Personal tools
You are here: Home documentation manuals Pure Data Chapter 3: Getting Pd to Run
Document Actions

Chapter 3: Getting Pd to Run

by Not dedicated last modified 2006-09-05 03:13 PM Miller S. Puckette
Pd Documentation chapter 3: Getting Pd to run
back to table of contents

Pd runs under Irix, Microsoft Windows, Linux, and Mac OS 10.2 (Jaguar). How to get Pd up and running depends on your operating system, but the overall strategy is the same. You must first get and install it, and then untangle whatever problems arise in handling audio and MIDI input and output, and finally get Pd to meet its real-time obligations reliably. In case of trouble also consult the Pd mailing list archive on , which often has late-breaking news about configuration problems and solutions.

You may be interested in getting only audio output or audio input, or you may need both to run simultaneously. By default, Pd will try to run both, but if you don't need either input or output, you may find that Pd runs more reliably, or at least more efficiently, with the unused direction turned off. This is controlled by Pd's command line flags.

Depending on your application you will have a more or less stringent latency requirement. Ideally, when any input (audio, MIDI, keyboard, network) is available, the outputs (in particular the audio output) should react instantly. In real life, it is necessary to buffer the audio inputs and outputs, trying always to keep some number of milliseconds ahead of real time to prepare for the inevitable occasions where the CPU runs off to service some different task from Pd. How small this latency can be chosen depends on your OS and your audio driver.

To test audio and MIDI, start Pd and select "test Audio and MIDI" from the "help" menu.

TIP: If Pd starts up but you get distortion or glitches in the audio output, this could be either because the "audio I/O buffer" isn't big enough, or else because the CPU load of the patch you're running is too great for the machine you have, or else because the ADC and DAC are out of sync or even at different sample rates. To test for the first possibility, try increasing the "-audiobuf" parameter in the command line (but see also under your OS below.) For the second, start up your favorite performance monitor program; and for the third, try starting Pd up with ADCs disabled.

Here are instructions for getting and installing Pd for the four operating systems it runs on: IRIX, MS Windows, Linux, and Max OSX.

3.1. IRIX (SGI machines)

(NOTE: as of release 0.35 I haven't had an IRIX machine to compile Pd on. Soeren Bovbjerg has kindly compiled 0.35 and 0.36 for IRIX; you can find these at )

Download Pd, which will be a "tar.Z" file. You can unpack this by typing "zcat [name].tar.Z | tar xf -" to a shell. This creates a directory named "pd".

Starting with release 0.25, Pd should come in "n32" and "o32" versions. "o32" is the default and will run on IRIX 5.x and up. "n32" runs faster, but only on 6.x and up. Also, "externs" have to be updated for n32. The "pd" executable (bin/pd in the distribution) is a symbolic link to either "pd-o32" or "pd-n32."

NOTE: "externs" appear to be broken in the N32 version... I'm not sure how long this has been true. If you want to use external objects, you have to use the O32 version.

Please note that the path to the Pd executable program can't contain space characters; don't put it in a directory named "Program Files" for example.

If for example you put Pd in ~, the executable program will be ~/pd/bin/pd. The program looks at its command line to figure out where it is, so it's best to invoke Pd by its full pathname. You should always invoke Pd from a Unix shell because many important messages appear on the standard error.

The simplest way to invoke Pd is to make an alias in your ".cshrc" file (assuming you use the "c" shell) such as:

    alias pd ~/pd/bin/pd

(assuming your Pd distribution landed in ~, for example).

Pd will open the "default" audio input and output devices, without regard for whether they are in sync or not. This will be bad if they aren't; use the "-noadc" or "-nodac" flag to disable either the input or output. Pd is supposed to handle up to 8 channels of audio in and/or out. (But at least one user had to recompile Pd on his Onyx to get 8 channels working.)

As to MIDI, Pd simply attempts to open all available MIDI devices for input and output, which is probably very bad on anything more recent than my Indy. If any MIDI ports fail to open either for input or output, all MIDI is disabled.

Pd has not been fixed to request real-time priority from Irix; it will compete with all other processes on your machine for CPU time.

Audio and MIDI in IRIX

Pd takes command line arguments to set the number of input and output channels and the sample rate. These don't affect the SGI's audio settings, which you have to set separately using the "audio panel." Pd does detect the audio sample rate if you don't specify one on the command line.

