Leading items

Updating and rebuilding Android

This is one in a series of articles on working with the Android Developer Phone (ADP1) device. In the previous episode, your editor went through the process of updating the software on the ADP1. This time around, we'll look at the latest software builds, then take a beginning look at the process of actually building new software for the device.

Your editor started by testing out the Android 1.5 images provided by HTC, the manufacturer of the ADP1. The provision of these images is a nice step forward by HTC; thus far, ADP1 owners have felt somewhat left out when new versions of the firmware have been released. This time around, they have the new software at about the same time as everybody else. The 1.5 update is done in the usual way: use the "adb" tool to copy it to /sdcard/update.zip on the phone, then reboot into the recovery image to actually install the new code. Two such iterations are required this time around; there is an update to the (closed-source) radio code which must be applied first.

So what's new in Android 1.5? The biggest user-visible feature is almost certainly the on-screen keyboard. It's no longer necessary to open the keyboard to send a quick text message. The on-screen keyboard is somewhat painful for your fat-fingered editor to use when the phone is in the portrait orientation, but it works better in the landscape mode. One has to wonder, though, what inspired the Android developers to dedicate a significant chunk of scarce screen space to a "smiley" key. There are plenty of characters which would have been rather more useful in that position.

Beyond that, the 1.5 release includes "Latitudes" support for those of you who want to continuously report your real-world location to the Google mother ship. There are simple screen effects which come into play when switching between applications and orientations. Holding the power button gives quick access to "airplane mode." The camera is quite a bit more responsive. The zoom icons are smaller and more discrete. GPS acquisition is said to be faster; your editor has not had a chance to test that claim, but it would certainly be a welcome improvement. The orientation-awareness (turn the phone on its side and the display switches to landscape mode) that has been a feature of the JesusFreke builds for a while is now part of the core platform. And so on. Mostly small stuff, but it's enough to make for a nicer feel to the platform overall.

Speaking of the JesusFreke builds, the JFv1.50 build, based on 1.5, is also available; your editor promptly installed it. This build is basic Android 1.5 with a number of additions, including multitouch support, tethering support, an augmented init daemon, a whole set of busybox-based command-line utilities, and more; see the full list for the details. As usual, these builds add a number of nice features to the phone; anybody who is interested in really playing with the device will likely prefer the JF version of the software.

Remaking JF builds

Playing with new builds is fun, but this is free software. The real fun comes from rebuilding the software from the source, perhaps with specific changes. There are two levels at which this can be done. The first is to use the JesusFreke "build environments." Essentially, the build environment is a tarball containing the modifications made to create the JF image, along with the necessary scripts. There's a new kernel containing multitouch and unionfs support, along with the patches needed to create it. Busybox is found there, as are a number of other useful diagnostic tools, an ssh client, and more.

To create a new build, it is necessary to get the associated official build, place it within the build environment, then run make. With any luck, the end result is an update.zip file ready to be flashed into the phone.

One of the interesting things your editor learned from looking at the images (and from talking to Mr. Freke) is that the JF builds do not actually involve rebuilding much of the Android system. It's mostly a matter of unpacking an official build and making a few creative substitutions. The kernel has been remade, as has the browser application (to support multitouch zooming). Everything else is just a matter of shuffling files around. So the JF build environments can be useful for somebody else wanting to do the same kind of manipulations, but more extensive changes require building the system at a lower level.

Building Android from source

Remaking Android from the source code turns out to be a bit of a challenge. What follows here is a brief set of instructions derived from the Android "building for Dream" page, some hints helpfully provided by GeunSik Lim, and a fair amount of painful experience. In summary: most of the code needed to rebuild the platform is available, but (1) it's not a quick or simple process, and (2) there are a few pieces missing.

