RFR: JDK-8308659: Use CSS scroll-margin instead of flexbox layout in API documentation
Jonathan Gibbons
jjg at openjdk.org
Thu Oct 5 18:12:43 UTC 2023
On Thu, 28 Sep 2023 19:50:21 GMT, Hannes Wallnöfer <hannesw at openjdk.org> wrote:
> A few years ago we switched to [Flexbox Layout](https://developer.mozilla.org/en-US/docs/Glossary/Flexbox) to implement the sticky header in the API docs generated by `javadoc`. This solved the problem of anchor link targets [not being positioned correctly](https://bugs.openjdk.org/browse/JDK-8223378), but it also introduced a few new ones:
>
> - It required a workaround to get browser history to work (JDK-8249133, JDK-8250779, 8286832)
> - It changed certain aspects of scrolling behavior in the browser (JDK-8301080)
> - It changed the way some CSS properties are interpreted (JDK-8315800)
>
> The reason for most of these problems is that the layout paradigm used by Flexbox is very different from traditional layout of HTML pages. The `scroll-margin-*` CSS properties introduced by the [CSS Scroll Snap Module](https://www.w3.org/TR/css-scroll-snap-1/) provide a simpler and less intrusive way to implement link target offsets in combination with sticky elements implemented using [`position: sticky`](https://developer.mozilla.org/en-US/docs/Web/CSS/position). However, although it is implemented [in all supported browsers](https://caniuse.com/?search=scroll-margin), it comes with its own challanges and quirks, which are explained below.
>
> The first problem to overcome was that `position: sticky` is more fragile on mobile browsers than Flexbox. If some part of the page content overflows the allotted horizontal space, the whole page can be zoomed out to view the whole content. This causes the header to become unsticky and the link target offsets to become erroneous. It was therefore necessary to make sure page content never overflows its allotted horizontal space, either by adding line break opportunities, or by making all or part of the page horizontally scrollable when its content overflows. Line break opportunities are difficult to add especially in preformatted code, so I opted for making the parts of the page most likely containing long lines of code scrollable.
>
> The next problem was that enabling horizontal scrolling on an element or one of its containing elements breaks the `scroll-margin-top` property in Chrome due to a browser quirk (both desktop and mobile versions). This means that the scrolling must occur in a child element rather than the sections or other elements serving as link targets.
>
> When enabling horizontal scrolling on the contents of sections containing user-provided content, another problem is that it disables [Margin Collapse](https://www.joshwc...
src/jdk.javadoc/share/classes/jdk/javadoc/internal/doclets/formats/html/MethodWriter.java line 106:
> 104:
> 105: for (Element method : methods) {
> 106: currentMethod = (ExecutableElement)method;
As a separate RFE, _it would be nice_ if `getVisibleMembers` took an optional `Class<T>` argument that is used to cast the result of the method to `List<T extends Element>` instead of just `Element`. This would enable to eliminate downstream casts, like here on line 106, to get the subtype.
See a similar technique in `Utils` for accessing sublists of a list of `DocTree` nodes,
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/15969#discussion_r1347805280
More information about the javadoc-dev
mailing list