Perl & LWP

When direct database access to the information you need isn't available, but web pages with the right data are, you might pursue "screen-scraping" -- fetching a web page and scanning its text for the appropriate pieces of text in order to do further processing. LWP (Library for WWW access in Perl) is a collection of module to help you do this. mir writes: " Perl & LWP is a solid, no-nonsense book that will teach you how to do screen-scraping using Perl. It describes how to automatically retrieve and use information from the web. An introduction to LWP and related modules from simple to advanced uses and various ways to extract information from the returned HTML." Perl & LWP author Sean M. Burke pages 264 publisher O'Reilly and Associates rating 9 reviewer mir ISBN 0596001789 summary Excellent introduction to extracting and processing information from web sites.

The good: The book has a nice style and good coverage of the subject, includes introduction to all the modules used, reference material and includes good, well-developed examples. I really liked the way the authors describe the basic methodology to develop screen-scraping code, from analyzing an HTML page to extracting and displaying only what you are interested in.

The bad: Not much is bad, really. Some chapters are a little dry, though, and sometimes the reference material could be better separated from the rest of the text. The book covers only simple access to web sites; I would have liked to see an example where the application engages in more dialogue with the server. In addition, the appendixes are not really useful. More Info:

If it had not been published by O'Reilly, Perl and LWP could have been titled Leveraging the Web: Object-Oriented techniques for information re-purposing, or Web Services, Generation 0. An even better title would have been Screen-scraping for fun and profit: one day we might all use Web Services and easily get the information we need from various providers using SOAP or REST, but in the meantime the common way to achieve this goal is just to write code to connect to a web server, retrieve a page and extract the information from the HTML. In short, "screen-scraping." This will teach you all about using Perl to get Web pages and extract their "substantifique moëlle" (the pith essence, the essentials) for your own usage. It showcases the power of Perl for that kind of job, from regular expressions to powerful CPAN modules.

At 200 pages, plus 40 pages of appendices and index, this one is part of that line of compact O'Reilly books which covers only a narrow topic in each volume but which covers those topics well. Just like Perl & XML , its target audience is Perl programmers who need to tackle a new domain. It gives them a toolbox and basic techniques that to provide a jump start and avoid many mistakes.

Perl & LWP starts from the basics: installing LWP, using LWP::Simple to retrieve a file from a URL, then goes on to a more complete description of the advanced LWP methods for dealing with forms and munging URLs. It continues with five chapters on how to process the HTML you get, using regular expressions, an HTML tokenizer and HTML::TreeBuilder, a powerful module that builds a tree from the HTML. It goes on with an explanation of how to allow your programs to access sites that require cookies, authentication or the use of a specific browser. The final chapter wraps it all up in a bigger example: a web-spider.

The book is well-written and to-the-point. It is structured in a way that mimics what a programmer new to the field would do: start from the docs for a module, play with it, write snippets of code that use the various functions of the module, then go on to coding real-life examples. I particularly liked the fact that the author often explains the whys, and not only the hows, of the various pieces of code he shows us.

It is interesting to note that going from regular expressions to ever more powerful modules is a path followed also by most Perl programmers, and even by the language itself: when Perl starts being applied to a new domain first there are no modules, then low-level ones start appearing, then, as the understanding of the problem grows, easier-to-use modules are written.

Finally I would like to thank the author for following his own advice by including interesting examples and above all for not including anything about retrieving stock-quotes.

Another recommended book on the subject is Network Programming with Perl by Lincoln D. Stein, which covers a wider subject but devotes 50 pages to this topic and is also very good.

Breakdown by chapter:

• 1. Introduction to Web Automation (15 pages): an overview of what this book will teach you, how to install Gisle Aas' LWP, some interesting words of caution about the brittleness of screen-scraping code, copyright issues and respect for the servers you are about to hammer, and finally a very simple example that shows the basic process of web automation.

  2. Web Basics (16p): describes how to use LWP::Simple, an easy way to do some simple processing.

  3. The LWP Class Model (17p): a slightly steeper read, closer to a reference than to a real introduction that lays out the ground work for the good stuff ahead.

  4. URLs (10p): another reference chapter, this one will teach you all you can do with URLs using the URI module. Although the chapter is clear and complete it includes little explanation as to why you will need to process URLs and it is not even mentioned in the introduction roadmap.

  5. Forms (28p): a complete and easy to read chapter. It includes a long description of HTML form fields that can be used as a reference, 2 fun examples (how to get the number of people living in any city in the US from the Census web site and how to check that your dream vanity plate is available in California) and how to use LWP to upload files to a server. It also describes the limits of the technique. I appreciated a very educative section showing how to go from a list of fields in a form to more and more useful code that queries that form.

