This file describes implementation and API changes; stuff more visible to the
user appears in the "release notes" instead.  See the bottom of this file
for original notes on source style and organization.

0.42.0 

changed definition of t_float, t_sample, t_floatarg so that they can be
set via #defines (PD_FLOATTYPE, etc).

0.41.0

add support for callback-based audio I/O
headers & code changed to use t_float or t_sample instead of float (patches
by zmoelnig).


0.40.0

0.39.0

canvas_dspstate and signalinlet_new() exported to m_pd.h

new function verbose() added to m_pd.h 

0.38.0

finally figured out how to do "-enable-", etc., flags in the configure
script correctly.

The scheduler now has a hook (set_so you can add polling routines) :
    sys_idlehook().

I'm now uploading directly to CVS repository ("main" and "stable_0_38"
branches.)  There are still problems keeping CVS's and my versions of
portaudio the same (CVS bashes the "ident" lines).

t_int to int in binbuf_addv prototype

64-bit fix to externs makefiles

Pd now uses portaudio out-of-the-box; customized files are moved to
"src" directory.

All "tags" are printf'd as %lx to make 64-bit safe.  

GUI queueing mechanism added: sys_queuegui() etc. 

massive rewrite of array code to unify arrays and scalars.

fixed empty lists automatically to call "bang" method if one is supplied.

rewrote the "namelist" stuff to facilitate preference saving (s_stuff.h,
s_path.c, s_file.c, s_main.c)

0.37.2

expr() "exp" temporary variables renamed to avoid compilation problems

0.37.1

makefile.in: MANINSTDIR = $(DESTDIR)/$(MANDIR) changed to
    $(DESTDIR)/$(INSTALL_PREFIX)/$(MANDIR) (thx. Mathieu Bouchard)

applied 2 jack patches from Luke Yelavich

add -fno-strict-aliasing to config script (linux&mac) to improve underflow,
etc., protection

add underflow protection to vcf~ object; rewrote underflow protection to be
faster in throw~/catch~ and send~/receive~

fixed bug in -inchannels/-outchannels arg parsing

fixed u_main.tk to make "apple" key work to accelerate menus on MACOS

fooled with MIDI to try to get sysex and other system messages to work.
Needs lots of testing now...

finally fixed OSS to open audio with NODELAY... also cleared dup-on-exec flag.

bug fix in scalar_properties

major editions to the IEM GUIs to fix bugs in how "$" variables are handled.
The code still isn't pretty but hopefully at least works now.

tried to get alsa noninterleaved access to work (needed for RME).  Failed
to get my RME card to load under ALSA and gave up for now.

fixed scalar drawing to fail gracefully when the template canvas disappears

bug fix in vd~ for very small delays (d_delay.c)

set up sys_oldtclversion flag to correct for changed text selection
(u_main.tk, s_main.c, g_rtext.c)

tried again to add "readn" support to s_audio_alsa.c: coded, but failing so far.

fixed broken octave divider example

removed "-Werror" from default makefile; fixed  configure script to respect
"CFLAGS" environment variable instead.  Suggest developers should use
"setenv CFLAGS -Werror".

added "-alsaadd" flag so people can specify alsa devnames to add to list.
fixed some problems with Pd crashing when ALSA failed to open.

took out the 2-pixel padding for MSW in g_canvas.g (HORIZBORDER/VERTBORDER)

went back to s_midi_mmio (portaudio version got assertion errors and anyway
I could never get sysex working in it as I had wanted.)

Took bug fixes from s_midi_pm.c, s_audio_jack.c, s_inter.c from "devel" branch;
also added "static" flag to configure.in (but the devel configure.in as a whole
doesn't seem to work for OSX, for me at least.)
 
Might have fixed a bug where labels disappear in buttons, etc, when saved
and reloaded.

brought s_audio_alsa.c up to alsa 1.0.0 compatibility

fixed "-alsaadd" (never worked before)

fooled with macintosh audio.  Fixed some (not all) of the audio I/O APIs
to deal with open failures better (reducing sys_{in,out}channels accordingly)

In the Alsa API, the synchronization test was too stringent and was loosened
to 3*DACBLKSIZE/2.

'make install' fixed to deal with 'extra' correctly.

one more improvement in jack support (guenter)

make an "nrt" flag so mac can disable pthread_setschedparam call if yu want.

------------------- original source notes -------------

0.  structure definition roadmap.  First, the containment tree of things
that can be sent messages ("pure data").  (note that t_object and t_text,
and t_graph and t_canvas, should be unified...)

