The reports of Polypaudio's death are greatly exaggerated.

We are proud to announce the release of Polypaudio 0.8, our networked sound daemon for Linux, other Unix-like operating systems, and Microsoft Windows. Since the last official release, 0.7, more than a year has passed. In the meantime Polypaudio experienced major improvements. Major contributions have been made by both Pierre Ossman and me. Pierre is being payed by Cendio AB to work on Polypaudio. Cendio distributes Polypaudio along with their ThinLinc Terminal Server.

Some of the major changes:

• New playback buffer model that allows applications to freely seek in the server side playback buffer (both with relative and absolute indexes) and to synchronize multiple streams together, in a way that the playback times are guaranteed to stay synchronized even in the case of a buffer underrun. (Lennart)
• Ported to Microsoft Windows and Sun Solaris (Pierre)
• Many inner loops (like sample type conversions) have been ported to liboil, which enables us to take advantage of modern SIMD instruction sets, like MMX or SSE/SSE2. (Lennart)
• Support for channel maps which allow applications to assign specific speaker positions to logical channels. This enables support for "surround sound". In addition we now support seperate volumes for all channels. (Lennart)
• Support for hardware volume control for drivers that support it. (Lennart, Pierre)
• Local users may now be authenticated just by the membership in a UNIX group, without the need to exchange authentication cookies. (Lennart)
• A new driver module module-detect which detects automatically what local output devices are available and loads the needed drivers. Supports ALSA, OSS, Solaris and Win32 devices. (Lennart, Pierre)
• Two new modules implementing RTP/SDP/SAP based multicast audio streaming. Useful for streaming music to multiple PCs with speakers simultaneously. Or for implementing a simple "always-on" conferencing solution for the LAN. Or for sharing a single MIC/LINE-IN jack on the LAN. (Lennart)
• Two new modules for connecting Polypaudio to a JACK audio server (Lennart)
• A new Zeroconf (mDNS/DNS-SD) publisher module. (Lennart)
• A new module to control the volume of an output sink with a LIRC supported infrared remote control, and another one for doing so with a multimeda keyboard. (Lennart)
• Support for resolving remote host names asynchronously using libasyncns. (Lennart)
• A simple proof-of-concept HTTP module, which dumps the current daemon status to HTML. (Lennart)
• Add proper validity checking of passed parameter to every single API functions. (Lennart)
• Last but not least, the documentation has been beefed up a lot and is no longer just a simple doxygen-based API documentation (Pierre, Lennart)

Sounds good, doesn't it? But that's not all!

We're really excited about this new Polypaudio release. However, there are more very exciting, good news in the Polypaudio world. Pierre implemented a Polypaudio plugin for alsa-libs. This means you may now use any ALSA-aware application to access a Polypaudio sound server! The patch has already merged upstream, and will probably appear in the next official release of alsa-plugins.

Due to the massive internal changes we had to make a lot of modifications to the public API. Hence applications which currently make use of the Polypaudio 0.7 API need to be updated. The patches or packages I maintain will be updated in the next weeks one-by-one. (That is: xmms-polyp, the MPlayer patch, the libao patch, the GStreamer patch and the PortAudio patch)

A side note: I wonder what this new release means for Polypaudio in Debian. I've never been informed by the Debian maintainers of Polypaudio that it has been uploaded to Debian, and never of the removal either. In fact I never exchanged a single line with those who were the Debian maintainers of Polypaudio. Is this the intended way how the Debian project wants its developers to communicate with upstream? I doubt that!

How does Polypaudio compare to ESOUND?

Polypaudio does everything what ESOUND does, and much more. It is a fully compatible drop-in replacement. With a small script you can make it command line compatible (including autospawning). ESOUND clients may connect to our daemon just like they did to the original ESOUND daemon, since we implemented a compatibility module for the ESOUND protocol.

Support for other well known networked audio protocols (such as NAS) should be easy to add - if there is a need.

For a full list of the features that Polypaudio has over ESOUND, see Polypaudio's homepage.

How does Polypaudio compare to ALSA's dmix?

Some people might ask whether there still is a need for a sound server in times where ALSA's dmix plugin is available. The answer is: yes!