  6. Simple HTML processing with Regular Expressions (15p): how to extract info from an HTML page using regexps. The chapter starts with short sections about various useful regexp features, then presents excellent advice on troubleshooting them, the limits of the technique and a series of examples. An interesting chapter, but read on for more powerful ways to process HTML. On the down side, I found the discussion of the s and m regexp modifiers a little confusing.

  7. HTML processing with Tokens (19p): using a real HTML parser is a better (safer) way to process HTML than regexps. This chapter uses HTML::TokeParser. It starts with a short, reference-type intro, then a detailed example. Another reference section describes the methods an alternate way of using the module, with short examples. This is the kind of reference I find the most useful, it is the simplest way to understand how to use a module.

  8. Tokenizing walkthrough (13p) a long Example showing step-by-step how to write a program that extracts data from a web site, using HTML::TokeParser. The explanations are very good, showing _why_ the code is built this way and including alternatives (both good and bad ones). This chapter describes really well the method readers can use to build their code.

  9. HTML processing with Trees (16p): even more powerful than an HTML tokenizer: HTML::TreeBuilder (written by the author of the book) builds a tree from the HTML. This chapter starts with a short reference section, then revisits 2 previous examples of extracting information from HTML using HTML::TreeBuilder.

  10. Modifying HTML with Trees (17p): More on the power of HTML::TreeBuilder: a reference/howto on the modification functions of HTML::TreeBuilder, with snippets of code for each function I really like HTML::TreeBuilder BTW, it is simple yet powerful.

  11. Cookies, Authentication and Advanced Requests (13p): Back to that LWP business... this chapter is simple and to-the-point: how to use cookies, authentication and referer to access even more web-sites. I just found that it lacked a description on how to code a complete session with cookies.

  12. Spiders (20p): a long example describing how to build a link-checking spider. It uses most of the techniques previously described in the book, plus some additional ones to deal with redirection and robots.txt files.

• Appendices

  I think the Appendices are actually the weakest part of the book, most of them are not really useful, apart from the ASCII table (every computer book should have an ASCII table IMHO ;--).

  • A. LWP modules (4p): the list and one line description of all modules in the LWP library, long and impressive! But not very useful,
  • B. HTTP status (2p): available elsewhere but still pretty useful,
  • C. Common MIME types (2p): lists both the usual extension and the MIME type,
  • D. Language Tags (2p): the author is a linguist ;--)
  • E. Common Content Encodings (2p): character set codes,
  • F. ASCII Table (13p): a very complete table, includes the ascii/unicode code, the corresponding HTML entity, description and glyph,
  • G. User's View of Object-Oriented Modules (11p): this is a very good idea. A lot of Perl programmers are not very familiar with OO, and in truth they don't need to be. They just need the basics of how to create an object in an existing class and call methods on it. I found the text too be sightly confusing though, in fact I believe it is a little too detailed and might confuse the reader.
  • Index (8p): I did not think the index was great (code is listed with references to 5 seemingly random pieces of code, type=file, HTML input element is listed twice, with and without the comma...), but this is not the kind of book where the index is the primary way to access the information. The Table of Content is complete and the chapters are focused enough that I have never needed to use the index.

You can purchase Perl & LWP from bn.com. Slashdot welcomes readers' book reviews -- to see your own review here, read the book review guidelines, then visit the submission page.

Comments

Doesn't seem to discuss the legalities

[burgburgburg]

I'm suprised that there isn't any discussion about the potential legal pitfalls in all of this repurposing listed in the contents. I'm not saying that scraping is illegal, but at least a mention of the possible claims and counter-arguments might have been called for.

LWP is great!

[JediTrainer]

Perl's been a wonderful tool in my situation. There's been a situation in my company where we needed to gather data from a (large) supplier, who was unwilling to provide us with a CSV (or otherwise easily parseable) file. Instead, we had to 'log in' to their site, and get the data as an HTML table from the browser.

In one evening, I wrote a quick Perl routine to perform the login and navigation to the appropriate page by LWP, download the needed page, and use REs to extract the appropriate information (yes, traditional screen scrape)

The beauty was that it was easy. I don't usually do Perl, but in this case it proved to be a wonderful tool creation tool :) LWP was a lifesaver here, and that script has worked for over a year now!

Worth learning LWP instead of doing it manually?

[Etcetera]

I've done a whoooole lot of screen-scraping working for a company that shall remain nameless :) and I've generally always used "lynx --source" or curl to download the file and parse/grep it manually.

Can anyone discuss if it's worth it to learn this module and convert HTML the "right" way? Does it provide more reliability, easy of use or deployment, or other spiffiness? Or is it just a bloated Perl module that slaps a layer of indirection onto what is sometimes a very simple task?

Regulare Expressions for HTML?

[Anonymous Coward]

Anyone using RegExps to parse HTML should be shot on sight.

This is so wrong in so many ways.
For starters, you cannot parse dyck-languages with regular expressions.

