components icon indicating copy to clipboard operation
components copied to clipboard

cdk-virtual-scroll-viewport does not handle css position: sticky cleanly

Open bdirito opened this issue 6 years ago • 58 comments
trafficstars

What is the expected behavior?

using the css attribute position: sticky should allow for elements to remain displayed according to their sticky parameters.

What is the current behavior?

This works for a while but then breaks after a point. Im assuming its a matter of the 'stickied' dom elements getting recycled.

What are the steps to reproduce?

https://stackblitz.com/edit/angular-wnar6b-wzjdmh

This is a slightly modified version of an example from the docs. As you scroll the cdk-virtual-scroll-viewport you will notice the light blue headers stick to the the top of their area as expected. Once you exceed a certain amount of scrolling however the headers disappear. For me items 0-2 work but the 3rd one breaks.

Down below is an example using the [non-virtual] scrolling. Here the css attribute works as expected.

Which versions of Angular, Material, OS, TypeScript, browsers are affected?

Forked from https://stackblitz.com/angular/apvvljrovad from the docs. no versions changed.

Is there anything else we should know?

https://github.com/angular/material2/issues/11621 also gets at my base requirement; wanting some elements of the virtual scroll list to 'stick around'. It would be ideal if standard css properties worked as via this ticket but that might be hard as the whole point of virtual-scroll is to remove things from the dom that 'shouldn't' be rendered. The position: sticky css attribute would seem to break those assumptions.

bdirito avatar Jan 15 '19 01:01 bdirito

any luck? looking for same solution.

ke1co1 avatar Feb 05 '19 16:02 ke1co1

Just following up on this, any update. This is must requirement for table with large dataset.

gurshinder78 avatar Mar 07 '19 21:03 gurshinder78

any luck? looking for same solution.

What i did to keep my header sticky was to use a sticky: true attribute in html. example:

Hope this helps!

ke1co1 avatar Mar 13 '19 15:03 ke1co1

@kelz2o3 I don't really get what your solution is. Could you add an example?

Dassderdie avatar Mar 13 '19 15:03 Dassderdie

i added this to my header row like so

    <tr
      mat-header-row
      *matHeaderRowDef="displayedColumns; sticky: true"
    ></tr>

ke1co1 avatar Mar 14 '19 14:03 ke1co1

sticky: true" is working fine with mat table if we don't use virtual scrolling, but it is messed up with virtual scrolling. Header starts moving when you start scrolling.

gurshinder78 avatar Mar 14 '19 18:03 gurshinder78

My use case is that I need a sticky header always present above a virtually-scrolling list ; I'm probably not the only one. Would you guys accept a PR that allows such a behavior (namely, a sticky header)?

florenthobein avatar Mar 18 '19 18:03 florenthobein

Here too, same problem now. @florenthobein which kind of workaround have you come up with? What's the root cause of this behavior?

Edit: okay, it seems transformY is the cause.

lppedd avatar Apr 30 '19 13:04 lppedd

It's the transformY indeed, that gets updated whenever elements of the list are replaced (remove+append).

I don't really have a neat solution. Tried to meddle with extending the FixedSizeVirtualScrollStrategy and move back a provided sticky header whenever onRenderedOffsetChanged was triggered but it's ugly and still too far from when transformY is applied, resulting in random blips unsuited for production.

So for now my "workaround" consists in:

  • having a header that is outside of the CDK Virtual Scroll and absolutely positioned over the list. Note: the header height has to be a multiple of your cdk itemSize.
  • prepending the list with "fake" elements that are gonna be hidden behind the header (hence the need for a header that's the total height of those fake elements).

