On Wed, 3 Feb 2021 04:28:21 GMT, Jonathan Gibbons <j...@openjdk.org> wrote:
> Please review an update to improve the support, where appropriate, for nested
> inline tags.
>
> It has always been the case that certain inline tags allow text and HTML to
> appear within them. With tags like `{@return}` and `{@summary}` it becomes
> desirable to also generally allow nested inline tags to appear in those
> places as well. The work for this was started with the support for
> `{@return}` [JDK-8075778](https://bugs.openjdk.java.net/browse/JDK-8075778),
> but applying the work more generally was out of scope at the time. This
> change completes the work that was started then.
>
> The work can be grouped into 4 parts, in 3 commits.
>
> ## Commit 1
>
> * Update `DocCommentParser` to syntactically allow nested inline tags in
> situations that permit text and HTML
> * Update the downstream code to semantically limit nestg where it does not
> make sense, and provide new tests to verify the behavior.
>
> A family of new tests are added, each testing the ability to put an inline
> tag within another of the same kind, with and without doclint being enabled.
> In addition, test cases are added placing a simple instance `{@value}` in an
> enclosing tag: this is a useful case just because the expansion is plain text
> and therefore valid in all situations. Additional tests and test cases can
> be added as needed.
>
> This commit left the `{@index}` tag generating "bad" code when it was nested.
> The error was "reference to an undeclared ID". The (temporary) solution was
> to disable the automatic link checking for this specific subtest.
>
> ## Commit 2
>
> * `HtmlDocletWriter` and `TagletWriterImpl ` pass around a pair of booleans
> `isFirstSentence` and `inSummary` to help determine the output to be
> generated. Conceptually, a third value is added to that group: a set
> containing the set of nested tag kinds, so that it is possible to determine
> the enclosing tags for a tag. But, rather than add a third parameter to be
> passed around, the 3 are grouped into a new class `TagletWriterImpl.Context`
> which encapsulates the two booleans and the new set. The new class is added
> in a way to minimize code churn. No tests are affected by this change: all
> continue to pass.
>
> ## Commit 3
>
> * The new `Context#inTags` field is used to help improve the behavior of
> nested `{@index}` tags even when used incorrectly, with warnings disabled.
> As a result, the temporary change in the first commit to disable automatic
> link checking in one of the test cases is reverted.
>
> <hr>
>
> The introduction of the new Context class is arguably more general than we
> need at this time, but it clears up some erratic and inconsistent use of the
> `isFirstSentence` and `inSummary` booleans. The new class also provides a
> better framework for any complex new inline tags we may add in future. We
> might want to change the `Set<DocTree.Kind>` to some other collection at some
> point, if needs be (a stack, for example.) We might also want to move more
> information into the `Context`, such as the related `Element` that is
> otherwise ubiquitously passed around.
>
> The overall cleanup also revealed some latent bugs in the code, that were
> hidden in some of the tests. Most notable was that there were still some
> cases were `<` and `>` were not being correctly escaped as `<` and `>`
> leading to output in some tests of the form `List<String>` ! This triggered
> a minor cleanup/rewrite of the beginning of
> `HtmlDocletWriter.seeTagsToContent` which was previously a bit too liberal
> with the use of `new RawHtml`! The other minor cleanup was more consistent
> handling of whitespace at the end of the first sentence, as will be seen in a
> couple of places in one of the tests that was updated.
Although this patch generally looks reasonable, I'm surprised to see some of
the behavior it introduces. For example, consider composition of `@link` tags
from the included tests:
/** First sentence. {@link #m1() ABC {@link #m2() DEF} GHI} */
Before the patch that doc comment resulted in the following HTML:
First sentence. <a href="#m1()"><code>ABC {@link #m2() DEF} GHI</code></a>
After the patch that same doc comment results in HTML that looks like this:
First sentence. <a href="#m1()"><code>ABC <a
href="#m2()"><code>DEF</code></a> GHI</code></a>
I wouldn't expect that latter behavior. Not only nested `<a>` are invalid HTML,
but they are also rendered strangely in browsers. In this particular case the
trailing `GHI` appears as normal text, not as a hyperlink.
-------------
PR: https://git.openjdk.java.net/jdk/pull/2369