There's a number of tools which must be installed on a Linux system to rebuild the Android platform. These include flex, bison, git, and the Sun Java system. Beyond that, one must grab the repo tool. Repo is Google's answer to the problem of managing a whole set of related repositories; essentially it is a tool which sits on top of git and manages a whole set of git repositories in parallel. Once repo has been installed, the meta-repository is set up with a command like:

    repo init -u git://android.git.kernel.org/platform/manifest.git

This command pulls down the manifest file describing all the repositories needed to build the platform. Note that if a branch other than the trunk is desired, it must be obtained during this stage with the -b option; repo apparently cannot switch branches in an existing source tree.

One then obtains the code by running "repo sync" and going out for coffee.

Incidentally, when you go out for coffee, you need not hurry back. It's entirely possible to fly to a different continent, harvest the coffee by hand (after waiting for it to reach the perfect ripeness), fly home, and roast it yourself. You'll still probably have time for a second cup before the downloading of the source is complete. You are, after all, not just downloading a huge pile of source files. You are, instead, downloading over 100 independent git repositories, each containing a long trail of history - about 2.4GB worth of stuff. It takes a while. And, needless to say, some disk space.

To make things worse, you still don't have all the source; there are a few components of the binary platform for the ADP1 which have not been released as free software. You cannot download those binary components from anywhere; instead, what's needed is to obtain them from a working phone. To that end, the file vendor/htc/dream/extract-files.sh contains a script which will pull the needed components from a USB-connected ADP1 device. These components vary from files containing mixer settings to programs for controlling Bluetooth, the GPS receiver, firmware for the wireless network adapter, a camera control library, and more. The dream directory also contains a binary driver module (wlan.ko) for the WIFI adapter, despite the fact that said driver is open source and included in the distribution.

After that, it's a matter of copying build/buildspec.mk.default to buildspec.mk in the top-level directory, editing it to set TARGET_PRODUCT to htc_dream, and typing make. And going out for more coffee, of course. At the end of the process, with luck (a fair amount of luck may be required), there will be new system and boot images which can be flashed into the phone with the fastboot tool. A reboot will run the new code.

At that point, of course, there are some surprises to be found. One is that the newly built software is lacking a number of features found in an official build. The reason for this is simple: several of the applications which run on Android phones are not open source. These include the Gmail client (which your editor will happily do without), Maps (which cannot be done without, at least until AndNav progresses a little further), and more. These applications can generally be recovered by grabbing the associated package files from an official build and slipping them into the build environment. See this article for a terse description of how that is done.

It took your editor a little while to figure out another little surprise: despite the fact that the Android source tree includes a kernel repository, the build process does not actually build the kernel. One might think that it would be hard to miss something the size of a kernel build, but ... did your editor mention that the Android source tree is big? The Android build system which goes with this source tree is quite a piece of work; there must be people working full time on it, and they probably hate their lives.

Trying to figure out what is happening in an Android build requires digging through many thousands of lines of makefile rules. What your editor finally discovered is that the build system simply pulls a binary kernel from a special "prebuilt" repository (that repository also contains a cross-compiling toolchain for the creation of ARM executables). The kernel source tree, seemingly, is just there for show. Using something other than the prebuilt kernel requires making it separately and pointing a build-system variable at the location of the result.

[PULL QUOTE: It's clear that even people who remake Android are not, in general, expected to remake the kernel. END QUOTE] It's clear that even people who remake Android are not, in general, expected to remake the kernel. The kernel repository pulled down by the repo command does contain the Android-specific patches, but it lacks nice things like branches (even "master" is missing) or tags. There are some remote branches with names like korg/android-msm-2.6.27 which contain lines of development for various kernel versions; the 2.6.27 one appears to be, as of this writing, the one which is best supported on real hardware. But, within those branches, there are (unlike the situation with the rest of the Android code) no tags associated with releases. Nothing in the repository will tell a developer which kernel was shipped with a given version of Android.

