Offset: Operate "relative to document" correctly #3096
Conversation
|
Thank you for your pull request. It looks like this may be your first contribution to a jQuery Foundation project, if so we need you to sign our Contributor License Agreement (CLA). 📝 Please visit http://contribute.jquery.org/CLA/ to sign. After you signed, the PR is checked again automatically after a minute. If there's still an issue, please reply here to let us know. If you've already signed our CLA, it's possible your git author information doesn't match your CLA signature (both your name and email have to match), for more information, check the status of your CLA check. |
|
I don't know why "CLA:Error"... 😢 |
Related to: jquery#3080 Remove unnecessary code that subtracts `document.documentElement.clientTop/clientLeft` (i.e. `border-width` of `<html>`) from coordinates. In Firefox, those are not subtracted because `clientTop/clientLeft` are `0` always by accidentally bug.
|
@anseki See http://contribute.jquery.org/CLA/status/?owner=jquery&repo=jquery&sha=359da5dd3b080c4c6367ec904fa9325a1d4f0305 . We ask for your real email address and real name (or consistently-used pseudonym), and require that the git commits match the CLA. Is "AnSeki" the best name for you to use for assigning contributions? |
|
Thank you @gibson042. What should I do? |
|
The current exception is about the name field. Do you have a more "traditional" name, or is "AnSeki" the best descriptor for us? |
|
I've been using this name anytime, and I hope use this. |
Fix: incorrect indention
|
|
||
| for ( flags = 0; | ||
| flags <= ( FLAG_DOC_MARGIN | FLAG_DOC_BORDER | FLAG_DOC_PADDING | FLAG_BODY_MARGIN | FLAG_BODY_BORDER | FLAG_BODY_PADDING ); | ||
| flags++ ) { |
There was a problem hiding this comment.
This is a very good start, but is a bit hard to read and hard to reason about. I now think every interesting case here can operate in a page with nonzero, non-pixel (to catch inappropriate parseFloats) values for all of margin, border, and padding on both <html> and <body>, with the only variation being position values for those two elements. You can achieve that by means of a single iframe fixture (setting position styles before making any other calls), but I would like them in distinct testIframe invocations (see effects.js for an example of such process-constructed QUnit.test input). The cases, and my observations of them, are as follows:
<html> position |
<body> position |
.offset origin |
|---|---|---|
| static | static/relative/absolute/fixed | document (i.e., outer edge of <html> margin) |
| relative/absolute/fixed | static/relative/absolute/fixed | inside <html> border |
That's a total of 16 cases, each case with its own testIframe. I'm aware that <body> position doesn't actually seem to matter, but it's probably worth covering anyway.
There was a problem hiding this comment.
Ok, I try to rewrite tests.
You said that situations() should be called in each case of that 16 cases. Is my interpretation correct?
Also, that 16 cases include two cases in current code "CASE 1" and "CASE 2".
Is "CASE 3" in current code unnecessary?
Or, should two cases marker1.style.position = "absolute" and marker1.style.position = "static" be tested in each case of that 16 cases? That is, situations() is called in each case of 2x16 cases.
|
It seems "CLA:Error" was solved by second commit. Thank you. |
- Add more test cases of `position` variation of `<html>` and `<body>`. - Do `testIframe()` in each case.
|
I added test cases, then
Total: 8192 cases Also, I tried to change messages. |
Add comment as information for tester who see the test code also when the test failed.
| FLAG_DOC_PADDING = 4, | ||
| FLAG_BODY_MARGIN = 8, | ||
| FLAG_BODY_BORDER = 16, | ||
| FLAG_BODY_PADDING = 32, |
There was a problem hiding this comment.
You said that
situations()should be called in each case of that 16 cases. Is my interpretation correct?