...yeah. 😓 And a bit of a mess to have the good width for the header as it's not positioned inside the list.

florenthobein avatar Apr 30 '19 18:04 florenthobein

@fabioloreggian Thanks for the explanation! However, in the end I gave up, I don't want to use workaround that break over time

lppedd avatar May 09 '19 20:05 lppedd

@florenthobein I found a solution to the tranformY compensation, but for some reason it's not sufficient, as once I started scrolling too far down, the headers will scroll with the content. Regardless, this is what I came up with after fiddling around a bit with the CDK scroll view port.

cdk-virtual-scroll-viewport {
    height: 100%;

    table {
        height: 100%;
        width: 100%;
        border-collapse: separate;
        border-spacing: 0;

        thead {
            tr {
                th {
                    height: 59px;
                    border-bottom: 1px solid #d0d0d0;
                    padding: 7px 20px 7px 7px;
                    position: sticky;
                    background-color: #ffffff;
                    top: 0;
            }
        }
    }
}
<cdk-virtual-scroll-viewport #scrollViewport itemSize="59" (scrolledIndexChange)="scrollIndexChanged($event)">
  <table>
    <thead>
      <tr>
        <th [style.transform]="inverseTranslation">Header 1</th>
        <th [style.transform]="inverseTranslation">Header 2</th>
      </tr>
    </thead>
    <tbody>
      <tr *cdkVirtualFor="let movement of movements; even as even; index as i" [class.even]="even">...</tr>
    </tbody>
  </table>
</cdk-virtual-scroll-viewport>
@ViewChild(CdkVirtualScrollViewport)
public viewPort: CdkVirtualScrollViewport;

public get inverseTranslation(): string {
    if (!this.viewPort || !this.viewPort["_renderedContentTransform"]) {
        return "translateY(0px)";
    }
    return this.viewPort["_renderedContentTransform"].replace(/translateY\((\d+)px\)/, "translateY(-$1px)");
}

That gets rid of the weird effects, but only until you scrolled down for a while. Not sure what I'm doing wrong and if that's even a problem I could solve. I'm not too familiar with sticky positions, it's still sometimes a mystery how that works exactly.

wartab avatar May 10 '19 12:05 wartab

@wartab your solution is partially worked. When we scroll down after few moment headers disappear. Can you explain your code so i can find some solution. I'm also facing same challenge.

monu0751 avatar Jun 27 '19 11:06 monu0751

@monu0751 Sadly I do not know how to solve this. That is what I mean with "but for some reason it's not sufficient, as once I started scrolling too far down, the headers will scroll with the content".

This technique is to apply the opposite transform: translateY() to the column headers that CDK applies when using the virtual-scroll to create the scrollbar. This is done by adding a - sign in the regex replace. That CSS transform is contained in a private field _renderedContentTransform in the VirtualScrollViewPort.

wartab avatar Jun 27 '19 12:06 wartab

FWIW, I noticed that it does work as expected in Safari if you use position: -webkit-sticky;. Not sure if Webkit is the only one doing it right, or if they're just doing it in a different way that happens to work better for this case.

literalpie avatar Jun 27 '19 20:06 literalpie

I have been somewhat successful in getting around this issue by forcing translateY(0) and then creating a component with height: Xpx at the beginning of the viewport (or in the first row of a table). I bind X to the value from virtualScrollViewport that is usually used for the translate value.

literalpie avatar Jun 27 '19 20:06 literalpie

@florenthobein I found a solution to the tranformY compensation, but for some reason it's not sufficient, as once I started scrolling too far down, the headers will scroll with the content. Regardless, this is what I came up with after fiddling around a bit with the CDK scroll view port.

cdk-virtual-scroll-viewport {
    height: 100%;

    table {
        height: 100%;
        width: 100%;
        border-collapse: separate;
        border-spacing: 0;

