ChatGPT解决这个技术问题 Extra ChatGPT

<fieldset> resizes wrong; appears to have unremovable `min-width: min-content`

Problem

I have a <select> where one of its <option>’s text values is very long. I want the <select> to resize so it is never wider than its parent, even if it has to cut off its displayed text. max-width: 100% should do that.

Before resize:

https://i.stack.imgur.com/RIv8L.png

What I want after resize:

https://i.stack.imgur.com/DJnrc.png

But if you load this jsFiddle example and resize the Result panel’s width to be smaller than that of the <select>, you can see that the select inside the <fieldset> fails to scale its width down.

What I’m actually seeing after resize:

https://i.stack.imgur.com/fcGsM.png

However, the equivalent page with a <div> instead of a <fieldset> does scale properly. You can see that and test your changes more easily if you have a <fieldset> and a <div> next to each other on one page. And if you delete the surrounding <fieldset> tags, the resizing works. The <fieldset> tag is somehow causing horizontal resizing to break.

The <fieldset> acts is as if there is a CSS rule fieldset { min-width: min-content; }. (min-content means, roughly, the smallest width that doesn’t cause a child to overflow.) If I replace the <fieldset> with a <div> with min-width: min-content, it looks exactly the same. Yet there is no rule with min-content in my styles, in the browser default stylesheet, or visible in Firebug’s CSS Inspector. I tried to override every style visible on the <fieldset> in Firebug’s CSS Inspector and in Firefox’s default stylesheet forms.css, but that didn’t help. Specifically overriding min-width and width didn’t do anything either.

Code

HTML of the fieldset:

<fieldset>
    <div class="wrapper">
        <select id="section" name="section">
            <option value="-1"></option>
            <option value="1501" selected="selected">Sphinx of black quartz, judge my vow. The quick brown fox jumps over the lazy dog.</option>
            <option value="1480">Subcontractor</option>
            <option value="3181">Valley</option>
            <option value="3180">Ventura</option>
            <option value="3220">Very Newest Section</option>
            <option value="1481">Visitor</option>
            <option value="3200">N/A</option>
        </select>
    </div>
</fieldset>

My CSS that should be working but isn’t:

fieldset {
    /* hide fieldset-specific visual features: */
    margin: 0;
    padding: 0;
    border: none;
}

select {
    max-width: 100%;
}

Resetting the width properties to the defaults does nothing:

fieldset {
    width: auto;
    min-width: 0;
    max-width: none;
}

Further CSS in which I try and fail to fix the problem:

/* try lots of things to fix the width, with no success: */
fieldset {
    display: block;
    min-width: 0;
    max-width: 100%;
    width: 100%;
    text-overflow: clip;
}

div.wrapper {
    width: 100%;
}

select {
    overflow: hidden;
}

More details

The problem also occurs in this more comprehensive, more complicated jsFiddle example, which is more similar to the web page I’m actually trying to fix. You can see from that that the <select> is not the problem – an inline-block div also fails to resize. Though this example is more complicated, I assume that the fix for the simple case above will also fix this more complicated case.

[Edit: see browser support details below.]

One curious thing about this problem is that if you set div.wrapper { width: 50%; }, the <fieldset> stops resizing itself at the point then the full-size <select> would have hit the edge of the viewport. The resizing happens as if the <select> has width: 100%, even though the <select> looks like it has width: 50%.

https://i.stack.imgur.com/ne1lu.png

If you give the <select> itself width: 50%, that behavior does not occur; the width is simply correctly set.

https://i.stack.imgur.com/kcHbr.png

I don’t understand the reason for that difference. But it may not be relevant.

I also found the very similar question HTML fieldset allows children to expand indefinitely. The asker couldn’t find a solution and guesses that there is no solution apart from removing the <fieldset>. But I’m wondering, if it really is impossible to make the <fieldset> display right, why is that? What in <fieldset>’s spec or default CSS (as of this question) causes this behavior? This special behavior is probably be documented somewhere, since multiple browsers work like this.

Background goal and requirements

The reason I’m trying to do this is as part of writing mobile styles for an existing page with a big form. The form has multiple sections, and one part of it is wrapped in a <fieldset>. On a smartphone (or if you make your browser window small), the part of the page with the <fieldset> is much wider than the rest of the form. Most of the form constrains its width just fine, but the section with the <fieldset> does not, forcing the user to zoom out or scroll right to see all of that section.

I’m wary of simply removing the <fieldset>, as it is generated on many pages in a big app, and I’m not sure what selectors in CSS or JavaScript might depend on it.

I can use JavaScript if I need to, and a JavaScript solution is better than nothing. But if JavaScript is the only way to do this, I’d be curious to hear an explanation for why this is not possible using only CSS and HTML.