Not exactly., I'm saying that situations and "CASE 1"/"CASE 2"/"CASE 3" should be removed in favor of a simpler approach. The testIframe callback shouldn't need to do too much:
// Assume initial conditions include html/document margin/border/padding:
// html { font-size: 10px; margin: 1em; border: 2em solid red; padding: 4em; }
// body { margin: 8em; border: 16em solid blue; padding: 32em; }
// …and a static element and an absolutely-positioned element:
// #static { position: static; }
// #absolute { position: absolute; top: 2em; left: 2em; }
// Establish document-relative origins for children of <body>
docElem.style.position = docPosition;
body.style.position = bodyPosition;
var bodyContentOrigin = 10 + 20 + 40 + 80 + 160 + 320;
var origin = 0 +
(docPosition === "static" ? 0 : 10 + 20) +
(bodyPosition === "static" ? 0 : 40 + 80 + 160);
// Check offsets
var absoluteOffset = $( "#absolute" ).offset();
assert.deepEqual(
$( "#static" ).offset(),
{ top: bodyContentOrigin, left: bodyContentOrigin },
"offset of position:static element includes <html> and <body> box styles" );
assert.deepEqual( absoluteOffset, { top: origin + 20, left: origin + 20 },
"offset of position:absolute element ignores box styles of position:static ancestors" );
$( "#absolute, #static" ).offset( absoluteOffset );
assert.deepEqual( $( "#absolute" ).offset(), absoluteOffset, "offset() round-trips" );
assert.deepEqual( $( "#static" ).offset(), absoluteOffset, "offset() is transitive" );
$( "#static" ).css( "position", "static" );
// Reposition html and body, tracking origin adjustments given margin/border/padding
var originAdjust = 0;
$( docElem ).css( { top: "1.5em", left: "1.5em" } );
if ( docPosition !== "static" ) {
originAdjust += 15 - (docPosition === "relative" ? 0 : 10);
}
$( body ).css( { top: "3em", left: "3em" } );
if ( bodyPosition === "fixed" || bodyPosition === "absolute" && docPosition === "static" ) {
// html box styles no longer matter
origin = 80 + 160;
bodyContentOrigin = origin + 320;
originAdjust = 30;
} else if ( bodyPosition !== "static" ) {
originAdjust += 30 - (bodyPosition === "relative" ? 0 : 40);
}
// Recheck offsets
absoluteOffset = $( "#absolute" ).offset();
assert.deepEqual(
$( "#static" ).offset(),
{ top: bodyContentOrigin + originAdjust, left: bodyContentOrigin + originAdjust },
"offset of position:static element respects ancestor positioning" );
assert.deepEqual(
absoluteOffset,
{ top: origin + originAdjust + 20, left: origin + originAdjust + 20 },
"offset of position:absolute respects ancestor positioning" );
$( "#absolute, #static" ).offset( absoluteOffset );
assert.deepEqual( $( "#absolute" ).offset(), absoluteOffset, "offset() still round-trips" );
assert.deepEqual( $( "#static" ).offset(), absoluteOffset, "offset() is still transitive" );And there could easily be ways to simplify even further.
There was a problem hiding this comment.
Sorry, I have to learn English...
So, please check following:
- I interpreted your comment in that issue as that the code should be tested with changing
marginandborder. Did I mistake again? - All
assert.deepEqual()s that are passed return value from.offset()failed.
I found this:
https://github.com/jquery/jquery/blob/master/external/qunit/qunit.js#L1628
if ( a.constructor === b.constructor ) {
Expected values that were made in the test code and return values from.offset()that were made iniframehave constructor in different namespaces (i.e. window), thereforea.constructor === b.constructorisfalse.
Return values from.offset()have to be copied to be checked byassert.deepEqual().
supportjQuery.extend({}, $( "#static" ).offset()) - This test failed when
html{position}isstaticandbody{position}isnon-static.
assert.deepEqual( absoluteOffset, { top: origin + 20, left: origin + 20 },
"offset of position:absolute element ignores box styles of position:static ancestors" );For example, in a case html{position:static} body{position:relative}:
Expected:
{
"left": 300,
"top": 300
}
Result:
{
"left": 330,
"top": 330
}
I think, this code is incorrect:
var origin = 0 +
(docPosition === "static" ? 0 : 10 + 20) +
(bodyPosition === "static" ? 0 : 40 + 80 + 160);If body has position:non-static, the element is positioned relative to body without being affected by position of <html>. body was already positioned with margin, border and padding of <html>.