On SGI machines, you have to work to get MIDI running. Before you start Pd, verify that least one MIDI port is configured open. Pd opens the FIRST MIDI port that's open. You might want to get rid of the "software" MIDI port if you're running 6.x. On Indys, the usual practice is to open serial port number 2 because some systems configure port 1 as "console" by default. You can use the GUI if you want, or else just type

    startmidi -d /dev/ttyd2

to get port 2 speaking MIDI, and


to stop it. You can test whether MIDI is configured by typing,

    ps -dafe | grep midi

and looking for "startmidi" processes.

It's a good idea to connect your serial port to your MIDI interface before typing the "startmidi" command, not afterward, at least in 5.x. We use the Opcode Studio 3 interface but in principle any Mac-compatible one should work.

The O2 apparently has RS232 ports, not RS422. I think SGI's web site says something about how to deal with this.

3.2. Microsoft Windows

Pd is compiled under NT, but shoould work under any version of Windows since 95. Pd will appear as a "zip" file. Unzip this, creating a directory such as \pd. (You can put it wherever you like but the path should have no spaces in it; so "Program Files" would be a bad place.)

If for example you put Pd in C:\pd, the executable program will be C:\pd\bin\pd. You can simply adjust your path to include C:\pd\bin and just invoke "pd" in a command prompt window. You can also make a "command prompt" shortcut to start Pd.

Pd requires "TCP/IP networking" to be turned on. This doesn't mean you have to be on a real network, but simply that Pd actually consists of two programs that make a "network link" to intercommunicate.

The vanishing window

Pd is a "command line" program. Most error and diagnostic messages from Pd appear on the command prompt window Pd runs from.

If you start Pd from the "run" menu or as a shortcut, and if there's a problem with run-time flags (see the Pd command line below), Pd will print an error and exit. You won't see this error unless you arrange for the "command prompt" or "msdos" window to stay open after Pd exits. One way to do this is to make a "batch" file ("run.bat", say) containing the Pd command line.

Audio in Microsoft Windows

You can ask for a list of audio and MIDI devices by typing "pd -listdev"; you can then specify which audio and MIDI device to use. Type "pd -help" (or make any mistake) to get the syntax for specifying which device to use.

Most PC sound cards seem to have MIDI built in; you don't seem to have to do anything special to get Pd to send and receive MIDI. You can list and choose MIDI devices in the same way as audio.

MIDI timing is very poor if you are using simultaneous audio input and output; if you suppress either audio input or output things will improve somewhat under NT; you can apparently get the jitter down to ~40 msec. On W95 performance is simply terrible. W98, with either audio input or output suppressed, offers fairly good MIDI timing (~5 msec jitter) but crashes occasionally.

Some NT and W98 drivers greet you with a constant trail of "resyncing audio" messages. Sometimes you can fix this by invoking Pd with the "-noresync" flag.


As of version 0.35 Pd supports ASIO. Invoke Pd as "pd -asio" and, if needed, specify "-sounddev" (etc.) flags to specify which device (see "the Pd command line" below.) You can also specify a "-blocksize" different from the default (256 samples) and "-audiobuf" in milliseconds. Pd will round this down to a power of two buffers, each of "-blocksize" in sample frames.

Using an RME Hammerfall, and specifying "-audiobuf 5 -blocksize 32" I was able to get about 7 milliseconds of throughput delay (as measured by the latency-measurement patch in 7.stuff/tools.) As always, you can specify "-channels" to any even number up to the maximum (32, I think) or can specify channel count separately for input and output (-inchannels and -outchannels).

The special joys of Windows 95

On Windows 95 you can expect a hard time. Every user who tries it seems to encounter a new problem. The best way to run Pd is to get into the "MSDOS Prompt" program and type \pb\bin\pd to it (or whatever the path ends up being.) You can probably put pd's "bin" directory in your path so that you just type "pd" to the prompt.

You don't want to run Pd from the "run" menu because if it fails to start up the window holding the error message will disappear instantly. Ditto for clicking on "batch files" or on the Pd executable itself.

The most common reason Pd might fail to start up in W95 is not having "networking" turned on. Pd is actually two programs that establish an IP interconnection. Beware that this sometimes fools Windows into calling your ISP for no reason.