------------ BFFORE 0.35: ---------
m_pd.h	    t_pd    	    	    anything with a class
    	    	t_gobj	    	    "graphic object"
    	    	    t_text  	    text object
g_canvas.h  
    	    	    t_glist 	    list of graphic objects
g_canvas.c  	    	t_canvas    Pd "document"

------------ AFTER 0.35: ---------
m_pd.h	    t_pd    	    	    anything with a class
    	    	t_gobj	    	    "graphic object"
    	    	    t_text  	    patchable object, AKA t_object
g_canvas.h     	    	t_glist     list of graphic objects, AKA t_canvas

... and other structures:
g_canvas.h  t_selection -- linked list of gobjs
    	    t_editor -- editor state, allocated for visible glists
m_imp.h     t_methodentry -- method handler
    	    t_widgetbehavior -- class-dependent editing behavior for gobjs
    	    t_parentwidgetbehavior -- objects' behavior on parent window
    	    t_class -- method definitions, instance size, flags, etc.


1.  C coding style.  The source should pass most "warnings" of C compilers
(-Wall on linux, for instance; see the makefile.)  Some informalities
are intentional, for instance the loose use of function prototypes (see
below) and uncast conversions from longer to shorter numerical formats.
The code doesn't respect "const" yet.

1.1.  Prefixes in structure elements.  The names of structure elements always
have a K&R-style prefix, as in ((t_atom)x)->a_type, where the "a_" prefix
indicates "atom."  This is intended to enhance readability (although the
convention arose from a limitation of early C compilers.)  Common prefixes are
"w_" (word), "a_" (atom), "s_" (symbol), "ob_" (object), "te_" (text object),
"g_" (graphical object), and "gl_" (glist, a list of graphical objects).  Also,
global symbols sometimes get prefixes, as in "s_float" (the symbol whose string
is "float).  Typedefs are prefixed by "t_".  Most _private_ structures, i.e.,
structures whose definitions appear in a ".c" file, are prefixed by "x_".

1.2.   Function arguments.  Many functions take as their first
argument a pointer named "x", which is a pointer to a structure suggested
by the function prefix; e.g., canvas_dirty(x, n) where "x" points to a canvas
(t_canvas *x).

1.3.  Function Prototypes.  Functions which are used in at least two different
files (besides where they originate) are prototyped in the appropriate include
file. Functions which are provided in one file and used in one other are
prototyped right where they are used.  This is just to keep the size of the
".h" files down for readability's sake.

1.4.  Whacko private terminology.  Some terms are lifted from other historically
relevant programs, notably "ugen" (which is just a tilde object; see d_ugen.c.)

1.5.  Spacing.  Tabs are 8 spaces; indentation is 4 spaces.  Indenting
curly brackets are by themselves on their own lines, as in:

    if (x)
    {
	x = 0;
    }

Lines should fit within 80 spaces.

2.  Max patch-level compatibility.  "Import" and "Export" functions are
provided which aspire to strict compatibility with 0.26 patches (ISPW version),
but which don't get anywhere close to that yet.  Where possible, features
appearing on the Mac will comeday also be provided; for instance, the connect
message on the Mac offers segmented patch cords; these will devolve into
straight lines in Pd.  Many, many UI objects in Opcode Max will not appear in
Pd, at least at first.

3.  Compatibility with Max 0.26 "externs", i.e., source-level compatibility. Pd
objects follow the style of 0.26 objects as closely as possible, making
exceptions in cases where the 0.26 model is clearly deficient.  These are:

3.1.  Anything involving the MacIntosh "Handle" data type is changed to use
char * or void * instead.

3.2.  Pd passes true single-precision floating-point arguments to methods;
Max uses double.
Typedefs are provided:
    t_floatarg, t_intarg for arguments passed by the message system
    t_float, t_int for the "word" union (in atoms, for example.)

3.3.  Badly-named entities got name changes:

    w_long --> w_int (in the "union word" structure)

3.4.  Many library functions are renamed and have different arguments;
I hope to provide an include file to alias them when compiling Max externs.

4.  Function name prefixes.
Many function names have prefixes which indicate what "package" they belong
to.  The exceptions are:
    typedmess, vmess, getfn, gensym (m_class.c)
    getbytes, freebytes, resizebytes (m_memory.c)
    post, error, bug (s_print.c)
which are all frequently called and which don't fit into simple categories.
Important packages are:
(pd-gui:)   pdgui -- everything
(pd:)	    pd -- functions common to all "pd" objects
    	    obj -- fuctions common to all "patchable" objects ala Max
    	    sys -- "system" level functions
    	    binbuf -- functions manipulating binbufs
    	    class -- functions manipulating classes
    	    (other) -- functions common to the named Pd class

5. Source file prefixes. 
PD:
s    system interface
m    message system
g    graphics stuff
d    DSP objects
x    control objects
z    other

PD-GUI:
t    TK front end