You *have* to use a proper HTML-parser (that is tolerant to some extent), otherwise your program is simply wrong and I can always construct a proper HTML page that will break your regexp parser.

For those who are really hot on doing information extraction on web pages:
In my diploma thesis I found some methods to extract data from web pages that is resistant to a lot of changes.
I.e. if the structure of a web page changes, you can still extract the right information.
So you can do "screen-scraping" if you really want to, but it should be easier to contact the information provider directly.

Another resource

[merlyn]

In addition to the Perl & LWP book, about half of my 150+ columns have been about LWP in one way or another. Enjoy! (And please support the magazines that still publish me: Linux Magazine and SysAdmin Magazine).

Parsing HTML in Perl

[Animats]

I parse large amounts of difficult HTML (and SGML) inside the Downside system. I do it in Perl, so I've been down this road. A few comments:

• Parsing into a tree is the way to go. HTML is obviously a tree structure, but not many applications parse it into a tree and work on it that way.
• HTML::TreeBuilder is useful, but has problems with badly constructed HTML. I wrote my own parser (SGML::TreeBuilder), which is much more robust, and will parse HTML, SGML, and XML into a tree of HTML::Element reliably without knowing anything about the tags. The key concept is to defer deciding whether a tag needs a matching close tag until you see a close tag for either that tag or an enclosing tag. When you find the close tag, you assume that all the unclosed tags within it didn't need a matching close at all. This minimizes the impact of unclosed tags; for example, if someone fails to close an <I>, it's ignored, rather than putting the entire remaining document in italics. (I really should put that up on CPAN. E-mail me if you want a copy.)
• One of the things you learn from this is that Perl doesn't do "get next character" very well. Some parsers for HTML use a C subroutine(!) to do the low-level tokenizing. It's embarassing that Perl, a string language, does this so badly.
• The output of parsing is a tree of Perl objects. Perl does objects, but not very well. It takes a lot of memory to represent HTML this way.
• High-volume parsing is better done in Java or C++. Perl's regular expressions don't help much, and the weak object system is an obstacle. Doing the whole job in Perl can work, but there's a sizable performance penalty.

Re:Parsing HTML in Perl

[Matts]

Try XML::LibXML instead. It parses HTML, uses a DOM tree, and is all in C code, so uses about twice the memory of your source document, instead of about 8 times for a pure perl DOM.

Big Time Scraping

[deathcow]

I work for a telecom company. You wouldnt believe the scope of devices which require screen scraping to work with. The biggest one that comes to mind that _can_ require it is the Lucent 5ESS telecom switch. While the 5ESS has an optional X.25 interface (for tens of thousands $), our company uses the human-ish text based interface.

Lets say a user (on PC) wants to look up a customers phone line. They pull up IE, go to a web page, make the request into an ASP page, it gets stored in SQL Server.

Meanwhile, a perl program retrieves a different URL, which gives all pending requests. It keeps an open session onto the 5ESS, like a human would. It then does the human tasks, retrieves the typically 1 to 10 page report and starts parsing goodies out of it for return.

More than just 5ESS switches -- DSC telecom switches, some echo cancellers, satellite modems, lots of other devices require scraping to work with.

LWP

[linuxelf]

Our company recently switched from Netscape Mail Server to Exchange 5.5. They then turned off all non-Microsoft protocols, like IMAP and POP, so suddenly, my beutiful Linux machine couldn't get mail, and they were making me switch to Outlook on Windows (ick) However, they did leave webmail as an option for me, woohoo. So, I just broke out my Perl and LWP and now I have a script that quite handily grabs all my mail from the webmail interface and populates it into my standard Unix mail spool. Problem solved.

Great Book, Cool author

[Pinball Wizard]

I started working through this book the day it came out. So far, great book - I used to write bots in C++ of all languages because I had a decent C++ library to work with, but since I switched to Perl and LWP its been a lot easier and more productive. I was able to put some things to work almost immediately retrieving info from sites like amazon.com and abebooks.com(I program for an online bookstore so the examples in this book were very useful) The author happens to live in the same city as me and even stopped by my work to chat for 1/2 hour! Great guy, great book.

In fact I wanted to write a review for this book, but obviously got beaten to the punch. My only wish(2nd edition perhaps) for this book is that it spent a little more time dealing with things like logging into sites, handling redirection, multi-page forms, dealing with stupid HTML tricks that try to throw off bots, etc. But for a first edition this is a great book.

Why HTML::TokeParser?

[Anonymous Coward]

I never liked that module. The tokens it returns are array references that don't even bother to keep similar elements in similar positions, thus forcing you to memorize the location of each element in each token type or repeatedly consult the docs. If you refuse to do event driven parsing, at least use something like HTML::TokeParser::Simple which is pretty cool as it's a factory method letting you call accessors on the returned tokens. You just memorize the method names and forget about trying to memorize the token structures.