Firefox/Projects/Network Error Pages

< Firefox‎ | Projects
Revision as of 22:10, 17 April 2009 by Cbartley (talk | contribs)

Summary

When a requested web page cannot be loaded, Firefox should display an error page that not only describes the problem, but which also presents the user with tools to resolve the issue. It also makes sense to extend this functionality to webserver 404 (File not found) responses as well.

Introduction

When Firefox cannot display a requested web page because of a network error, it shows a network error page. This page describes the problem, but it offers only one “tool” for resolving the problem, the ‘’’Try Again’’’ button. This is of limited utility, in fact, for a large class of errors, this is the one thing guaranteed not to work (cf. misspelled URLs). A better network error page would provide a variety of tools for resolving the problem.

Additionally, web pages that result in 404 errors have a lot in common with network error pages, except that the error page is provided by the server. In many case this is a default 404 error page for the server which is not very helpful. The other major browsers will substitute a browser-provided 404 error page in many cases (the typical rule seems to be not to override server-provided error pages over about 500 bytes in length, since those are likely to be customized). Firefox should provide similar functionality.

Team

cbartley  developer, sprint lead
boriss    UI
johnath   technical advisor
beltzner  Google Link Doctor liaison

Target Release

Firefox.next (Firefox 3.6?)

Status

  • A Places-based suggestions feature and 404 redirection were tested using an extension. It seemed to work well for cases where it made sense at all.
  • Initial page design for DNS Not Found and 404 errors with Places-based suggestions, Search, and Try Again.
  • Entity definitions for the initial page design have been added to "netError.dtd".
  • Now have a C++ implementation of the Levenshtein edit distance algorithm, which is accessible in Places Sqlite queries.
  • 404 errors are intercepted and redirected to the new error page, but only the Places-based suggestions feature is functional.
  • DNS Not Found errors use the new page design, but it’s not functional at all yet.
  • Discussions with the Google Link Doctor team are ongoing.

Near Term Goals

Address just DNS Not Found errors and 404 Errors.

  • Settle on a page design (mostly shared) for DNS Not Found errors and 404 Errors.
  • Get the implementation in good working order.
    • Clean up the 404 interceptor/redirector code and find an appropriate location for it in the source tree.
    • Get Places-based suggestions working for DNS Not Found errors.
    • Add a real search box, using the user’s preferred search engine (but not Wikipedia?).
  • Solicit feedback on functionality, implementation, and page design.
  • Identify an implementation that we can land on trunk in the near term.

Longer Term Goals

  • Add more tools to the new error page design.
    • Google Link Doctor
    • Herdict
    • Other tools TBD
  • Move the other error pages to the new design and unify the implementation.

Possible Tools

  • Retry button
  • Suggested alternatives drawn from Places DB
  • Suggested alternatives drawn from the Google Link Doctor service
  • Suggested alternatives using simple rules of thumb (e.g. “ww.example.com” ? www.example.com)
  • Search box, pre-populated with the failed URL or even anchor text, if available
  • Herdict-based feedback (Herdict provides crowd-source site accessibility data)
  • Other tools, for example the ErrorZilla extension provides Google Cache, Coral Cache, Wayback Machine, Ping, Trace, and Whois.

Caveats, Restrictions, Outstanding Questions

  • Wholesale override of 404 pages would be unpopular, and there is no reliable way to tell the difference between a 404 error page provided by a default install of the web server and a custom 404 error page. The other major browsers seem to look at file size and if it’s below a certain threshold, assume that it’s a default page and overridable. Additionally a Firefox override might be useful even for custom error pages in some circumstances.
  • The old error page, by design, runs without chrome privileges, and the new page must also be chromeless. This is to prevent an XSS attach from escalating into a privilege escalation attack.
  • There are privacy issues around using Google Link Doctor that will need to be addressed if we are to use it automatically.
  • The current 404 error page interception/redirection code just does a document.location.replace on EndDocumentLoad to redirect to the Firefox 404 error page. This has some limitations, and it may be necessary support this functionality at deeper level in the docshell. This should be possible in the Firefox.next timeframe.
  • One click access to the original server-supplied 404 error page might be desirable.
  • We could provide xx

UI Designs

UI designs have been broken up into two categories: official designs, and speculative designs. Official designs are displayed first, in reverse chronological order, so that the current design target is obvious. Speculative designs are below, also in reverse chronological order, since that’s the most likely order they will be discussed in. "Out-of-date" designs may get deleted as work progresses.