It is often necessary to specify a huge audio buffer to get steady audio output in W95; see the command line arguments below.

3.3. Linux

What to do depends on which flavor of Linux you are running (e.g., Debian or Red Hat). The instructions here should work for Pd 0.33 and up regardless of your situation. (If not, you can read the Pd mailing list archives for recent problems; if you have found a new problem you're welcome to post it to the list.)

Before you start, you might want to check that you have the resources Pd needs. The main things you need are the C compiler, X windows (including the X development package for Pd to link against) and TK. If you're running Redhat or Mandrake 7.x or up, I think these are all present by default. The RedHat X client developer "RPM" package is called XFree86-devel.

You don't absolutely have to have the X server package running; you can run Pd on the microprocessor in your refrigerator as long as it can connect to an X server on another machine.

If you're running RedHat you might want to use RPM to install Pd. For other linux distributions, download the "tar.gz" version and compile it.

Getting Pd as an RPM

Download Pd, perhaps from , to a file such as "pd-0.33-0.i386.rpm". Open a "shell" window, cd to the directory containing the file, and type the command,

    rpm -i pd-0.33-0.i386.rpm

(substituting the real file name.) Then you should be able to type "pd" to a shell and watch the Pd main window appear.

Getting Pd as a .tar.gz

Download Pd, perhaps from , to file such as "pd-linux-033.tar.gz". Open a "shell" window, cd to the directory containing the file, and type the command,

    zcat pd-linux-033.tar.gz | tar xf -
which creates a directory named "pd". I do this from my home directory. Next, compile it. "CD" to pd and read the INSTALL.txt, or else just cd to "pd/src" and type
make depend

You can pass flags to "configure" to customize your compilation:

    To enable ALSA 0.9x (the latest one), add "--enable-alsa".
    To enable the older ALSA 0.5x, add "--enable-old-alsa". 
    To enable Ritsch's RME 9652 driver, add --enable-rme".
    To put Pd in /usr/bin instead of /usr/local/bin, add "--prefix=/bin".

After "make", just type "~/pd/bin/pd" to run pd.

Alternatively, as superuser, you can run "make install" after "make depend" and then anyone on your system can just type "pd" to run it.

Testing audio and MIDI.

Next try audio. We want to know whether audio output works, whether audio input works, and whether they work simultaneously. First run "aumix" to see audio input and output gains and which device is "recording". Then test audio output by running

    pd -noadc
and selecting "test audio and MIDI" from the "help" menu. You should see a patch. Turn on the test tone and listen. Do the usual where's-the-signal business.

Then quit Pd and test audio input via

    pd -nodac
Re-open the test patch and hit "meter"; look at the levels. 100 dB is a hard clip; arrange gains so that the input signal tops out around 80 or 90, but no higher.

Now see if your audio driver can do full duplex by typing "pd" with no flags. If you see error messages involving /dev/dsp or /dev/dsp2, you're probably not able to run audio in and out at the same time. If on the other hand there's no complaint, and if the audio test patch does what you want, you might wish to experiment with the "-audiobuffer" flag to see what values of audio latency your audio system can handle.

Audio hardware in Linux

Be forewarned: installing and testing audio and MIDI drivers in Linux can take days or weeks. There apears to be no single place where you can get detailed information on Linux audio.

Depending on your hardware and software, you might or might not be able to run "full duplex," i.e., use audio input and output at the same time. For many applications it's important to be able to do this, but if by any chance you don't need simultaneous input and output you will have much less trouble than if you do.

There are two widely-used driver sets, called "OSS" and "ALSA". OSS is included in the standard Linux kernels since version 2.2. However, for some audio cards you can find newer versions than are included in the kernel releases. You can get ALSA from .

(There is also a commercial version of the OSS drivers which costs $30 (slightly more for certain audio cards.) Hit . These might be easier to use than the free OSS drivers, but I've never tried them.)

ALSA is able to emulate OSS, so that you can usually run Pd using the default "OSS" settings even if it's actually ALSA that's running.