        thead {
            tr {
                th {
                    height: 59px;
                    border-bottom: 1px solid #d0d0d0;
                    padding: 7px 20px 7px 7px;
                    position: sticky;
                    background-color: #ffffff;
                    top: 0;
            }
        }
    }
}
<cdk-virtual-scroll-viewport #scrollViewport itemSize="59" (scrolledIndexChange)="scrollIndexChanged($event)">
  <table>
    <thead>
      <tr>
        <th [style.transform]="inverseTranslation">Header 1</th>
        <th [style.transform]="inverseTranslation">Header 2</th>
      </tr>
    </thead>
    <tbody>
      <tr *cdkVirtualFor="let movement of movements; even as even; index as i" [class.even]="even">...</tr>
    </tbody>
  </table>
</cdk-virtual-scroll-viewport>
@ViewChild(CdkVirtualScrollViewport)
public viewPort: CdkVirtualScrollViewport;

public get inverseTranslation(): string {
    if (!this.viewPort || !this.viewPort["_renderedContentTransform"]) {
        return "translateY(0px)";
    }
    return this.viewPort["_renderedContentTransform"].replace(/translateY\((\d+)px\)/, "translateY(-$1px)");
}

That gets rid of the weird effects, but only until you scrolled down for a while. Not sure what I'm doing wrong and if that's even a problem I could solve. I'm not too familiar with sticky positions, it's still sometimes a mystery how that works exactly.

The problem with the solution you posted is that you're using style.transform. A position: sticky evaluates to a position: fixed once you hit the part where it would scroll, and that doesn't work right with a transformation.

Instead use style.top and set it to the inverse number of pixels, similar to what you're doing now.

lonestarjdavis avatar Jul 15 '19 16:07 lonestarjdavis

@lonestarjdavis Amazing, thanks for the heads up, that works like a charm!

wartab avatar Jul 16 '19 07:07 wartab

@wartab Could you please post a final solution, or create a stackblitz ?

Thanks.

artem-galas avatar Jul 16 '19 09:07 artem-galas

@artem-galas https://stackblitz.com/edit/components-issue-t3xvyz That works for my use case.

wartab avatar Jul 16 '19 15:07 wartab

@wartab Looks great with chrome. However, when drag the scroll bar with firefox, the table header is still not clean. Next if turn on the firefox debug inspector, it looks great again. Do you have any idea what cause the issue with firefox wilthout turn on the debug inspector?

lujian98 avatar Aug 20 '19 23:08 lujian98

@lujian98 Not sure what version of Firefox you are using, but it is not happening for me. I also do not really have much insight into Firefox since we only target Electron, which is Chromium. Apologies.

wartab avatar Aug 21 '19 07:08 wartab

Thank you @wartab I am using latest Firefox.

lujian98 avatar Aug 21 '19 16:08 lujian98

Nice one @wartab!

matthewiiv avatar Sep 19 '19 12:09 matthewiiv

@florenthobein I found a solution to the tranformY compensation, but for some reason it's not sufficient, as once I started scrolling too far down, the headers will scroll with the content. Regardless, this is what I came up with after fiddling around a bit with the CDK scroll view port.

cdk-virtual-scroll-viewport {
    height: 100%;

    table {
        height: 100%;
        width: 100%;
        border-collapse: separate;
        border-spacing: 0;

        thead {
            tr {
                th {
                    height: 59px;
                    border-bottom: 1px solid #d0d0d0;
                    padding: 7px 20px 7px 7px;
                    position: sticky;
                    background-color: #ffffff;
                    top: 0;
            }
        }
    }
}
<cdk-virtual-scroll-viewport #scrollViewport itemSize="59" (scrolledIndexChange)="scrollIndexChanged($event)">
  <table>
    <thead>
      <tr>
        <th [style.transform]="inverseTranslation">Header 1</th>
        <th [style.transform]="inverseTranslation">Header 2</th>
      </tr>
    </thead>
    <tbody>
      <tr *cdkVirtualFor="let movement of movements; even as even; index as i" [class.even]="even">...</tr>
    </tbody>
  </table>
</cdk-virtual-scroll-viewport>
@ViewChild(CdkVirtualScrollViewport)
public viewPort: CdkVirtualScrollViewport;

public get inverseTranslation(): string {
    if (!this.viewPort || !this.viewPort["_renderedContentTransform"]) {
        return "translateY(0px)";
    }
    return this.viewPort["_renderedContentTransform"].replace(/translateY\((\d+)px\)/, "translateY(-$1px)");
}