Therefore the code should be:
var origin =
bodyPosition !== "static" ? 10 + 20 + 40 + 80 + 160 : // `10 + 20 + 40` must be included
docPosition !== "static" ? 10 + 20 : // relative to inside `<html>`
0; // relative to document (i.e. outside `<html>`)- This test failed when
html{position}is notstaticorrelative.
assert.deepEqual(
supportjQuery.extend({}, $( "#static" ).offset()),
{ top: bodyContentOrigin + originAdjust, left: bodyContentOrigin + originAdjust },
"offset of position:static element respects ancestor positioning" );I think, this code is incorrect:
if ( docPosition !== "static" ) {
originAdjust += 15 - (docPosition === "relative" ? 0 : 10);
}If <html> has position:non-static, it is positioned with margin even if potision is other than relative.
Therefore the code should be:
if ( docPosition !== "static" ) {
originAdjust += 15;
}- This test failed in some cases:
assert.deepEqual(
absoluteOffset,
{ top: origin + originAdjust + 20, left: origin + originAdjust + 20 },
"offset of position:absolute respects ancestor positioning" );This is solved by fixing originAdjust and origin.
There was a problem hiding this comment.
The test code I wrote might do the almost same as your code, by removing code that changes margin and border.
But I didn't think that it should change top and left of <html> and body.
Therefore I think that I can't write PR. Maybe, I can't understand your request well by my very poor English.
Then, I hope that someone write PR.
There was a problem hiding this comment.
BTW, why the test doesn't need to change border?
I think that this is PR to fix bug with clientTop/clientLeft (i.e. border).
At least, it should test in cases of that <html> has border and no border.
There was a problem hiding this comment.
Sorry, I have to learn English...
So, please check following:
No problem. And it's not all your misunderstanding; my suggestions are changing as I learn new details of browser behavior around this issue.
I interpreted your comment in that issue as that the code should be tested with changing
marginandborder. Did I mistake again?
It was a partial mistake. I wasn't saying that margin and border need to change, just that test cases should cover both zero and nonzero values. Further, we already have tests for the no-margin no-border no-padding cases, and I have since concluded that the "only margin" and "only border" cases are unnecessary, so this PR can focus on situations in which all the box styles are nonzero.
Return values from
.offset()have to be copied to be checked byassert.deepEqual()…supportjQuery.extend({}, $( "#static" ).offset())
👍
If
bodyhasposition:non-static, the element is positioned relative tobodywithout being affected bypositionof<html>.bodywas already positioned withmargin,borderandpaddingof<html>.
Therefore the code should be:var origin = bodyPosition !== "static" ? 10 + 20 + 40 + 80 + 160 : // `10 + 20 + 40` must be included docPosition !== "static" ? 10 + 20 : // relative to inside `<html>` 0; // relative to document (i.e. outside `<html>`)
I definitely mis-defined origin, and I agree that your fix is correct.
I think, this code is incorrect:
if ( docPosition !== "static" ) { originAdjust += 15 - (docPosition === "relative" ? 0 : 10); }If
<html>hasposition:non-static, it is positioned withmargineven ifpotisionis other thanrelative.
Therefore the code should be:if ( docPosition !== "static" ) { originAdjust += 15; }
Again, a correct fix for my mistake.
The test code I wrote might do the almost same as your code, by removing code that changes
marginandborder.
But I didn't think that it should changetopandleftof<html>andbody.
Therefore I think that I can't write PR. Maybe, I can't understand your request well by my very poor English.
Then, I hope that someone write PR.