Design Questions

  • Is it possible to link to a page with more information about the problem?
  • Should indication that the error is "404" or “DNS Not Found” be provided in the first screen?
  • How many alternatives should be shown, and how closely should those match user input?
  • What distinction should be drawn from the user's perspective between 404 errors and “DNS Not Found” errors?
  • Which tools should be provided?

The Official UI Design

 

  • Unfortunately, there don't seem to be many useful tools we can provide here, beyond a suggestion and a search.
  • It's been noted that having only one suggestion on this page as a bulleted item is visually jarring. Earlier mockups had this - my logic was that using this standard format means there's consistency between this and the 404 page above it, which could be brought to the other warnings. Also, this simple bullet list of "tools" can be easily expanded as new link tools may be added. However, I moved it back in this version and lined up the search box. Basically I thought that with only four elements it was too much to vary the indentation and spacing of them all. So if there were more than one "suggestion" it would go back to the 404 format above, but here there's only one (searching Google), so all is left aligned.
  • Next step: links to a page with an explanation of the error and more details of what might have gone wrong


  This is a design taking into account some recent decisions based on what could be done for the code freeze vs trunk:

  • The middle block of text can't be altered except as a whole piece, and since there's not enough time to make a separate website explaining DNS error, removing it could leave a newer lost - it's kept in this mockup
  • Google search can be added - it's given as a prominent option and the most visually distinctive of the options
  • Try again is still present. I experimented making it a button much like it is now and like "Search" is after Google search, but that didn't really work into the style of a bullet list well

Speculative UI Designs












Overview

Sprint lead: curtis
Sprinters: johnath, boriss

Description:

More useful/helpful set of network error pages that offer tools helping users get to a useful end point via search, analysis of Places DB, or automatic redirects/reloads.

Status

Strings and String Freeze

The strings only patch, bug 484313, missed string freeze. It's been RESOLVED INVALID, and the patch has been moved to the main bug, 482874, per beltzner's request. If the strings patch is to land for 3.5 it will have to go through the exception process. Since the strings are a prerequisite for the new network error pages functionality, it is all effectively gated on the strings exception process.

The downside is that the network error pages functionality proposed here might not make 3.5. The upside is that we have a little more time to work on the UI design. mconnor in particular thinks the existing UI design could benefit from more work.

Implementation

Now have a Firefox patch that directly supports 404 error redirection natively (the 404 prototype extension is no longer needed). This patch also includes a C++ implementation of Levenshtein edit distance, accessible from Sqlite. The alternative URLs query now uses this function. Note that the DNS Not Found error is still using dummy data for alternative URLs although it's using the same new error page as 404 errors (with only slight differences in wording).

Project steps

  • String Freeze for Beta 4: Mar 19, 2009
  • Code Freeze for Beta 4: April 6, 2009

Completed

  • Create experimental 404 extension and send it out for comments

      Completed, comments were:
    • Works well for simulated small errors in URLs
    • Hard to say how well it works for real errors in the wild.
    • Drawing alternatives from the Places DB looks like a practical tool we can use for suggesting alternatives.
  • Discuss possible 404 implementation with bzbarsky and others

      Completed, comments were:
    • 404 interception unified with the existing error handling architecture is too deep a change to make for 3.1.
    • A less intrusive approach similar to the one the experimental 404 extension uses may be possible -- need to investigate further.

In progress

  • Discuss with Google Link Doctor team

      In progress, next: follow up on experimental access to the service.
  • Implement Levenshtein Distance algorithm in C++

      Completed
    • Followup: Plug algorithm into Firefox+Sqlite, providing better performance and accuracy compared to the approach used in the extension.

        In progress
  • Proposed a plan for actual implementation

      In progress (see below)

To do

  • Follow up with contact on Link Doctor team about experimental access to the Link Doctor service.

      To do
  • Extend the 404 extension to intercept other error types so we can use it to prototype improved error page designs.

      To do
  • Figure out how to plumb 404 interception into Firefox proper (talk to Johnath).

      To do
  • Investigate implementing a proper edit-distance algorithm in C++ that can be used with Sqlite.

      To do
  • Come up with a plan for a production-quality Firefox 3.5 patch.

      To do
  • (Boriss) design a 404 & dnserror explanation website

      To do (and decide if it is part of scope)

Proposed Plan

3.5

For 3.5, we're only adding a Google search box to the dnsnotfound page.

Trunk/after 3.5

We're targeting two error types:

  1. 404 errors.
  2. DnsNotFound errors.

Tools to draw on:

  1. Places alternatives: Places DB + Edit Distance -- can be called automatically.
    • Good for errors on URLs that the user has visited before
    • Might be especially useful for misspelled hostnames.
  2. Google Link Doctor
    • Assuming partner issues are worked out with Google.
    • Can't be called automatically for privacy reasons.
    • Available with one click (similar to pre-loaded Google search box)
  3. Other easy to implement tools
    • common transformations (www..example.com --> www.example.com, etc.)
    • pre-loaded search box
    • etc.