So it's hard to build a kernel which corresponds to the one found within an official release. But not impossible: most of the official releases include the git commit ID in the kernel version. So by digging down into the settings menus, your editor was able to determine that the HTC 1.5 build came from commit 8312baf. After checking out that commit, one can do a make msm_defconfig to configure the kernel properly. Then it's just a matter of setting the ARCH and CROSS_COMPILE environment variables and doing the build. If you have a 32-bit Linux environment, the prebuilt ARM toolchain provided with the Android source does the job just fine.

Once the kernel build is done, it's possible to build a new set of firmware images which can be loaded into the device with fastboot. That's easy to say, but it can be harder to do; the sources from the repository often do not build, and it's not always easy to get all the pieces together to make a working image for the ADP1. Making it possible for people outside of the core Android project to build and install the platform appears to be an afterthought, at best.

Android and the ADP1

In truth, Google does not really support the ADP1 as a system people can develop and run on; this situation was somewhat explained by Jean-Baptiste Queru, who is easily the most helpful Google developer on the mailing lists:

So, while the ADP1 is one of the most open cellular phone platforms yet to appear, it does not, yet, represent a fully-open system in the way the OpenMoko phone does. Most people wanting to do things with this device are likely to end up starting with the official, binary builds and tweaking things around the edges, much like as has been done with the JF builds.

That said, there is a lot of fun to be had with this device. It's fully hackable at the kernel level now, and more hardware information is becoming available, which raises the hope of gaining more control over the low-level system in time. About the time the ADP1 becomes fully obsolete and unobtainable, we should have it figured out pretty well. With any luck at all, at least one of the devices which replaces it will be more open from the outset.

Comments (24 posted)

Unladen swallow: accelerating Python

Google uses Python for many of its engineering projects, from internal server monitoring and reporting to outward-facing products like Google Groups, so it is no surprise that the company wants to improve Python application performance. A group of Google developers is working on a new optimization branch of Python dubbed Unladen Swallow, with the goal of a five-fold speed increase over the trunk. It will achieve that goal by adding just-in-time compilation and a new virtual machine design, all while retaining source compatibility for Python application developers.

Unladen Swallow's lead developers Collin Winter, Jeffrey Yasskin, and Thomas Wouters have long been core developers for the CPython project, the reference implementation and most widespread interpreter for the Python language. All three are Google employees, and others contribute their "twenty percent time" to Unladen Swallow, but the group insists that it is a Python project, not an effort owned by Google.

Winter said the origin of the idea dates back to his work on the web-based code review tool Mondrian, when the team's attempts at optimization repeatedly hit limitations in CPython, such as the Global Interpreter Lock (GIL), the mutex that prevents concurrency on multiprocessor or multi-core machines. While researching potential speed-ups and changes, Winter and the other Google engineers eventually decided that the long-range ideas they had in mind were significant enough to warrant making a separate branch. Plus, doing so would give them the chance to stress-test their ideas before trying to roll them back into CPython.

The Concept: a bird's eye view

The core of the Unladen Swallow team's planned improvements is to remove performance bottlenecks in the Python virtual machine (VM) design, leaving the rest of the interpreter — not to mention the substantial runtime library — relatively untouched. The long-term plan is to replace CPython's existing stack-based VM with a register-based VM built with code from the Low Level Virtual Machine (LLVM) project, and to add a just-in-time compiler (JIT) on top of the new VM. Other performance-based improvements are welcome at the same time, and the team has several in store based on their talks with heavy Python users.

Using a JIT will speed up execution by compiling to machine code, thus eliminating the overhead of fetching, decoding, and dispatching Python opcodes. "In CPython," Winter explained, "this overhead is significant; some minor tweaks were made to CPython 2.7 that netted a 15% speed-up with relatively little work."

Adding the JIT presents a good opportunity to switch from a stack-based VM to LLVM's register-based design, which Winter said will net its own performance benefits. The merits of stack- versus register-based VMs is an ongoing debate, but Winter cites a 2005 study [PDF] from the Lua project showcasing the empirical benefits of the register-based design.

