Skip to main content

Kimberly Blessing

Code Archaeology: Dusting Off the Line Mode Browser

6 min read

Photo of an old IBM computer running the line mode browser, with the explanation page displayedThe line mode browser (LMB) info page, running on the line mode browser.

It's been over a month since I returned from my trip to CERN to participate in the line-mode browser hack days and I'm finally getting my photos and my notes in order. If you haven't read much about the event, Jeremy wrote an excellent summary of the whole experience.

I was on the "coding" team and spent a good deal of time reading through and tracing the logic of the line-mode browser (LMB) source code. It felt like every few minutes I was finding something new or interesting, and pointing it out to the rest of the team, so we could be sure to make note of a feature we needed to implement or something that might be a pitfall. I referred to this work as "code archaeology". Here are some of my more interesting findings:

  • The earliest style sheet (compiled into the browser) had six styles: normal, list, glossary, monospace, address, and heading. (Heading had a further seven nested styles for TITLE and H1-H6.) What changed with each of these styles? Left indent (two levels), right margin, alignment, capitalize states (two, one for each left indent), double spacing, number of blank lines before and number of blank lines after.
  • There's code in place to handle H0 as a heading level. It was mapped to the title heading style -- so if both TITLE and H0 were present, H0 would overwrite what was shown in the TITLE location (at the top right of the screen (see picture).
  • If you've been doing web development for long enough, you'll probably recall the ISINDEX element -- and perhaps, like me, wondered it was supposed to do. If it existed in a page, it added a keyword command to the list of options in the prompt (see picture). It doesn't look like this feature was fully implemented in the LMB -- I don't see any code that would return a list of pages matching the query. In addition, the LMB only recognized ISINDEX if the opening tag was in all caps.
  • Here's an HTML element that was new to most of us: LISTING, which is supposed to render plain text. The LMB code we were looking at had a bug, however: only the opening tag was parsed and any closing </LISTING> tag was left to render in the page (see picture).
  • Another element that we'd never heard of before was HP, for highlighted phrase. HP comes from SGML. It had the effect of forcing text into all caps. The LMB that we reviewed parsed <hp1> and <hp2> only.
  • Because of the way the LMB parses a document, HTML comments (e.g. <!-- -->) are ignored -- even though there were no comments in any of the early HTML documents we reviewed. In fact, any unknown tag is thrown away as a "junk tag", yet its contents (what's between the opening and closing tags) are still rendered to the screen. Thus, a modern web page with a <style> or <script> block would end up showing the user a whole lot of code!

My handwritten notes from my time at CERN My handwritten notes tracing the logic of the LMB, with Tim Berners-Lee's original proposal in the background

I'm not sure that we explicitly discussed it, but it soon became clear that we were building a simulator that we expected could render both current web pages as well as the earliest markup found in the first web site. This led to review of some very old markup, written by Tim Berners-Lee himself, and lots of experimentation with the old IBM system we had with us, which was running the LMB. Some more fun discoveries:

  • The HEAD element doesn't yet appear in HTML -- it would have been parsed out as a "junk tag". That said, in some old HTML documents, you'll find HEADER tags surrounding the TITLE and NEXTID. Fortunately modern browsers handle this old markup pretty well, so we didn't have to code anything special for our simulator.
  • These days, we're used to surrounding paragraphs of text with opening and closing tags... and, back in the day, we just putt hem at the start of paragraphs. Originally <P> tags were used as paragraph separators, to create space between paragraphs. This made the work of writing the simulator CSS, to render both old and new code correctly, a little tricky.
  • The LMB indicated a hyperlink by placing a number in brackets following the link text, e.g. [1]. If your link text was more than one word, it might be confusing or hard to determine what text was part of the hyperlink. That said, the list command would show all of the hyperlinks in a document... but it was only helpful if your URLs were descriptive (see picture). The LMB could be started with flags to disable the showing of hyperlinks, which may have been useful when also running in non-interactive mode -- perhaps when trying to cat contents to files for storage or printing.

You, too, can get the line-mode browser source code and read through the main file (www.c). See what jumps out at you!

As I've expressed before, I'm concerned that many web professionals don't understand the history of the web and thus are doomed to repeat past mistakes. My biggest hope is that anyone who writes code for the web today will spend just a few minutes using the LMB simulator to browse the first web site as well as to check out their own site -- and notice, at the very least, that well-marked up content was and still must be the foundation of the web, in order to have a decent experience, regardless of browser capabilities. Good markup never goes out of style.