404 redirection

Design scope

  • Will use 404-specific error page for 3.1 rather than trying to re-use the generic network error page. The goal is to give the user better indication of the problem and provide tools for next steps if needed.
  • Will not redirect obvious custom server error pages. To this end, we'd like to use a heuristic beyond simple page size for identifying default 404 pages. Possibilities:
    1. Compare to standard Apache and IIS 404 pages, possibly fuzzy (use edit istance?)
    2. Use EndDocumentLoad (or other event) + redirect
    3. Implement listener in C++, should be able to access 404 status directly from dcument.
    4. Deep docshell changes (Not possible for 3.1}
  • Will allow original error page to be available at one click
  • Will unify with existing error page architecture in v.next.

Open questions

  • Which tools will be provided?
  • Should indication that the error is "404" be provided in the first screen?
  • Is it possible to link to a page with more information about the problem?
  • How many alternatives should be shown, and how closely should those match user input?
  • What distinction should be drawn from the user's perspective between this and a DnsNotFound error?
  • Problem: The 404 error page can't be a chrome page, since that could expose us to privilege escalation attacks piggy-backed on XSS attacks.

      Solution: Use an about: URL to access the 404 error page.


Resources

Sketches

Notes

Most current design

 

  • Offers the best suggestion prominently
  • Links to the domain
  • Searches the domain for the directory
  • Offers a Google search box filled with the text entry
  • Next step: links to a page with an explanation of the error and more details of what might have gone wrong


DnsNotFound

Design scope

  • Will use the DnsNotFound error page to give the user better indication of the problem and provide tools for next steps if needed
  • Will provide the user tools via one-click access to Google Link Doctor
  • Will provide the user with better link options with data drawn from Places

Open questions

  • Which tools will be provided?
  • Should indication that the error is "DnsNotFound" be provided in the first screen?
  • Is it possible to link to a page with more information about the problem?
  • How many alternatives should be shown, and how closely should those match user input?
  • What distinction should be drawn from the user's perspective between this and a 404 error?

Resources

Sketches

Notes

Most current design

 

  • Unfortunately, there don't seem to be many useful tools we can provide here, beyond a suggestion and a search.
  • It's been noted that having only one suggestion on this page as a bulleted item is visually jarring. Earlier mockups had this - my logic was that using this standard format means there's consistency between this and the 404 page above it, which could be brought to the other warnings. Also, this simple bullet list of "tools" can be easily expanded as new link tools may be added. However, I moved it back in this version and lined up the search box. Basically I thought that with only four elements it was too much to vary the indentation and spacing of them all. So if there were more than one "suggestion" it would go back to the 404 format above, but here there's only one (searching Google), so all is left aligned.
  • Next step: links to a page with an explanation of the error and more details of what might have gone wrong

  This is a design taking into account some recent decisions based on what could be done for the code freeze vs trunk:

  • The middle block of text can't be altered except as a whole piece, and since there's not enough time to make a separate website explaining DNS error, removing it could leave a newer lost - it's kept in this mockup
  • Google search can be added - it's given as a prominent option and the most visually distinctive of the options
  • Try again is still present. I experimented making it a button much like it is now and like "Search" is after Google search, but that didn't really work into the style of a bullet list well





References

Notes

Can we use Places autocomplete for link correction? A mis-typed link will often be in error in only the last couple of characters -- we can truncate the last 2 or 3 characters and rely on Places auto-complete to give us recommendations...

http://mxr.mozilla.org/mozilla-central/source/toolkit/components/places/tests/autocomplete/

A related alternative would be to construct an edit-distance function and register it with SqlLite. We can then do a query on minimal edit distance (async of course) to get the best link correction matches.

Problem is, how do we hook link correction into the error page? We'd like to provide suggestions inside the web page similar to Google's "did you mean" on the search results page.

Google Chrome LinkDoctor Service (linkhelp.clients.google.com)

 wget -U "Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US) AppleWebKit/525.13 (KHTML, like Gecko) Chrome/0.2.149.27 Safari/525.13" "http://linkhelp.clients.google.com/tbproxy/lh/fixurl?sourceid=navclient%20&hl=en&sd=com&error=http404&url=http://news.speeple.com/sunflowers"
 
 wget -U "Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US) AppleWebKit/525.13 (KHTML, like Gecko) Chrome/0.2.149.27 Safari/525.13" "http://www.google.com"