Unladen Swallow is based on Python 2.6.1, which is not the most recent release. Python 3.0 was released in December of 2008, implementing the backward-incompatible 3.0 version of Python. Because the majority of Python code in the wild — and in use at Google — is still written for Python 2.x, the Unladen Swallow team decided to focus its efforts on the earlier version where more benefits would be felt. By using the CPython source as its base, Python users can expect Unladen Swallow to retain 100% source compatibility.

Still, Winter said, the team does keep in close contact with Python designer Guido van Rossum (himself a Google employee) and other members of the CPython team. "In our discussions with Guido and others about how and where to merge our changes back into CPython, the idea has been proposed that Unladen Swallow should merge into 3.x. 3.x is the future of the language, and if 3.x is significantly faster than 2.x, that's an obvious incentive to port applications and libraries to 3.x. None of that is set in stone, and Guido may well change his mind."

Recent sightings

The team has set a tight development schedule for Unladen Swallow, making quarterly milestone releases. The first release, 2009Q1, was limited in scope, aiming for a 25 to 35% speed increase over vanilla CPython by making less than drastic changes to the code. The changes include a new eval loop reimplemented using vmgen, several improvements to the garbage collector — better tracking long-lived objects so that the garbage collection can make fewer collection runs — and to the data serialization module cPickle, which the developers said will benefit web applications in particular. Several obscure Python opcodes were also removed and replaced with functionally-equivalent Python functions, which reduces code size without affecting performance.

Unladen Swallow 2009Q1 is available as source code only for the time being, and can be checked out as a branch from the project's public Subversion repository. No specific compilation instructions are provided because this release closely follows the upstream CPython, but the developers do recommend building in 64-bit mode in order to take the fullest advantage of the performance increases.

Since speed of execution is the goal, the team performs regular benchmarks on the code. The thirteen benchmark tests in the suite are based on real-world performance tests designed to highlight practical application tasks, particularly for web applications. The results of the tests on Unladen Swallow 2009Q1 versus CPython 2.6.1 are posted on the project wiki; Unladen Swallow ranges from 7.43% faster to 157.17% faster, beating CPython on every benchmark.

Work is underway now on Unladen Swallow 2009Q2, which will focus on replacing the existing CPython VM with an equivalent built using LLVM.

Elsewhere in the ecosystem

Other open source projects have sought to improve Python application execution using some of the same ideas. Psyco was an earlier JIT for Python, but which was later superseded by the PyPy project. PyPy's primary goal is not performance, though, rather it is to build a Python implementation in Python itself. Stackless Python implements concurrency through the use of its own scheduler and special primitives called "tasklets." Finally, the Parrot project is implementing Python on its own register-based VM.

In some ways, Unladen Swallow is more ambitious than these other projects, particularly when you consider the rapid pace of development laid out in the road map. On the other hand, Unladen Swallow starts from the CPython 2.6.1 code base, and incorporates many CPython developers, which greatly improves the chances that its changes will one day be blessed as the official CPython release. Many of the 2009Q1 changes have already been sent upstream to CPython, and the door is still wide open for the 3.0 series should the JIT and VM performance deliver real-world performance increases anywhere close to the expected 400 percent.

Comments (18 posted)

Tomboy, Gnote, and the limits of forks

Your editor has long been a user of the Tomboy note-taking tool. Tomboy makes it easy to gather thoughts, organize them, and pull them up on demand; it is, beyond doubt, a useful productivity tool. But all is not perfect with Tomboy. Some people have complained about its faults for a while; Hubert Figuiere, instead, chose to do something about it in the form of the Gnote utility. So now, of course, people are complaining about Gnote instead.