Installing and configuring FREE OSS

OSS is really a collection of loadable device drivers. The commands for loading and unloading the drivers are "insmod" and "rmmod". You can see if the audio drivers are running using "lsmod" (as root.) If you see something like:

Module         Pages    Used by
eepro100           3            1 (autoclean)
opl3               3            0
opl3sa2            1            0
ad1848             4    [opl3sa2]       0
mpu401             5    [opl3sa2]       0
sound             15    [opl3 opl3sa2 ad1848 mpu401]    0
soundcore          1    [sound] 6
soundlow           1    [sound] 0
aic7xxx           23            2

then OSS is running, and if all you see is:

eepro100           3            1 (autoclean)
aic7xxx           23            2

then it isn't. You can turn OSS off by running "rmmod" repeatedly, starting with "opl3" (or whatever) so as not to remove any module before you remove all the modules that depend on it. In the above listing, "opl3*" is device dependent and you might see different names.

The file, "/etc/modules.conf" apparently controls which sound drivers are started at boot time. The sndconfig program updates this file but you can also change things manually, for instance to switch between two different sound cards. In Redhat 6.x and earlier, the file is named "conf.modules."

Here is a modules.conf file for OSS:

alias eth0 e100
alias parport_lowlevel parport_pc
alias char-major-81 bttv
alias usb-controller usb-uhci
alias sound-slot-0 i810_audio
alias sound-slot-1 es1371

Here the two sound cards are the (motherboard resident) i810 driver and an ensoniq es1371.

In RedHat at least, the "sndconfig" program tries to automatically search for your soundcard. Unfortunlately it only finds the "first" one which is often not the one you want to use!

Under OSS, programs can stream sound using either "block" or "stream" mode. Stream mode is the more modern and better of the two, but the majority of drivers, even for new sound cards, only support "block." Pd makes "block" the default.

ALSA (Advanced Linux Sound Architecture)

ALSA is newer, hence less stable and harder to use, than OSS. Alsa comes in a "finished" version (0.5.x) and a different, redesigned, "beta" version, 0.9. Installing ALSA can be tricky and/or confusing.

As of version 0.37 Pd works only with 0.9.x versions. The RPM version of Pd is compiled for 0.9.x.

By default, Pd uses OSS. If you are running ALSA, Pd will use ALSA's OSS emulation. To make Pd use ALSA "natively", i.e., the way ALSA is designed to be used, include the "-alsa" flag in the command line.

In ALSA, you can specify which sound card to use using the "-alsadev" flag. So, for instance, "-alsadev 3" means your third card, counting from one. You can also specify it the ALSA way: "-alsadev hw:3,0".

Which sound card?

Here's a rundown on my experiences with sound cards so far. See also the Pd mailing list archives.

RME 9652 (Hammerfall)

This is the best sound card out there; it costs around $500 and has 3 ADAT I/O ports and one SPDIF. There is a "baby hammerfall" also, which I think is the "9632." DO NOT CONFUSE THE 9652/9632 WITH OTHER RME BOARDS WHICH MIGHT NOT WORK WITH PD.

Guenter Geiger has an OSS driver for Hammerfall for 2.4 kernels (such as RedHat 7.1 and up). You have to download and compile it: .

You must then run Pd using the "-32bit' flag, because this uses a non-standard extension of OSS to 32 bit samples.

Hammerfalls now have an ALSA driver; from what I hear it won't work yet with Pd. I was unable to install the ALSA driver on the two machines I tried ("no such device").

Midiman sells devices with between 4 and 12 analog channels in and out, for which there are ALSA drivers. It seems to work fine with the old ALSA driver (0.5). I'm running mine in Alsa 0.9 beta 10. The driver name is "ice1712".

Alsa provides an "envy24control" program (in "utils". You should run this and check that your ice1712's sync source is internal if you have no SPDIF input, or "SPDIF" if you do. I think the default is now "internal" but don't take it for granted...

In RedHat 7.0, motherboards with native i810 audio systems don't work in full duplex (they crash linux). Either run Pd -noadc or else (better) install ALSA.

3.4. Macintosh OSX

Pd version 0.35 and up support Macintosh OSX, although there are still some problems. You can always just download the sources and compile it yourself, or (easier) find a MacOSX-style "package". The first package was put together by Adam Lindsay and can be found on The package simply installs itself and you needn't follow the directions below.
To install on OSX from source:
Whether you've downloaded the source or the "package" you can always compile Pd for yourself, whether to make your own improvements, or possibly so that you can get the newest version before it shows up compiled for Mac OS X.

To be able to compile Pd, you must have installed Tcl/Tk specifically in /Applications/Wish and /Library/Frameworks/Tk.framework and /Library/Frameworks/Tcl.framework.