That gets rid of the weird effects, but only until you scrolled down for a while. Not sure what I'm doing wrong and if that's even a problem I could solve. I'm not too familiar with sticky positions, it's still sometimes a mystery how that works exactly.

inverseTranslation function can be improved as follows:

  get inverseTranslation(): string {
    const offset = this.viewPort.getOffsetToRenderedContentStart();

    return `-${offset}px`;
  }

or

get inverseTranslation(): string {
    return `-${this.viewPort.getOffsetToRenderedContentStart()}px`;
  }

fernandomeridamatilla avatar Oct 16 '19 07:10 fernandomeridamatilla

@mmalerba Can you get this on the virtual scrolling bug list https://github.com/angular/components/projects/20

bdirito avatar Oct 16 '19 17:10 bdirito

@artem-galas https://stackblitz.com/edit/components-issue-t3xvyz That works for my use case.

that was a good example and it actually keeps the header in top, but in my case I have resizable columns and header position fails when moving :( look here https://gifyu.com/image/kyj4

diegoale1994 avatar Nov 01 '19 16:11 diegoale1994

how I can chnage the color of # @this Capture

KheireddineAzzez avatar Jan 20 '20 12:01 KheireddineAzzez

I am in need of this feature as well.

@wartab Can I ask _renderedContentOffset is a private property ? Is there a "clean" way to access it (without string literals) ?

Also, it throws this pretty often:

ERROR Error: ExpressionChangedAfterItHasBeenCheckedError: Expression has changed after it was checked. Previous value for 'top': '-700px'. Current value: '-1200px'.

Yohandah avatar Aug 21 '20 12:08 Yohandah

@florenthobein I found a solution to the tranformY compensation, but for some reason it's not sufficient, as once I started scrolling too far down, the headers will scroll with the content. Regardless, this is what I came up with after fiddling around a bit with the CDK scroll view port.

cdk-virtual-scroll-viewport {
    height: 100%;

    table {
        height: 100%;
        width: 100%;
        border-collapse: separate;
        border-spacing: 0;

        thead {
            tr {
                th {
                    height: 59px;
                    border-bottom: 1px solid #d0d0d0;
                    padding: 7px 20px 7px 7px;
                    position: sticky;
                    background-color: #ffffff;
                    top: 0;
            }
        }
    }
}
<cdk-virtual-scroll-viewport #scrollViewport itemSize="59" (scrolledIndexChange)="scrollIndexChanged($event)">
  <table>
    <thead>
      <tr>
        <th [style.transform]="inverseTranslation">Header 1</th>
        <th [style.transform]="inverseTranslation">Header 2</th>
      </tr>
    </thead>
    <tbody>
      <tr *cdkVirtualFor="let movement of movements; even as even; index as i" [class.even]="even">...</tr>
    </tbody>
  </table>
</cdk-virtual-scroll-viewport>
@ViewChild(CdkVirtualScrollViewport)
public viewPort: CdkVirtualScrollViewport;

public get inverseTranslation(): string {
    if (!this.viewPort || !this.viewPort["_renderedContentTransform"]) {
        return "translateY(0px)";
    }
    return this.viewPort["_renderedContentTransform"].replace(/translateY\((\d+)px\)/, "translateY(-$1px)");
}

That gets rid of the weird effects, but only until you scrolled down for a while. Not sure what I'm doing wrong and if that's even a problem I could solve. I'm not too familiar with sticky positions, it's still sometimes a mystery how that works exactly.

The problem with the solution you posted is that you're using style.transform. A position: sticky evaluates to a position: fixed once you hit the part where it would scroll, and that doesn't work right with a transformation.

Instead use style.top and set it to the inverse number of pixels, similar to what you're doing now.

The inverseTranslation method will be call on every view check, maybe it is not ideal regarding performances.

