Slashdot Mirror

That's almost a flamebait in the original post, because it's so utterly unprincipled, ineffective and inefficient.

• Ineffective: Most importantly, it doesn't actually search a term index, but you can only search for file names (so you have to know already what you're searching for). There is no good desktop search tool for UNIX that I'm aware of (although I've used SWISH-E to index plain text document collections, but that's still different from a tool intended to index whole directory trees for full-text search.
• Inefficient: The find/locate commands don't use an index. People below have proposed updatedb, but I doubt that uses incremental index updating, which can become essential if you run it once per night on a large machine. Full-text indexing is much more resource intensive than just indexing file names, so you want to be even more sure that when tomorrows cron job starts, today's will have finished.
• Unprincipled: You could actually find a pipeline of UNIX system commands that implement full-text indexing and search, but that's not a good way to do it. I am aware of the power and versatility of the pipe paradigm, but search is such a fundamental (pervasive, important) problem that it licenses a dedicated development.

Ideally, there'd be a search engine which is part of the operating system, and Microsoft has recognised this and has been working on it for quite some time now. It will be a major selling point of Longhorn, and I predict it will dramatically enhance Windows usability compared to Linux.

Unfortunately, the open source community has not recognised the problem as a whole, but I'm aware the people on the ReiserFS file system have ambitious future plans to include features in that direction (but that might come too late), and I wouldn't count on the likes of Yahoo/Google to deliver the ultimate UNIX/Linux search solution.

--
Try Nuggets , the first SMS search engine -- text your questions, get your answers from the Web.

= 9J =

I'm also toying with a proposal for Testing Perl, though it may be more of a niche market than Software Development with Perl or Extreme Perl. Maybe not.

There are plenty of programs out there that can work well with just an HTML+JavaScript interface, especially if you have a small database (even a DB_File!) on your machine, and an interpreter for a scripting language like Perl or Python.

I'm curious to see whether it does anything more than Jellybean can... there's something compelling about a tiny local web server with the power of mod_perl and a simple interface that lets you build persistent, network aware applications that can replicate data between clients. With XPCOM, it's certainly possible to write a nicer interface than one that only has HTML Form widgets and some onClick handlers.

--

If you ask Jon Udell, the web services are already here. The latest buzzword advances with XML, SOAP, XML-RPC, and friends are all just further refinement and evolution of the interface. Also, Udell's book, Practical Internet Groupware, talks extensively about adapting existing sites into web services. For example, a site like MetaCrawler demonstrates this in how it uses search engines' HTML "interface" to scoop up search results. Or, take the scripts that query news sites without the benefit of RDF or RSS, parsing HTML to scoop up and aggregate news headlines. These are all primitive web services.

And this is not to mention app servers such as Zope and Frontier, which are already built to offer web services natively. It just seems irresistable to use all of these simple building blocks to create neato keen distributed systems...

If you ask Jon Udell, the web services are already here. The latest buzzword advances with XML, SOAP, XML-RPC, and friends are all just further refinement and evolution of the interface. Also, Udell's book, Practical Internet Groupware, talks extensively about adapting existing sites into web services. For example, a site like MetaCrawler demonstrates this in how it uses search engines' HTML "interface" to scoop up search results. Or, take the scripts that query news sites without the benefit of RDF or RSS, parsing HTML to scoop up and aggregate news headlines. These are all primitive web services.

And this is not to mention app servers such as Zope and Frontier, which are already built to offer web services natively. It just seems irresistable to use all of these simple building blocks to create neato keen distributed systems...

Jon Udell talks a lot about groupware solutions using 'simple' Internet protocols such as NNTP and HTTP. Collaboration tools, examples and discussions of this are available in his new book Practical Internet Groupware - see http://www.oreilly.com/catalog/pracintgr/ for more details.

This may help, and fit in nicely with the move to Linux. Good luck

dj (not related to Jon, just interested in what he has to say).