Firstly, Polypaudio is networked, which dmix is not. However, there are many reasons why Polypaudio is useful on non-networked systems as well. Polypaudio is portable, it is available not just for Linux but for FreeBSD, Solaris and even Microsoft Windows. Polypaudio is extensible, there is broad range of additional modules available which allow the user to use Polypaudio in many exciting ways ALSA doesn't offer. In Polypaudio streams, devices and other server internals can be monitored and introspected freely. The volume of the multiple streams may be manipulated independently of each other, which allows new exciting applications like a work-alike of the new per-application mixer tool featured in upcoming Windows Vista. In multi-user systems, Polypaudio offers a secure and safe way to allow multiple users to access the sound device simultaneously. Polypaudio may be accessed through the ESOUND and the ALSA APIs. In addition, ALSA dmix is still not supported properly by many ALSA clients, and is difficult to setup.

A side node: dmix forks off its own simple sound daemon anyway, hence there is no big difference to using Polypaudio with the ALSA plugin in auto-spawning mode. (Though admittedly, those ALSA clients that don't work properly with dmix, won't do so with our ALSA plugin as well since they actually use the ALSA API incorrectly.)

How does Polypaudio compare to JACK?

Everytime people discuss sound servers on Unix/Linux and which way is the right to go for desktops, JACK gets mentioned and suggested by some as a replacement for ESOUND for the desktop. However, this is not practical. JACK is not intended to be a desktop sound server, instead it is designed for professional audio in mind. Its semantics are different from other sound servers: e.g. it uses exclusively floating point samples, doesn't deal directly with interleaved channels and maintains a server global time-line which may be stopped and seeked around. All that translates badly to desktop usages. JACK is really nice software, but just not designed for the normal desktop user, who's not working on professional audio production.

Since we think that JACK is really a nice piece of work, we added two new modules to Polypaudio which can be used to hook it up to a JACK server.

Get Polypaudio 0.8, while it is hot!

BTW: We're looking for a logo for Polypaudio. Feel free to send us your suggestions!

Update: The Debian rant is unjust to Jeff Waugh. In fact, he had informed me that he prepared Debian packages of Polypaudio. I just never realized that he had actually uploaded them to Debian. What still stands, however, is that I've not been informed or asked about the removal.

In response to mjg59's rant about controlling the LCD brightness on laptops, I invested some time to reverse engineer the Windows driver of my MSI S270 laptop which implements changing LCD brightness. It requires some ugly fiddling with registers of the "embedded controller" on ports 0x62 and 0x66. The result of my work is s270ctrl, a small userspace utility. I'm planning to turn this into a proper kernel module soon.

If you have access to the current issue (03/06) of the german Linux Magazin make sure to read the two extensive articles about Avahi (p.64 and p.90). Daniel S. Haischt wrote the second article, I wrote the other. Both are a worthy read!

Due to the great work of Sebastien Estienne there are now Debian packages of mod_dnssd and mod_mime_xattr available from my little Debian package repository. They've been uploaded to Ubuntu as well, but we are still looking for some Debian developer who would be willing to upload them to Debian proper. Feel free to contact me if you are interested!

I updated my little Apache module mod_mime_xattr to be compatible with Apache 2.0.

What is it useful for? Linux (2.4 with patch, 2.6 out-of-the-box) has been supporting extended attributes for files (EAs) for ages, but very few applications use them. To change that I wrote a small module for Apache which interpretes the EA user.mime_type and uses its value as MIME type for all files served by Apache. The EA has been standardized by the XDG MIME system, but apparently neither Gnome nor KDE support it right now.

Usage of mod_mime_xattr is simple. To enable interpretation of the EA on the entire tree use something like this in your Apache configuration file:

<Directory />
XAttrMimeType On
</Directory>

That's all that is required to make use of user.mime_type on all files where it is set. To set the EA use a command like this one:

setfattr -n "user.mime_type" -v "text/html" foo.txt

And foo.txt will become a file with the MIME type of text/html, although its suffix is .txt!

The first release of mod_dnssd is now available. It adds DNS-SD based Zeroconf support to Apache 2.0 using Avahi.

This work has been inspired by Sander Temme's and Sebastien Estienne's mod_zeroconf module, but supersedes it in every way. MacOSX ships with mod_rendezvous/mod_bonjour, but mod_dnssd is much more powerful than this piece of software as well. In short: mod_dnssd is definitely the greatest way to add Zeroconf support to Apache available today.