The CdkVirtualScrollViewport has a ScrollStrategy private property offering a onRenderedOffsetChanged method we can use to update the header position:

       <tr>
        <th [style.top]="headerTop">Header 1</th>
        <th [style.top]="headerTop">Header 2</th>
      </tr>
    headerTop = '0px';

    ngAfterViewInit(): void {
        if (!!this.viewPort) {
            this.viewPort['_scrollStrategy'].onRenderedOffsetChanged = () =>
                (this.headerTop = `-${this.viewPort.getOffsetToRenderedContentStart()}px`);
        }
    }

ebrehault avatar Oct 01 '20 12:10 ebrehault

Working example with dynamic sticky items with correct placement: https://stackblitz.com/edit/angular-ivy-pxmnpx?file=src/app/app.component.html (css vars need angular 10+) An issue can arise where the sticky element will jump 1 px up and down if the rows get sub-pixel-height (in my project bc. the rows are calculated from viewport / 4). This can be avoided by making sure that the viewport height is dividable by e.g. 4 in my case (see https://stackoverflow.com/questions/37754542/css-calc-round-down-with-two-decimal-cases/64921523)

Remember to implement something like:

if (platform.SAFARI) {
this.offset = "0px";
}

wiegell avatar Mar 06 '21 00:03 wiegell

Hello there,

I came across a similar issue and used the useful advices from this thread (thanks very much for sharing btw !!!).

However, this doesn't work if your table is handled with an *ngIf directive. The sticky headers workaround gets broken, when the *ngIf directive removes the table from the dom, therefore the @ViewChild assignment "gets lost" as the variable value is not refreshed.

My workaround was to use a setter instead of a variable, in my component :

/** Instead of using a variable, use a setter */
@ViewChild('tableViewPort') private set tableViewPortContent(tbvp: CdkVirtualScrollViewport) {
    this.tableViewPort = tbvp;
    if (this.tableViewPort) {
      this.initiateStickyHeaders();
    }
  }

  public ITEM_SIZE = 50;
  public offset: number;
  private tableViewPort: CdkVirtualScrollViewport;
  private tableSubscribers: any = {};

[...]

public ngOnInit(): void {}

public ngOnChanges(): void {}

public ngOnDestroy(): void {
    this.tableSubscribers.forEach((sub: any) => {
      sub?.unsubscribe();
    });
}

private initiateStickyHeaders(): void {
    /** Remove previous possible existing subscriber for cleanup */
    _.forEach(this.tableSubscribers, (sub: any) => {
      sub?.unsubscribe();
    });

    let subsc = this.tableViewPort?.scrolledIndexChange
      .pipe(
        map(() => this.tableViewPort?.getOffsetToRenderedContentStart() * -1),
        distinctUntilChanged(),
      )
      .subscribe((offset) => {
        this.offset = offset;
      });
    this.tableSubscribers.scrolledIndexChange = subsc;

    subsc = this.tableViewPort?.renderedRangeStream?.subscribe((range) => {
      this.offset = range.start * -this.ITEM_SIZE;
    });
    this.tableSubscribers.renderedRangeStream = subsc;
}

In my template :

<cdk-virtual-scroll-viewport #tableViewPort id="emloyee-table" [style.height.px]="600" [itemSize]="ITEM_SIZE" matSort (matSortChange)="sortContent($event)">
        <table class="base-margin-top table" mat-table [dataSource]="getPagedEmployeesList()">

          <ng-container *ngFor="let header of headers" [matColumnDef]="header">
            
              <th [style.top.px]="offset"  style="background-color: white !important;" mat-header-cell *matHeaderCellDef>{{header}}</th>

            <td mat-cell *matCellDef="let emp"> {{emp[header]}} </td>
          </ng-container>

          <tr mat-header-row *matHeaderRowDef="headers; sticky: true" style="background-color: white !important;" [style.top.px]="offset"></tr>
          <ng-template let-row matRowDef cdkVirtualFor [matRowDefColumns]="headers" [cdkVirtualForOf]="getPagedEmployeesList()">
            <tr mat-row></tr>
          </ng-template>