First download and install TK for OSX ( Get a recent one compiled for OSX, by chasing through "latest file releases", and finding a "download" link for TCL. Then download "TclTkAqua-8.4.1-Jaguar.dmg" (the version number may change). Click on this file and a "folder" will open. Click on the "package" icon in that folder and an auto-installer will put Tcl/Tk on your system.

For old versions of Tcl/Tk you also had to get the "h" files from XFree86 and put them in /usr/X11R6/include. You can download just the H files from:
(the individual files seem to have adequate copyright notices so that I can just redistribute them.) ((I hope this is no longer necessary but I notice people keep downloading these files anyway, so I'll leave them there a while longer until I'm sure they're not needed.))

Then, just as for linux, just unload pd-whatever.tar.gz into a directory such as ~/pd-0.35-test17 , cd to pd-0.36-0/src, type "./configure" and "make". Then type ~/pd-0.36-0/bin/pd to a shell and enjoy!

If you wish you can put the line,

    alias pd ~/pd/bin/pd
in the file, ~/.tcshrc, so that you can later just type "pd" to a shell. (The shell only reads the ~/.tcshrc file on startup, so this won't take effect in any existing shells unless you specially type
    source ~/.tcshrc
to them.)

In some cases you have to explicitly give "-soundindev" and "-soundoutdev" flags for Pd to open audio correctly; "pd -listdev" should show you the correct device numbers.

To get MIDI working, you have to do the Mac OSX magic to get a USB MIDI interface installed. I've seen this done with Midisport devices and I think you just download the OSX driver and follow directions.

On the machine I tried, it was necessary to type,

    pd -midiindev 1 -midioutdev 2
to get MIDI working.

3.5. graphics rendering using GEM

GEM, originally by Mark Danks but now supported by Iohannes Zmoelnig, is essentially an extension of Pd that allows you to do OpenGL programming using a suite of "GEM objects" roughly parallel to the tilde objects built into Pd for audio. Find out more from Johannes's page.

3.6. The Pd command line

Pd is a "command line" program. The best way to run it is from your "terminal emulator," "shell," or "MSDOS prompt." The command line is:

    pd [options] [patches to open]

although you may have to specify a path so your command interpreter can find Pd (OS dependent.) Possible options include:

audio configuration flags:
-r            -- specify sample rate
-inchannels ...  -- number of audio in channels (by device, like "2" or "16,8")
-outchannels ... -- number of audio out channels (by device)
-channels ...    -- specify both input and output channels
-audiobuf     -- specify size of audio buffer in msec
-blocksize    -- specify audio I/O block size in sample frames
-sleepgrain   -- specify number of milliseconds to sleep when idle
-nodac           -- suppress audio output
-noadc           -- suppress audio input
-noaudio         -- suppress audio input and output (-nosound is synonym) 
-listdev          -- list audio and MIDI devices
-audioindev ...  -- sound in device list; e.g., "2,1" for second and first
-audiooutdev ... -- sound out device list, same as above 
-audiodev ...    -- specify both -audioindev and -audiooutdev together

(linux specific audio:)
-frags        -- specify number of audio fragments (defeats audiobuf)
-fragsize     -- specify log of fragment size ('blocksize' is better...)
-stream          -- use stream mode audio (e.g., for es1370 audio cards)
-alsa            -- use ALSA audio drivers
-alsadev      -- ALSA device # (counting from 1) or name: default hw:0,0

MIDI configuration flags:
-midiindev ...   -- midi in device list; e.g., "1,3" for first and third
-midioutdev ...  -- midi out device list, same format
-mididev ...     -- specify -midioutdev and -midiindev together
-nomidiin        -- suppress MIDI input
-nomidiout       -- suppress MIDI output
-nomidi          -- suppress MIDI input and output

general flags:
-path      -- add to file search path
-open      -- open file(s) on startup
-lib       -- load object library(s)
-font         -- specify default font size in points
-verbose         -- extra printout on startup and when searching for files
-d            -- specify debug level
-noloadbang      -- suppress all loadbangs
-nogui           -- suppress starting the GUI
-guicmd "cmd..." -- substitute another GUI program (e.g., rsh)
-send "msg..."   -- send a message at startup (after patches are loaded)
-rt or -realtime -- use real-time priority (needs root privilege)