I really think we're close. My complaint wan't about what you were testing, just how it was being tested. And as shown above, you clearly have the right insights.
There was a problem hiding this comment.
Thank you very much for your very attentive care.
I understood that the test doesn't have to change border.
I have more questions.
- Why does the code change the
top/leftofhtml/body?
I think that these properties affectgetBoundingClientRect()method of browser, and.offset()method of jQuery calculates with a return value of thegetBoundingClientRect()without consideringtop/left.
That is, if the result become incorrect by changingtop/left, it is bug ofgetBoundingClientRect()method (i.e. browser), not jQuery.
Also,top/leftofhtml/bodyare usually not changed. - Why is the code that passes a return value of
.offset()to another element unnecessary?
https://github.com/anseki/jquery/blob/offset-html-border/test/unit/offset.js#L650
This is not "round-trip". I thought that the test should compare an element that was positioned by.offset()and another element that was positioned by native CSS, forsetOffset().
setOffset()must position an element at the same position as another element that was positioned by various CSS properties (position,border, etc.).
I think that a cases of an element that hasposition:staticare important becausesetOffset()calculates distance from current coordinates.offsetmethod returns incorrect value in Firefox #3080 (comment)
That is, it compares coordinates that was got without.offsetand coordinates that was changed by.offsetthat should not be affected by border of<html>.
There was a problem hiding this comment.
Why does the code change the
top/leftofhtml/body?
I think that these properties affectgetBoundingClientRect()method of browser, and.offset()method of jQuery calculates with a return value of thegetBoundingClientRect()without consideringtop/left.
That is, if the result become incorrect by changingtop/left, it is bug ofgetBoundingClientRect()method (i.e. browser), not jQuery.
If there are such bugs, we need to know so we can work around them (or document our failure to do so).
Also,
top/leftofhtml/bodyare usually not changed.
Nor do html or body elements usually have borders. The point is verifying that our functions perform as claimed.
Why is the code that passes a return value of
.offset()to another element unnecessary?
…
I think that a cases of an element that hasposition:staticare important becausesetOffset()calculates distance from current coordinates. #3080 (comment)
I agree. We should do both $( "#absolute" ).offset( absoluteOffset ) and $( "#static" ).offset( absoluteOffset ) and expect the subsequent .offset() to match absoluteOffset. I updated my suggestions accordingly.
|
I reconsidered about setter. Therefore if getter works correctly, things that affect setter are:
I think that the test doesn't need to change I will rewrite the test later. |
- Remove the code that changes box-properties of `<html>` and `body`. - Add `testIframe` to test the setter.
|
BTW, I found another issue. I found that I send this as "new issue" later. |
We don't need to change |
| html { | ||
| margin: 1px; | ||
| border-width: 2px; | ||
| padding: 4px; |
There was a problem hiding this comment.
It is important to me that all of these units are not pixels.
|
I see. |
- Copy the code from jquery#3096 (comment) - Wrap `.offset()` by `supportjQuery.extend`. - Fix `origin` calucuration. - Fix `originAdjust` calucuration.
| "offset of position:static element includes <html> and <body> box styles" ); | ||
| assert.deepEqual( absoluteOffset, { top: origin + 20, left: origin + 20 }, | ||
| "offset of position:absolute element ignores box styles of position:static ancestors" ); | ||
| $( "#absolute, #static" ).offset( absoluteOffset ); |
|
As discussed at today's meeting, we'll hold off on changing the |
Summary
Fix:
.offset()doesn't operate "relative to document" correctlyChecklist
Mark an
[x]for completed items, if you're not sure leave them unchecked and we can assist.Thanks! Bots and humans will be around shortly to check it out.
Related to: #3080
Remove unnecessary code that subtracts
document.documentElement.clientTop/clientLeft(i.e.border-widthof<html>) from coordinates.In Firefox, those are not subtracted because
clientTop/clientLeftare0always by accidentally bug.