        </table>
</cdk-virtual-scroll-viewport>

DVelbeck avatar Jan 27 '22 15:01 DVelbeck

Here is my solution using a directive. It uses the MutationObserver api to detect attributes changes on the scroll view port wrapper and then update the header top position.

directive:

import { Directive, ElementRef, OnDestroy, OnInit } from '@angular/core';

const TRANSLATE_Y_REGEX = /translateY\((\d+)px\)/;

const MUTATION_ATTRIBUTES_TYPE: MutationRecordType = 'attributes';

@Directive({
  selector: '[appTableStickyHeader]',
})
export class TableStickyHeaderDirective implements OnInit, OnDestroy {

  private translateYObserver: MutationObserver;

  constructor(
    private readonly virtualScrollViewPort: ElementRef,
  ) {
  }

  ngOnInit() {
    this.initStickHeader();
  }

  ngOnDestroy() {
    if (this.translateYObserver) {
      this.translateYObserver.disconnect();
    }
  }

  private initStickHeader() {
    const contentWrapper = this.virtualScrollViewPort.nativeElement.firstChild;
    const tableHeader = this.virtualScrollViewPort.nativeElement.querySelector('thead');

    this.translateYObserver = new MutationObserver((mutations =>
      this.onMutationsChange(mutations, contentWrapper, tableHeader)),
    );
    this.translateYObserver.observe(contentWrapper, {
      attributes: true,
    });
  }

  private onMutationsChange(mutations: MutationRecord[], contentWrapper: HTMLElement, tableHeader: HTMLElement) {
    mutations
      .filter(it => it.type === MUTATION_ATTRIBUTES_TYPE)
      .forEach(() => this.setHeaderTopPosition(contentWrapper, tableHeader));
  }

  private setHeaderTopPosition(contentWrapper: HTMLElement, tableHeader: HTMLElement) {
    const match = contentWrapper.style.transform.match(TRANSLATE_Y_REGEX);
    if (!match) {
      return;
    }

    const translate = parseInt(match[1]);
    tableHeader.style.top = `-${translate}px`;
  }
}

css:

thead {
  position: sticky;
}

ojacquemart avatar May 11 '22 12:05 ojacquemart

Once #24394 is released (I guess 14.1), it is possible to have sticky elements before the scroll-viewport but within the scrolling element (adopted from the dev-app):

<div class="demo-viewport" cdkVirtualScrollingElement>
  <p class="position: sticky; top: 0">Content before virtual scrolling items</p>
  <cdk-virtual-scroll-viewport itemSize="50">
    <div *cdkVirtualFor="let size of fixedSizeData; let i = index" class="demo-item"
         [style.height.px]="size">
      Item #{{i}} - ({{size}}px)
    </div>
  </cdk-virtual-scroll-viewport>
  <p>Content after virtual scrolling items</p>
</div>

So with this it should be possible to implement sticky headers.

However, using position:; sticky on items within the scroll-viewport is still not working. I guess this cannot be simply fixed within the cdk, but at least it could provide developers a better way to deal with it. As described above, the translate value needs to be compensated in top: ?.

@mmalerba what do you think about this approach: Exposing the current translate value as a css variable. Doing this would allow developers to make elements sticky like this, without the need of any additional js:

.sticky-element {
  position: sticky;
  top: calc(<wanted-top-value> + var(--cdk-scrolling-offset))
}

spike-rabbit avatar Jun 08 '22 09:06 spike-rabbit

That's an interesting idea. Let me add it to our agenda for team discussion since we don't currently have any APIs like this

mmalerba avatar Jun 16 '22 15:06 mmalerba

Hi @spike-rabbit, I had a chance to discuss this with the team. In general we're not opposed to this type of API in situations where it makes sense. Folks did have a couple of concerns:

  1. We're not sure if there would be performance implications of constantly updating a CSS variable. We'd need to do some testing to see if it causes some kind of style thrashing
  2. There was some concern that sticky items may have bigger issues than just the positioning, i.e. if the item scrolls out of range it could be removed from the DOM completely. This wouldn't be a concern using the virtual scroll in append only mode, but in that case the position should be trivial anyways.

