One of the advantages to being associated with Google is that we have access to a lot of information about the Web, and a lot of computers to test on. About once an hour, our distributed test infrastructure takes the very latest version of Google Chrome in development and uses it to automatically load a large number of the pages that Google has seen are most popular around the world. When it's done, it produces a report like this on the Buildbot waterfall that all our developers (and anyone else) can see:
Here the final test got through a bit over 100,000 pages before stopping to make way for the next build to be tested. And before each Dev, Beta, or Stable channel release, we run with a much larger number of URLs.
In addition, we "fuzz-test" the user interface, automatically performing arbitrary sequences of actions (opening a new tab, pressing the spacebar, opening various dialogs, etc. — a total of more than 30 possible actions). These are also run in our distributed testing architecture, so we can exercise thousands of combinations for each new version of Google Chrome in progress. The same report that shows the page-load results above collects these UI test results too:
This sort of large-scale testing is great for finding crashes that happen only rarely, or that only affect pages that developers wouldn't have visited as part of their haphazard manual testing. By catching a problem right away even if it's very rare, it's easier for developers to figure out what change caused the error and fix it before it ever gets close to showing up in Google Chrome itself.
Posted by Pamela Greene and Patrick Johnson, Software Engineers
The Hunspell dictionary maintainers have done a great job creating high-quality dictionaries that anybody can use, but one of the problems with any dictionary is that there are inevitably omissions, especially as new words appear or proper nouns come into common use. We at Google are in a good position to use our knowledge of the internet to identify and fix some of these omissions. The Google translation team used their language models to generate a sorted list of the most popular words in each language. This was cross-checked with the Hunspell dictionaries to generate a list of the top 1000 words not present in each dictionary. This list includes many popular words, but also common misspellings. To remove these words, each list was reviewed by specialist in that language. Generally, we tried to keep proper nouns and even foreign words as long as they were in common usage.
We hope that by using the the existing GPL/LGPL/MPL tri-license for our addition, our work can be picked up by other users of Hunspell. We also hope to make more improvements in the future, both for additional languages like Turkish, and to refine the word lists we already have. If you're passionate about your language, you can help out by writing affix rules for the added words or reviewing more word lists.
The recent dev-channel release of Google Chrome (2.0.160.0) has the additional words we generated for 19 of the languages. Hopefully, you'll see fewer common words marked as misspelled. For example, the English dictionary now includes "antivirus," "anime," "screensaver," and "webcam," and commonly used names such as "BibTeX," "Mozilla," "Obama," and "Wikipedia." For our scientific users, we even have "gastroenterology," "oligonucleotide," and "Saccharomyces"! We'd like to give special thanks to the great help we got from the translation team who generated the words and the language search specialists who reviewed the lists.
Posted by Brett Wilson and Siddhartha Chattopadhyay, Software Engineers
While the V8 team has been working hard to improve JavaScript performance, one part of the language that we have so far not given much attention is regexps. Our previous implementation was based on the widely used PCRE library developed by Philip Hazel at the University of Cambridge. The version we used, known as JSCRE, was adapted and improved by the WebKit project for use with JavaScript. Using JSCRE gave us a regular expression implementation that was compatible with industry standards and has served us well. However, as we've improved other parts of the language, regexps started to stand out as being slower than the rest. We felt it should be possible to improve performance by integrating with our existing infrastructure rather than using an external library. The SquirrelFish team is following a similar approach with their JavaScript engine.
A fundamental decision we made early in the design of Irregexp was that we would be willing to spend extra time compiling a regular expression if that would make running it faster. During compilation Irregexp first converts a regexp into an intermediate automaton representation. This is in many ways the "natural" and most accessible representation and makes it much easier to analyze and optimize the regexp. For instance, when compiling /Sun|Mon/ the automaton representation lets us recognize that both alternatives have an 'n' as their third character. We can quickly scan the input until we find an 'n' and then start to match the regexp two characters earlier. Irregexp looks up to four characters ahead and matches up to four characters at a time.