Edit: browser support

On the site, I need to support Internet Explorer 8 and later (we just dropped support for IE7), the latest Firefox, and the latest Chrome. This particular page should also work on iOS and Android smartphones. Slightly degraded but still usable behavior is acceptable for Internet Explorer 8.

I retested my broken fieldset example on different browsers. It actually already works in these browsers:

Internet Explorer 8, 9, and 10

Chrome

Chrome for Android

It breaks in these browsers:

Firefox

Firefox for Android

Internet Explorer 7

Thus, the only browser I care about that the current code breaks in is Firefox (on both desktop and mobile). If the code were fixed so it worked in Firefox without breaking it in any other browsers, that would solve my problem.

The site HTML template uses Internet Explorer conditional comments to add classes such .ie8 and .oldie to the <html> element. You can use those classes in your CSS if you need to work around styling differences in IE. The classes added are the same as in this old version of HTML5 Boilerplate.

congratulations to this excellent writeup
Your simple example: give <fieldset> (or div.wrapper) whatever width it needs, and select { width:100% }. max-width appears to give the element some % of its parent's original width, then does not scale, whereas width does scale with its parent. I think that will do what you want (but I could have misunderstood). In your more complex fiddle, try giving the fields in the left column a %-width, and a min- pixel-width. Then give the fields on the right 100 - (left-fields-%-width) % width.
I just had another thought: what if you set the fieldset's min-width to 0? That by itself probably won't solve the problem, but possibly worth trying to work out. I'm working down a different path now, but if it doesn't work I'll try that out too
Do you need the label and field to stay on the same line? For small viewports, they overlap. Why not make it somewhat responsive, and break to a new line if necessary?

J
Jordan Gray

Update (25 Sept 2017)

The Firefox bug described below is fixed as of Firefox 53 and the link to this answer has finally been removed from Bootstrap's documentation.

Also, my sincere apologies to the Mozilla contributors who had to block removing support for -moz-document partly due to this answer.

The fix

In WebKit and Firefox 53+, you just set min-width: 0; on the fieldset to override the default value of min-content

Still, Firefox is a bit… odd when it comes to fieldsets. To make this work in earlier versions, you must change the display property of the fieldset to one of the following values:

table-cell (recommended)

table-column

table-column-group

table-footer-group

table-header-group

table-row

table-row-group

Of these, I recommend table-cell. Both table-row and table-row-group prevent you from changing width, while table-column and table-column-group prevent you from changing height.

This will (somewhat reasonably) break rendering in IE. Since only Gecko needs this, you can justifiably use @-moz-document—one of Mozilla's proprietary CSS extensions—to hide it from other browsers:

@-moz-document url-prefix() {
    fieldset {
        display: table-cell;
    }
}

(Here's a jsFiddle demo.)

That fixes things, but if you're anything like me your reaction was something like…

What.

There is a reason, but it's not pretty.

The default presentation of the fieldset element is absurd and essentially impossible to specify in CSS. Think about it: the fieldset's border disappears where it's overlapped by a legend element, but the background remains visible! There's no way to reproduce this with any other combination of elements.

To top it off, implementations are full of concessions to legacy behaviour. One such is that the minimum width of a fieldset is never less than the intrinsic width of its content. WebKit gives you a way to override this behaviour by specifying it in the default stylesheet, but Gecko² goes a step further and enforces it in the rendering engine.

However, internal table elements constitute a special frame type in Gecko. Dimensional constraints for elements with these display values set are calculated in a separate code path, entirely circumventing the enforced minimum width imposed on fieldsets.

Again—the bug for this has been fixed as of Firefox 53, so you do not need this hack if you are only targeting newer versions.

Is using @-moz-document safe?

For this one issue, yes. @-moz-document works as intended in all versions of Firefox up until 53, where this bug is fixed.

This is no accident. Due in part to this answer, the bug to limit @-moz-document to user/UA stylesheets was made dependent on the underlying fieldset bug being fixed first.

Beyond this, do not use @-moz-document to target Firefox in your CSS, other resources notwithstanding.³

¹ Value may be prefixed. According to one reader, this has no effect in Android 4.1.2 Stock Browser and possibly other old versions; I have not had time to verify this.

² All links to the Gecko source in this answer refer to the 5065fdc12408 changeset, committed 29ᵗʰ July 2013; you may wish to compare notes with the most recent revision from Mozilla Central.

³ See e.g. SO #953491: Targeting only Firefox with CSS and CSS Tricks: CSS hacks targeting Firefox for widely referenced articles on high-profile sites.


