Skip to content

Undetectables

Jump to bottom
jdalton edited this page May 13, 2011 · 80 revisions

The Undetectables

These things cannot be feature detected. They must rely on either:

  1. UA sniffing
  2. Browser inferences
  3. Providing the same (slow) fix to everyone

Categories of Undetectables:

HTML5

  • registerProtocolHandler() support (existed in webkit but doesnt do anything, still live in iOS/Android likely.)
  • File Drag 'n Drop (webkit doesnt expose a global DataTransfer constuctor to inspect) issue 57
  • webforms UI: datepicker, colorpicker UIs cannot be detected, only the associated constraint validation
  • inability to edit things in contenteditable/designMode in mobile Webkit browsers

Audio/Video

  • timeupdate video/audio event fires inconsistently across webkit and moz
  • HTML5 video readyState is inconsistent across webkit(4) and moz(3)
  • canPlayType() results on mobile are unreliable
  • preload attribute support, can't be tested without bandwidth-consuming tests (Chrome has implemented the preload-IDL, but not the feature)
  • loop attribute support, can't be tested without bandwidth-consuming tests (support for this attribute is practically non-existent)

Networking

  • script tag onload/onerror/onreadystatechange actual behaviour (atomicity, order, etc etc)
  • link tag / stylesheet onload actual behaviour (exists in webkit / ff but never fires)
  • Network related metrics, a la the Browserscope Networking Results (|| script, css, etc)

Typography / Text

  • Support for &shy; or <wbr>
  • ClearType disabling during opacity changes
  • FOUT behavior of Gecko's @font-face support
  • font rendering (like smoothing, anti-aliasing, font-weight diffs among browsers/platforms)
  • consistent rendering ( in the page and alert() dialogs) of HTML entities
  • proper handling of CSS text-overflow:ellipsis behavior on text

Events

  • keyboard event bugs, especially related to the keydown/keypress events, and the keyCode/charCode event properties.
  • the order (and reliability) of mousedown, click, and mouseup events (especially if a mousedblclick is happening)
    • especially when fired on a form element like a checkbox, reliability of state update of the element compared to when the events fire
  • whether a scroll wheel event is coming from a continuous device like a trackpad or the Magic Mouse or from a traditional clicking scroll wheel (in webkit).
  • onhashchange is case insensitive in IE8 (maybe 9 too?), all others are case sensitive (HTML5 requires case sensitivity). Can't be detected without actually triggering a hashchange event.
  • IE8 in compatibility view mode has onhashchange, but doesn't support it. Workaround is to test for document.documentMode && document.documentMode >= 9.
  • support for DOMContentLoaded.
    • You can add an event listerner and hope, fall back to document.readyState or window.onload or in IE, the doScroll hack.
  • submit + required attribute for form inputs: browsers that support this HTML5 attribute can cancel the form submit if required values are missing, but the "submit" event still bubbles up (Opera, Firefox 4) and it can't be detected that preventDefault() is called on it

CSS

  • <select> display in IE6
  • PNG alpha transparency
  • The look of the various *-resize cursors is very different across browsers and platforms.
  • ie6 css bugs like duplicate characters bug
  • IE display transparency bug on height > 4096px
  • CSS Generated Content support
  • False positives for box-shadow, border-radius, etc (happens in Aurora/Midori/etc) issue 169
  • pointer-events false positive in Opera
  • position:fixed support in iOS and Android

Edge cases

  • IE CSS bug where certain types of positioning of an <input type=text> element can cause that element's autocomplete box to not render at the same location as the actual input box.
  • strange treatment of <!-- --> comment nodes in the DOM (in IE)... like in some cases, a comment node will be moved to be a child of the previous element even though it's not a child proper. Can cause things like :empty to fail differently.
  • resize event firing rate. e.g. you can't keep an element's position smoothly updated in response to resize or scroll.

General undetectables

  • Memory leaks
  • DOM/JS performance (e.g.: innerHTML vs. DOM manipulation)
    • Testable, but typically not feasible to test just to avoid code forking.
  • Hardware Acceleration
    • Whether a CSS transition can use (or did use) accelerated compositing. If an effect isn't going to be hardware-accelerated, then a setTimeout animation is often more performant than a non-accelerated CSS transition. No reliable way to measure the framerate of a CSS transition.

This might not be a complete list

Add to the list if you know more Undetectables. Likewise, link to possible solutions if you think an item listed here can be feature-detected.

Contributors: paulirish, jaubourg, rwaldron, zachleat, snover, devongovett, rgrove, savetheclocktower, getify, garann, jdalton, michaelbrundage, necolas, mislav

Clone this wiki locally