After optimization we generate native machine code which uses backtracking to try different alternatives. Backtracking can be time-consuming so we use optimizations to avoid as much of it as we can. There are techniques to avoid backtracking altogether but the nature of regexps in JavaScript makes it difficult to apply them in our case, though it is something we may implement in the future.
During development we have tested Irregexp against one million of the most popular webpages to ensure that the new implementation stays compatible with our previous implementation and the web. We have also used this data to create a new benchmark which is included in version 3 of the V8 Benchmark Suite. We feel this is a good reflection of what is found on the web.
If you want to try this out, and help us test it in the process, you can subscribe to the dev-channel and if you see problems that might be related to Irregexp consider filing a bug.
And BTW, we'll have sessions on V8 and other Chrome-related topics in May at Google I/O, Google's largest developer conference.
Posted by Erik Corry, Christian Plesner Hansen and Lasse Reichstein Holst Nielsen, Software Engineers
"To achieve the streamlined feel we were after … we had our own intuitions about what was and wasn't useful in current browsers, we had no idea how those ideas matched to reality. So in typical Google fashion, we turned to data; we ran long studies of the browsing habits of thousands of volunteers, compiled giant charts of what features people did and didn't use, argued over and incorporated that data into our designs and prototypes, ran experiments, watched how our test users reacted, listened to their feedback, and then repeated the cycle over and over and over again."
To provide some more insight into this process, I should explain what we mean by "data." The data we turn to is both quantitative and qualitative. Usage logs provide statistics such how many users have tried a feature and how frequently a feature gets used. These logs are collected only from people who have chosen to share usage statistics with us. This quantitative data tells us the "how" and the "when" but not the "why." For that, we use qualitative data gathered through research methods like surveys, interviews and contextual inquiry which involves observing people in their home or work environments. Often we bring people to one of our usability labs where we can observe their interactions and collect feedback on a new feature we are working on. Many times we employ an eye tracker where we can find out what exactly people are looking at on our user interface. By incorporating data from all these sources into our design process, we hope to provide a user experience that satisfies the needs of the many Google Chrome users out there.
In the future, we are planning on releasing some of our research on this blog and the on the UX Site to show how the data we are collecting is impacting the Chrome experience.
All of our research data comes from studying and observing people. But what kind of "people" do I mean? Probably someone just like you. So if you are interested in becoming a potential participant in a research study on Chrome or one of the many other Google products, I encourage you to sign up at google.com/usability.
Because we don't have those big Dot-Oh release milestones on the calendar, we don't have long periods of Beta testing new features. Instead we use automatic update channels to release Google Chrome to a community of early adopters. The channels are essentially a never-ending Beta test and a continuous feedback loop that lets us rapidly develop new ideas into solid product features.
You can subscribe to one of our update channels:
Stable channel. Everyone is on the Stable channel when they first install Google Chrome. The Stable channel is updated with features and fixes once they have been throughly tested in the Beta channel. If you want a rock solid browser but don't need the latest features, the Stable channel is for you.
Beta channel. People who like to use and help refine the latest features subscribe to the Beta channel. Every month or so, we promote stable and complete features from the Dev channel to the Beta channel. The Beta channel is more stable than Dev, but may lack the polish one expects from a finished product.
Developer preview channel. The Dev channel is where ideas get tested (and sometimes fail). The Dev channel can be very unstable at times, and new features usually require some manual configuration to be enabled. Still, simply using Dev channel releases is an easy (practically zero-effort) way for anyone to help improve Google Chrome.
To get more actively involved with Google Chrome, subscribe to the Dev or Beta channel. Just run a little program (found here) and that's it. After that, you'll automatically get early access updates.
If you're ready to try some new stuff, we've just released a Dev channel update that has a new version of WebKit, a new network stack, and some features like form autocomplete (read about it here). It's less polished than what Dev channel users have been getting during Google Chrome's Beta, so we've moved all of our existing Dev channel users to the Beta channel. If you were on the Dev channel, you can decide whether to switch to the new Dev channel or stay on the new Beta channel.