A few examples just to show how great mod_dnssd is:

DNSSDEnable On

This is everything you need to enable DNS-SD support in Apache after loading the module. It will publish all virtual hosts and all existing mod_userdir directories (i.e. ~/public_html) as services of type _http._tcp.

In case you want to publish some subdirectory of the web server as service, just place DNSSDServiceName inside a <Location> section for that path:

<Location /foobar>
	DNSSDServiceName "A special service called foobar"
</Location>

You can even use it to publish WebDAV shares using Apache's mod_dav module:

<Location /webdav>
	Dav On
	DNSSDServiceName "A WebDAV folder"
	DNSSDServiceTypes _webdav._tcp
</Location>

This especially cool since we now have a free software server counterpart for Gnome's and KDE's WebDAV client functionality.

Or to publish your blog as RSS service:

<Location /blog.cgi?rss>
	DNSSDServiceName "The blog"
	DNSSDServiceTypes _rss._tcp
</Location>

Get it while it is hot!

A few days ago we relased Avahi 0.6.3. This is an important bugfix release, everyone should update as soon as possible.

Avahi now has its own domain avahi.org and finally has a logo, thanks to the great work of Mathieu Drouet:

Avahi has moved from Debian Experimental to Unstable. Ubuntu moved it from Universe to Main since it successfully passed their security auditing. The Fedora Core development distribution contains it too, as does SuSE's and Gentoo's. But where's Mandriva? Apparently they are considering it, for whatever it is worth. FreeBSD Ports has it too. I guess this means that Avahi has now been accepted by all major distributions. Hurrah!

It's impressing how easy it is to draw fractals with Python. Using the ubercool Python Imaging Library and native complex number support in Python you can code an elaborate and easy to understand fractal generator in less than 50 lines of code:

#!/usr/bin/python
import Image, ImageDraw, math, colorsys

dimensions = (800, 800)
scale = 1.0/(dimensions[0]/3)
center = (2.2, 1.5)       # Use this for Mandelbrot set
#center = (1.5, 1.5)       # Use this for Julia set
iterate_max = 100
colors_max = 50

img = Image.new("RGB", dimensions)
d = ImageDraw.Draw(img)

# Calculate a tolerable palette
palette = [0] * colors_max
for i in xrange(colors_max):
    f = 1-abs((float(i)/colors_max-1)**15)
    r, g, b = colorsys.hsv_to_rgb(.66+f/3, 1-f/2, f)
    palette[i] = (int(r*255), int(g*255), int(b*255))

# Calculate the mandelbrot sequence for the point c with start value z
def iterate_mandelbrot(c, z = 0):
    for n in xrange(iterate_max + 1):
        z = z*z +c
        if abs(z) > 2:
            return n
    return None

# Draw our image
for y in xrange(dimensions[1]):
    for x in xrange(dimensions[0]):
        c = complex(x * scale - center[0], y * scale - center[1])

        n = iterate_mandelbrot(c)            # Use this for Mandelbrot set
        #n = iterate_mandelbrot(complex(0.3, 0.6), c)  # Use this for Julia set

        if n is None:
            v = 1
        else:
            v = n/100.0

        d.point((x, y), fill = palette[int(v * (colors_max-1))])

del d
img.save("result.png")

Some example pictures:

 .

I am doing a lot of embedded Linux work lately. The machines we use configure their hostname depending on some external configuration options. They boot from a CF card, which is mostly mounted read-only. Since the hostname changes often but we wanted to use sudo we had a problem: sudo requires the local host name to be resolvable using gethostbyname(). On Debian this is usually done by patching /etc/hosts correctly. Unfortunately that file resides on a read-only partition. Instead of hacking some ugly symlink based solution I decided to fix it the right way and wrote a tiny NSS module which does nothing more than mapping the hostname to the IP address 127.0.0.2 (and back). (That IP address is on the loopback device, but is not identical to localhost.)

Get nss-myhostname while it is hot!

BTW: This tool I wrote is pretty useful on embedded machines too, and certainly easier to use than setterm -dump 1 -file /dev/stdout | fold -w 80. And it does color too. And looping. And is much cooler anyway.

Avahi 0.6 is now officially released. Get it while it is hot!

Read the announcement.

In related news: I prepared a patch for distcc that adds Zeroconf support using Avahi.