-webkit-transform and input type=text do not work together. Transforms mangle the rendering of the inputs (bug1 and bug2), but also typing into the inputs mangles the transform (bug3 and bug4). I also include the expected behavior (bug0). This has been tested on Safari 3.2.1 (5525.27.1) and WebKit-SVN-r38707.dmg on Mac OS X 10.5.5 (intel macbook). Similar but more severe problems occur in Mobile Safari on iPhone OS 2.1, but I suspect that is outside the scope of this bugzilla. In case not: There bug3 and bug4 causes the iPhone to hang (the keyboard shows the letter being selected, but hangs there. Home key and sleep key are unresponsive; holding sleep then tapping home wakes it up). This depends on the keybaord being visible. If the keyboard is hidden (javascript is used to select the input), then bug3 and bug4 on Mobile Safari behave approximately like bug3 on Webkit SVN.
Created attachment 25499 [details] Testcase: Expected behavior (bug0) This is just a simple form. You should be able to type into the text box without problems. No transforms have been applied.
Created attachment 25500 [details] Testcase: mild rendering anomaly (bug1) This includes -webkit-transform: translate(0,0) and now there are mild rendering problems with the caret.
Created attachment 25502 [details] Testcase: Severe rendering anomaly (bug2) Now -webkit-transform: translate(-320px,0) causes severe rendering anomalies. The typed text does not appear after the first letter until the input control is blurred.
Created attachment 25503 [details] Testcase: Transform damaged (bug3) Another severe rendering anomaly. Now the transformed element is inside a div of fixed with with overflow: hidden. Once the first letter is typed into the input control, the entire fieldset vanishes.
Created attachment 25504 [details] Testcase: Interesting variant, walking input (bug4) This is an interesting variant where the width of the div is larger than the fieldset. Now the input "walks" slowly off the screen to the left as text is typed. One can use javascript to reset the transform to "translate(0,0)", but the result is still a severely translated element (though not at any particularly explainable position).
Fixed via bug 15671, bug 21906 and others.
I confirm the bug is fixed in the WebKit r39090 nightly. Thanks very much!