_symtype" another synonym for "t_symbol *"
- added forgotten dsp_free function for MaxMSP
- fixed forgotten __class__ member for MaxMSP libraries
- changed basic MaxMSP object to t_pxbox (ok, a bit more memory is used....)
- MaxMSP library can now be loaded from startup folder (but can't use aliases)
- if no handler is found for pure anything (just symbol, no args...) try list handler
- added interface for SndObj (http://www.may.ie/academic/music/musictec/SndObj/ )

0.3.2:
- Doxygen generated inline documentation / functional reference
- added some more functionality to the AtomAnything class
- bugfix for threaded methods with var list or anything arguments
- added threaded method for arbitrary (void *) data structs
- ThrMutex: added lock count functions Push() and Pop()
- eliminated potentially dangerous SetThing,GetThing and introduced DoBind,DoUnbind,GetBound instead
- fixed severe bug with the current threads list
- add path specification possibility for help symbols (in FLEXT_NEW*)
- use pthread_attr and DETACHED flag

0.3.1:
- added some more functionality to the AtomList class
- fixed forgotten Sleep implementation in pd@unix
- reorganized file structure and config/make procedures

0.3.0:
- added CYGWIN support

- added threaded methods along with a message queue for ToOut* functions (very unstable for MaxMSP!)
  to use threads compile flext with FLEXT_THREADS definition
- check/update function for buffer change (resize etc.)
- description text for inlets/outlets (e.g. for MaxMSPs assist function) - not fully implemented
- added buffer resize functions flext_base::buffer::Frames(int,bool)
- added some utility functions: Sleep, CopyAtom, CopyList
- added List manipulation classes: AtomList, AtomAnything
- Alias object names (simply specify with FLEXT_NEW*, separated by whitespace)
- float messages - int method is called if there is no float method
- MaxMSP: int messages - float method is called if there is no int method
- fixed type warning for class constructors with int arguments in PD
- fixed severe bug concerning symbol methods
- MaxMSP: use critical sections for message output (for more safety in overdrive mode)
- PD: default/manual definition of help file by flext_base::DefineHelp
- added GetThing/SetThing to access t_symbol's s_thing data member

- introduced FLEXT_NEW_DSP* and FLEXT_LIB_DSP* for FLEXT_NEW_TILDE* and FLEXT_LIB_TILDE* (the latter become deprecated)
- all variable argument defs (aka gimme) now have a V instead of G  (e.g. FLEXT_NEW_V)

0.2.3:
- restructured files and started usable inline documentation
- found that there was no method to output a bang?! Ugh! -> corrected
- finally eliminated awful t_flint type and all other schizophrenic flint functions (float and int exist now equally)
- now (finally) using type t_sample for sample values (should just be identical to float)
- added AddInBang/AddOutBang (same as Add*Symbol - unchecked)
- buffer class: added "bool Ok()" check function
- switched on error posting for unhandled messages
- added XletCode and AddInlets/AddOutlets for multiple Inlet/Outlet addition (max. 9 inlets/outlets)
- if float or int message is not handled explicitly then [list float/int( is tried

0.2.2:
- added xgimme argument type (supplies method handlers with symbol+gimme)
- more information on DSP system with flext_dsp object (block size,number of audio inputs and outputs)
- cleaner gcc makefile
- made NewAligned/FreeAligned static functions
- introduced FLEXT(_TILDE)_SETUP for class setup in a library setup function
- introduced external libraries for MaxMSP, changed library setup behavior (now via FLEXT_LIB_SETUP)
- included MaxMsp's inofficial buffer.h with consent by David Zicarelli
- changed dynamic casts in callback functions to static as gcc 3.0.4 has obvious bugs

0.2.1:
- fixed bug in message parsing code (anything messages were not correctly mapped to inlets)
- changed t_symtype to t_symptr (= t_symbol *), needed for method argument lists
- eliminated flint type (t_flint remains)

0.2.0:
- internal proxy objects for any non-leftmost inlets
- Max/MSP: all signal inlets can receive messages 
- method/argument parsing is done by flext - float/int are not distinguished, the first handler wins
- integrated more system functions into flext_base & eliminated superfluous #defines
- distribute list (into inlet 0) elements over inlets (right to left)
- added outlets for anythings
- defines for callback-to-method functions and method setup (FLEXT_CALLBACK*, FLEXT_ADD*)
- uses PD's or Max's memory allocation functions (for safety in Max's overdrive)
- no support for default arguments (A_DEFFLOAT and A_DEFSYMBOL).. use GIMME instead!
- better graphics update behavior for PD
- improved behavior for invalid/undefined buffers/arrays
- use MaxMSP internal z_disabled flag with flext_dsp for pausing/resuming dsp processing 
- included CHECK_TILDE, a test whether a tilde object (defined as FLEXT_TILDE_*) has a trailing ~. (debug mode only)
- changed notation of flext functions from to_out_float like to ToOutFloat like
- eliminated trivial shortcuts (F,I,V,...) for built-in types
- MaxMSP is only capable of 3 creation arguments... that should be sufficient - otherwise use GIMME
- Methods for aligned memory (NewAligned, FreeAligned)

0.1.1:
- documentation for flext.h
- more emancipation from GEM code
- virtually everything renamed
- abstraction for dsp processing
- makefile for BCC
- manual call of extern_setup or main unnecessary for single objects - only in pd libraries
- delayed buffer init (only name is set beforehand)
- loadbang also in PD
- introduced "compatibility mode" which denies platform-specific features 
- fixed severe dsp bug (symptom: "float method overwritten" warning in pd)
- fixed bug: wrong return code from buffer::Set

0.1.0: 
- max-pd 0.2 becomes flext 0.1.0
- heavy usage of unchanged GEM code


----------------------------------------------------------------------------

Notes:
- no support for default object arguments (A_DEFFLOAT, A_DEFSYMBOL) -> use variable argument lists instead

Platform specific:
- PD does not allow signal and message to go into the same inlet (except leftmost inlet)

Restrictions in compatibility mode:
- Max allows only 9 float/int inlets

Porting to new compilers/platforms:
- enums must be int-sized!!!
- compiler must support bool type
- an STL implementation must exist
- no need of C++ exceptions or RTTI (RTTI only for GUI objects)
	BUT: some libraries could use RTTI and then need run-time type information from the library! (this is known for gcc compiles)

----------------------------------------------------------------------------

TODO list:

for 0.4.8:
- SIMD for gcc
- Max/MSP MachO support

for 0.5.0:
- optimizations for object initialization and messaging

bugs:
- recreation of object with different argument line forgets about eventually present attributes

- can't use MFC libraries because of global new and delete overloadings

TEST:
- PD: problems with timed buffer redrawing (takes a lot of cpu time)
- hard thread termination upon object destruction doesn't seem to work properly -> crash
- Max rounding bug ... buffer resize could be one sample less!
- flext_dsp: Max/MSP doesn't correctly report in/out channel counts


general:
- documentation
- add log messages for debugging version
- use PD's t_float and t_int types (or comparable for 32-bit safety)

- add double handlers
- add signal in/out connection query function

- support for Max qelem style

- PD: with DSP objects all float messages to left input are currently converted to signal
		there should be a way to avoid this

- flext::post and flext::error should print via a worker thread (and should be unlimited in characters)

tests:
- PD: figure out what "pointer" messages do and where they occur
- buffer resize: flext_base::buffer::Frames(): should we use buffer or system sample rate?

- check whether m_dsp gets called upon deletion of a used buffer (PD and MaxMSP may behave differently).


features:
- manage running threads individually (stop, isrunning?, priority etc.)