I had just noticed that the fix broke in IE, only to come here and find out that you’ve already edited in the solution to that. Thanks! The Gecko hack solution works fine for me in each browser.
Was pulling my hair out on this issue, but your fix is perfect - except it only seems to work in firefox. Any suggestions for other browsers, especially mobile browsers? My biggest problem is resizing due to orientation change on mobile devices...
This is an absolutely superb answer - thorough, researched and concise. Thanks.
Of course it's an absolutely superb answer! Bootstrap official documentation points here
Well, I never knew that about fieldsets. I have learned something here today.
b
boars

Safari on iOS issue with selected answer

I found the answer from Jordan Gray to be particularly helpful. However it didn't seem to solve this issue on Safari iOS for me.

The issue for me is simply that the fieldset cannot have an auto width if the element within has a max-width as a % width.

Fix for issue

Simply setting the fieldset to have a 100% width of it's container seems to get around this issue.

Example

fieldset {
    min-width: 0; 
    width: 100%; 
}

Please refer to the below for working examples - if you remove the % width off the fieldset or replace it with auto, it will not continue to function.

JSFiddle | Codepen


This is a great addition to the general problem, I didn't think to test it in iOS. :) I almost think this might be worth you asking and self-answering as a new question, too. I'll try to do some extra testing this week and link to this answer if it works.
R
Rory O'Kane

I’ve struggled for many hours with this, and basically, the browser is applying computed styling that you need to override in your CSS. I forget the exact property that is being set on fieldset elements versus divs (perhaps min-width?).

My best advice would be to change your element to a div, copy the computed styles from your inspector, then change your element back to fieldset and compare the computed styles to find the culprit.

Hope that helps.

Update: Adding display: table-cell helps in non-Chrome browsers.


I compared every CSS property of the fieldset and div on this page in Firebug. It didn’t help. In Firebug, the only properties with different values were width and perspective-origin. The width was numerically different, but the fieldset’s width was auto, like the div’s. And the perspective-origin is just a function of the width – 50% of it by default.
However, the comparison did help in a way in Chrome Web Inspector. I found out right after opening my example that it is working in Chrome! I had added the min-width: 0; to my example after testing in Chrome, but it seems like that solved the problem in Chrome. The Web Inspector tells me that Chrome has the style min-width: -webkit-min-content;, just as I hypothesized. So now I only need to solve the problem in Firefox and possibly other browsers I haven’t tested in yet.
Glad it helped a little. I was able to solve my issue in other browsers with "display: table-cell". I don't think that it's not the best solution, but hopefully that works for you.
T
Trojan

.fake-select { white-space:nowrap; } caused the fieldset to interpret the .fake-select element by its original width, rather than its forced width (even when the overflow is hidden).

Remove that rule, and change .fake-select's max-width:100% to just width:100% and everything fits. The caveat is that you see all of the content of the fake-select, but I don't think this is all that bad, and it fits horizontally now.

Update: with the current rules in the following fiddle (which contains only real selects), the fieldset's children are constrained to correct widths. Other than removing rules for .fake-select and fixing comments (from // comment to /* comment */, I've noted changes in the fiddle's CSS.

I understand your problem better now, and the fiddle reflects some progress. I set default rules for all <select>s, and reserve .xxlarge for those which you know will be wider than 480px (and this only works because you know the width of #viewport, and can manually add the class to those too wide. Just requires a little bit of testing)

Proof


Whatever fix you use has to work on an actual <select>. The actual site has only <select>s. The point of fake-select was to have the layout rules of a <select> without actually being a <select>, just to show that this behavior isn’t a browser bug that only happens with <select> elements. So changing the styles of .fake-select to be less like a <select> defeats the purpose.
I replaced the .fake-select box with a <select> (identical to the one already present) to demonstrate. The elements are all constrained to the parent's width (except when you click on it, dropping down the options, which are each displayed on a single line - but I don't think you can control that). I also removed all rules for .fake-select. Does the current fiddle do what you need?
It does not. width: 100% works with the long menus on the page, but does not work right with short menus. It expands short menus to the full width of the page, but I want short menus to keep their short width. Menus should be their natural width if there is room, but 100% width if they are too big for their parent. That’s what max-width is supposed to do, but failing to do in this case.
I've made some more changes, separating "all <select>s" from "long <select>s", and just working on background width now.
#viewport does not have a fixed width, either. That’s why I put the Resize Viewport button on the page – so you can test that everything works no matter what the viewport width is. Just as .fake-select is a substitute for a real <select> for testing, #viewport is a substitute for a real viewport. (A “viewport” is basically a browser window). I included #viewport on the page so that you could easily see elements that stick out of the viewport, rather than having to scroll to see them.