I do see that a lot of the requests here have to do with making sticky header elements that are not virtualized items, but just persistent elements that appear within the scrolling element. I guess I'm curious what the reason for putting these headers inside the scrolling element is. Why not just have them outside so they're naturally "sticky"? (Sorry if I'm just missing something obvious)

<div class="box-with-border">
  <div>Header</div>
</div>
<cdk-virtual-scroll-viewport>
  <div *cdkVirtualFor="...">Item</div>
</cdk-virtual-scroll-viewport>

If there's a good motivation for needing them to be inside the scrolling element, and the performance concerns turn out to not be an issue, I could see adding this.

mmalerba avatar Jun 22 '22 19:06 mmalerba

@mmalerba I poorly reimplemented a use case that I have: https://stackblitz.com/edit/angular-ujjkis. In the real life application, there are much more items so that we needed to virtualize scrolling.

What I am trying to show there is, that sticky elements can be used in larger items that are scrolled, if they have some kind of title or other meta information that should be visible while the item is in the viewport.

Regarding your concerns:

  1. We have a custom virtual scrolling component using css variables running in production for a few years already. There are at least no obvious performance issues. To be honest, the application is only used on desktop computers and I never did actual performance tests.
  2. Based on my experiences, using sticky items works pretty well. It's ok, that the item is removed from the DOM when out of the viewport. The sticky element is not visible at that time anyway.

I hope this is understandable. Unfortunately, I cannot share the link to the real application as it is located in our Intranet.

spike-rabbit avatar Jun 24 '22 07:06 spike-rabbit

Ah ok, that seems like a reasonable use case. I wasn't thinking about a single item possibly being bigger than the viewport. I think we would probably be fine adding a CSS variable then. The next step would just be to do some quick testing to see if setting the variable somehow impacts performance

mmalerba avatar Jun 28 '22 18:06 mmalerba

Hi @spike-rabbit, I had a chance to discuss this with the team. In general we're not opposed to this type of API in situations where it makes sense. Folks did have a couple of concerns:

1. We're not sure if there would  be performance implications of constantly updating a CSS variable. We'd need to do some testing to see if it causes some kind of style thrashing

2. There was some concern that sticky items may have bigger issues than just the positioning, i.e. if the item scrolls out of range it could be removed from the DOM completely. This wouldn't be a concern using the virtual scroll in append only mode, but in that case the position should be trivial anyways.

I do see that a lot of the requests here have to do with making sticky header elements that are not virtualized items, but just persistent elements that appear within the scrolling element. I guess I'm curious what the reason for putting these headers inside the scrolling element is. Why not just have them outside so they're naturally "sticky"? (Sorry if I'm just missing something obvious)

<div class="box-with-border">
  <div>Header</div>
</div>
<cdk-virtual-scroll-viewport>
  <div *cdkVirtualFor="...">Item</div>
</cdk-virtual-scroll-viewport>

If there's a good motivation for needing them to be inside the scrolling element, and the performance concerns turn out to not be an issue, I could see adding this.

The reason this gets cited a lot is because specifically tables tend to misbehave frequently if you insert things that don't belong there. Addtionally, they make it quite trivial to have the header columns the same width as the body columns without forcing the width.

wartab avatar Jun 28 '22 18:06 wartab

@mmalerba so how do we proceed here? Shall I create a pull request, and you do the performance testing?

spike-rabbit avatar Jul 04 '22 12:07 spike-rabbit

@spike-rabbit sure, if you send me a PR I can look into how to test the performance

mmalerba avatar Jul 13 '22 20:07 mmalerba