So what are the problems with Tomboy? For your editor, at least, the biggest issue is the simple sluggishness of the tool. It is a large program which takes quite some time to start up. If one tries to bring up a note on a system which has been used for other things, there can be a lengthy delay before the menu from the taskbar icon appears. Rightly or wrongly, users blame this fundamental slowness on Tomboy's use of the Mono platform. Now, of course, use of Mono brings in a whole host of other complaints, but we need not consider those here. The simple fact is that Mono adds a great deal of baggage to what should be a small and simple application. A basic note-taking tool should not be a heavyweight program.

Gnote is a reimplementation of Tomboy's functionality using C++ and GTK+. In a sense, though, it is not an independently-developed application. Instead, Gnote is a straightforward port of the original C# code to C++. So it's not just a Tomboy work-alike; it's a true clone. There are advantages to this approach; it makes use of the experience which has gone into Tomboy development, and compatibility with the Tomboy file format is not a problem. This approach enabled Hubert to put together a working application in a relatively short time.

Some distributors (Fedora, at least) have made Gnote packages available. Your editor played with the Rawhide version of Gnote for a bit; it is, in general, indistinguishable from Tomboy. It does seem more responsive, especially when the system is under some memory pressure. Annoyingly, it does not (unlike Tomboy) dismiss notes when the escape key is hit. Beyond that, though, Tomboy users (at least those who do not use plugins) will be hard-put to tell the difference between the two.

It is said that imitation is the sincerest form of flattery; if that is true, one would expect the Tomboy developers to be flattered indeed. But a web log entry about the Tomboy 0.14.1 release made it clear that "flattered" may not be the operative word in the Tomboy camp:

It should not come as a surprise that this language inspired a lot of responses, on the original site and elsewhere. Reading through the discussions, your editor has come to a couple of conclusions:

• The Tomboy development community obviously sees Gnote as a threat. It's not entirely clear why that should be. If these developers are paying attention to what they are doing, and if the Mono platform is as "extremely productive" as they say, they should have no trouble staying ahead of Gnote. Beyond that, the existence of other, interoperable applications should serve to grow the community as a whole.

• Gnote clearly has added value for some users. There is a significant community out there which does not want to have Mono on its systems. One may or may not agree with their reasoning, but one cannot argue with the fact that these users exist; simply dismissing their concerns is unlikely to change their minds. Mono-free, Tomboy-like functionality adds value for those users.

It is evident that some developers and users in the Tomboy camp think that the creation of Gnote is an improper thing to do. The creation of a new application by translating code into another language seems unoriginal at best, and a misappropriation of the code at worst. The fact that the code has been relicensed (from LGPL 2+ to GPLv3) in the translation process has not helped. But it should be remembered that both the translation and the relicensing are acts which are allowed by the license chosen by the Tomboy developers. The LGPL license text packaged with the Tomboy code reads:

Other parts of Tomboy carry GPLv2+ or BSD licenses; it's actually quite a mixture, but they all allow conversion to GPLv3. So Hubert has only done what the original developers explicitly allowed him to do; about the only complaint one might make is that he appears to not have carried the copyright attributions over into his translated code. That, probably, is an omission which needs to be fixed; it would be hard to argue that Gnote is not derived from Tomboy.

Beyond that, there are concerns that Gnote will take developer attention away from Tomboy. That could be true, but chances are that any developers working on Gnote (and it's not clear that there are any, beyond Hubert) are unlikely to have been working on Tomboy previously. There is also concern about what happens if and when Tomboy and Gnote diverge from each other. That could well happen; Hubert does not appear to have promised to mirror Tomboy forever. Should things go that way, Gnote really just becomes another fork; it will live or die on its own merits.

It will take time to know whether hacking on Gnote is a wise use of Hubert's time. But it is a clear example of what is good about free software: a developer with a specific itch was able to scratch it (for himself and for others) without having to start from the beginning. Criticisms of Gnote would seem to be misplaced, and attempts to suppress it (and telling distributors that distributing Gnote is a threat to their users' data can only be seen as such an attempt) even more so. Free software means letting others take your code in directions you may not always approve of.

Comments (120 posted)