-frags        -- specify number of audio fragments (defeats audiobuf)
-fragsize     -- specify log of fragment size ('blocksize' is better...)
-blocksize    -- specify audio I/O block size in sample frames
-stream          -- use stream mode audio (e.g., for es1370 audio cards)
-alsa            -- use ALSA audio drivers
-alsadev      -- ALSA device # (counting from 1) or name: default hw:0,0

(MS Windows specific audio:)
-resync           -- resynchronize audio (default if more than 2 channels)
-noresync         -- never resynchronize audio I/O (default for stereo)
-asio             -- use ASIO audio driver (and not the 'MMIO' default)

MIDI configuration flags:
-midiindev ...   -- midi in device list; e.g., "1,3" for first and third
-midioutdev ...  -- midi out device list, same format
-nomidiin        -- suppress MIDI input
-nomidiout       -- suppress MIDI output
-nomidi          -- suppress MIDI input and output

general flags:
-path      -- add to file search path
-open      -- open file(s) on startup
-lib       -- load object library(s)
-font         -- specify default font size in points
-verbose         -- extra printout on startup and when searching for files
-d            -- specify debug level
-noloadbang      -- suppress all loadbangs
-nogui           -- suppress starting the GUI
-guicmd "cmd..." -- substitute another GUI program (e.g., rsh)
-send "msg..."   -- send a message at startup (after patches are loaded)
-rt or -realtime -- use real-time priority (needs root privilege)

Here are some details on some of the audio and MIDI options (but see also the next section on file management.)
sample rate
The sample rate controls Pd's logical sample rate which need not be that of the audio input and output devices. If Pd's sample rate is wrong, time will flow at the wrong rate and synthetic sounds will be transposed. If the output and input devices are running at different rates, Pd will constantly drop frames to re-sync them, which will sound bad. You can disable input or output if this is a problem.
audio buffer size
You can specify an audio buffer size in milliseconds, typically between 10 and 300, depending on how responsive your OS and drivers are. If this is set too low there will be audio I/O errors ("data late"). The higher the value is, on the other hand, the more throughput delay you will hear from the audio and/or control inputs (MIDI, GUI) and the audio coming out.

In Linux and Windows, you can also specify the audio block size in sample frames (but in Windows, this is only effective when using ASIO).

MIDI devices

You can specify multiple MIDI input and output devices. For example, "pd -midiindev 3 -midioutdev 4,2" asks for the third MIDI input device and the fourth and second MIDI output device. The "channel message" midi objects in Pd such as notein or pgmout will take channels 1-16 to mean the first open MIDI port, 17-32 the second one, and so on. The midiin, sysexin, midiout objects give you a separate inlet to specify which of the open MIDI port numbers you want.

In Linux, if you ask for "pd -midioutdev 1" for instance, you get /dev/midi0 or /dev/midi00 (or even /dev/midi). "-midioutdev 45" would be /dev/midi44. In NT, device number 0 is the "MIDI mapper", which is the default MIDI device you selected from the control panel; counting from one, the device numbers are card numbers as listed by "pd -listdev."

3.7. dealing with files

Pd has a search path feature; you specify the path on the command line using the "-path" option. Paths may contain any number of files. If you specify several files in a single "-path" option they're separated by colons in unix or semicolons in NT. When Pd searches for an abstraction or an "extern" it uses the path to try to find the necessary file. The "read" messages to qlists and arrays (aka tables) work the same way. NO SPACES MAY APPEAR ANYWHERE IN THE SEARCH PATH, e.g., "c:\my nonsense\goobers" won't work.

Filenames in Pd are always separated by (unix-style) forward slashes, even if you're on Windows (which uses backslashes). This is so that patches can be ported more easily between operating systems. On the other hand, if you specify a filename on the command line (as in "pd -path c:\pdlib") the file separator should agree with the operating system.

If a filename in a patch has any "/" characters in it, the "path" is not used; thus, "../sounds/sample1.wav" causes Pd only to look relative to the directory containing the patch. (You may also invoke externs that way.)

As of version 0.35, there may be spaces in the path to Pd itself; also, the "openpanel" and "savepanel" objects can handle spaces. But still not the search path.

Powered by IEM Powered by Plone Section 508 WCAG Valid XHTML Valid CSS Usable in any browser