maintainer: Ivica Bukvic ico@vt.edu
maintainer: Jonathan Wilkes jancsika@yahoo.com
- One Paragraph Overview
- Flavors of Pure Data
- Three Paragraph Overview
- Goals
- Installation Guide
- Contributor Guide
- Core Pd Notes
- GUI Message Spec
Pure Data (aka Pd) is a visual programming language. That means you can use it to create software graphically by drawing diagrams instead of writing lines of code. These diagrams show how data flows through the software, displaying on the screen what text-based languages require you to piece together in your mind.
There are currently three main distributions of Pure Data:
- Pd-L2Ork. Version used by Ivica Bukvic for his laptop orchestra. This guide is for Pd-L2Ork.
- Pure Data "Vanilla". Miller Puckette's personal version which he hosts on his website and maintains. It doesn't include external libraries like objects for doing graphics, video, etc.
- Pure Data Extended. A monolithic distribution which ships with lots of external libraries. As of August 2015 it hasn't been updated since January 2014.
Pd has been designed with an emphasis on generating sound, video, 2D/3D graphics, and connecting through sensors, input devices, and MIDI as well as OSC devices.
Pd has a special emphasis on generating audio and/or video in real time, with low latency. Much of its design focuses on receiving, manipulating, and delivering high-quality audio signals. Specifically, the software addresses the problem of how to do this efficiently and reliably on general purpose operating systems like OSX, Windows, Debian, etc.-- i.e., systems designed mainly for multi-tasking.
Pd can easily work over local and remote networks. It can be used to integrate wearable technology, motor systems, lighting rigs, and other equipment. Pd is also suitable for learning basic multimedia processing and visual programming methods, as well as for realizing complex systems for large-scale projects.
Pd-L2ork has the following goals:
- Documentation. We like documentation. It's like code, except friendly.
- Be reliable. Binary releases must be usable for performances and installations. The git repo must always be in a workable state that can be compiled. Regressions must be fixed quickly.
- Be discoverable. Undocumented features are buggy. Missing help files are bugs. Patches for new functionality that lack documentation are spam.
- Be consistent. Consistent interfaces are themselves a kind of documentation. We like documentation, so it follows that we like consistent interfaces
To install using a pre-compiled binary, follow these instructions: http://l2ork.music.vt.edu/main/?page_id=56
To set up a development environment, first make sure you have the following package dependencies listed here: http://l2ork.music.vt.edu/main/?page_id=56
Then follow the steps outlined here: http://l2ork.music.vt.edu/main/?page_id=56#install-dev
Contributing is easy:
- Join the development list: http://disis.music.vt.edu/cgi-bin/mailman/listinfo/l2ork-dev
- Tell us what you'd like to work on. Unfortunately there are lots of externals and even core features that are poorly documented. We can help make sure you aren't duplicating functionality (or that you at least know what's already been implemented).
- Send us your patch and we'll try it out. If it's well-documented and there aren't any bugs we'll add it to the software.
- If you want to do regular development and have commit access, just request it, then follow the Pd-L2Ork goals above.
Here are some of the current tasks:
- coming up with a better name than Pd-L2Ork. :)
- skills needed: creativity, basic knowledge about programming in Pd
- status: no work done on this yet
- writing small audio/visual Pd games or demos to include in the next release
- skills needed: ability to write Pd programs
- status: I wrote a little sprite-based game that will ship with the next version of Pd-L2Ork. In it, the character walks around in an actual Pd diagram shoots at the objects to progress, and to make realtime changes to the music. What I'd like is to include a new, smallish game with each release that has a link in the Pd console. It can be a little demo or game, just something fun that shows off what can be done using Pure Data.
- porting Pd-L2Ork's graphical user interface from Tcl/Tk to Qt.
- skills needed: knowledge about Qt5/QML, threading, and Pd's core design and deterministic message-dispatching and scheduling
- status: under active development
- designing/implementing regression test template
- skills needed: knowledge about... regression tests. :) But also some expertise in using Pd so that the tests themselves can be written in Pure Data. At the same time, they should be able to be run as part of the automated packaging process (i.e., in -nogui mode).
- status: some externals have their own testing environments, but they are limited as they require manual intervention to run and read the results inside a graphical window. Here's an email thread with Katja Vetter's design, which looks to be automatable: http://markmail.org/message/t7yitfc55anus76i#query:+page:1+mid:chb56ve7kea2qumn+state:results And Mathieu Bouchard's "pure unity" (not sure if this is the most recent link...): http://sourceforge.net/p/pure-data/svn/HEAD/tree/tags/externals/pureunity/pureunity-0.0/
The following is adapted from Pd Vanilla's original source notes. (Found in pd/src/CHANGELOG.txt for some reason...)
Sections 2-3 below are quite old. Someone needs to check whether they even hold true for Pd Vanilla any more.
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...)
BEFORE 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
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.
1.0 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.
2.0. 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 someday 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.
3.0. Compatibility with Max 0.26 "externs"-- 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.
4.0. 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
5.0. 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
Purpose: a set of functions to communicate with the gui without putting language-specific strings (like tcl) into the C code. The new interface is a step toward separating some (but not all) of the GUI logic out from the C code. Of course the GUI can still be designed to parse and evaluate incoming messages as commands. But the idiosyncracies of the GUI toolkit can be limited to either the GUI code itself or to a small set of modular wrappers around sys_vgui.
The public interface consists of the following:
gui_vmess(const char *msg, const char *format, ...);where const char *format consists of zero or more of the following:
- f - floating point value (t_float)
- i - integer (int)
- s - c string (char*)
- x - hexadecimal integer value, with a precision of at least six digits. (hex value is preceded by an 'x', like "x123456")
For some of Pd's internals like array visualization, the message length may vary. For these special cases, the following functions allow the developer to iteratively build up a message to send to the GUI.
gui_start_vmess(const char *msg, const char *format, ...);
gui_start_array(); // start an array
gui_f(t_float float); // floating point array element (t_float)
gui_i(int int); // integer array element (int)
gui_s(const char *str); // c string array element
gui_end_array(); // end an array
gui_end_vmess(); // terminate the messageThe above will send a well-formed message to the GUI, where the number of array elements are limited by the amount of memory available to the GUI. Because of the complexity of this approach, it may only be used when it is necessary to send a variable length message to the GUI. (Some of the current code may violate this rule, but that can be viewed as a bug which needs to get fixed.)
The array element functions gui_f, gui_i, and gui_s may only be used inside an array. Arrays may be nested, but this adds complexity and should be avoided if possible.
The public functions above should fit any sensible message format. Unfortunately, Pd's message format (FUDI) is too simplistic to handle arbitrary c-strings and arrays, so it cannot be used here. (But if it happens to improve in the future it should be trivial to make a wrapper for the public interface above.)
The current wrapper was made with the assumption that there is a Javascript Engine at the other end of the message. Messages consist of a selector, followed by whitespace, followed by a comman-delimited list of valid Javascript primitives (numbers, strings, and arrays). For the arrays, Javascript's array notation is used. This is a highly idiosyncratic, quick-and-dirty approach. But the point is that the idiosyncracy exists in a single file of the source code, and can be easily made more modular (or replaced entirely by something else) without affecting any of the rest of the C code.