RFR: JDK-8309566: Migrate away from TagletWriter and TagletWriterImpl

[Jonathan Gibbons]

On Mon, 10 Jul 2023 14:26:05 GMT, Pavel Rappo <prappo at openjdk.org> wrote:

>> Please review a medium-substantial cleanup of the internal `Taglet` world.
>> 
>> The initial motivation was to move tag-specific functionality from `TagletWriter[Impl]` to HTML-specific new subtypes of the individual `Taglet` classes, so that taglets are now better represented by a format-neutral base type and an HTML-specific subtype. The new subtypes are collected in a new `formats.html.taglets` package, and for the most part, they are accessed via their `Taglet` API.
>> 
>> A secondary motivation is to clean up handling of inline tags. Not all inline tags had corresponding taglets: most notably, `{@link}` was not modeled by a taglet.  Introducing `[Html]LinkTaglet` allowed more code to move from `TagletWriterImpl` to `HtmlLinkTaglet` ... and `HtmlSeeTaglet` and `HtmlSnippetTablet` now delegate to `HtmlLinkTaglet` to generate links. Also, in `HtmlDocletWriter` a notable visitor, in `commentTags toContent` had explicit `visit...` methods that effectively duplicated the functionality of `defaultAction`, so those methods can be and have been deleted.
>> 
>> Taglets used to be stateless, even though they are generally created once per doclet. (That was fixed, now, they _are_ created just once per doclet.) They now contain the standard long-lived members, like `configuration`, `utils`, `messages`, etc for the convenience of subtypes.  `TagletWriterImpl.Context` has always been effectively format-neutral and has been moved up to `Taglet.Context`.  
>> 
>> It had been hoped to replace the `TagletWriter` parameter to `Taglet::getInlineTagOutput` and `Taglet::getAllBlockTagOutput` with `Taglet.Context` perhaps calling with a HTML-subtype instance. But there is still enough useful functionality on `TagletWriter` that that is not practical at this time.
>> 
>> Taglets vary greatly in size, from small/trivial to large/complex. While it might seem unnecessary to use top-level classes for the small case, it seems better to go with a consistent uniform design rather than try and reduce any perceived overhead, perhaps by selectively using nested classes, as is often the case elsewhere in `jdk.javadoc` and `jdk.compiler`. Grouping the new `HTML...Taglet` classes in a new `formats.html.taglets` package seems like a good compromise.
>> 
>> **Note**: this is just "cleanup" and refactoring. There is no intentional change to any functionality, nor any added or removed.  If code appears to have been "deleted" it has either been moved elsewhere, or was effectively unused anyway. No test...
>
> src/jdk.javadoc/share/classes/jdk/javadoc/internal/doclets/toolkit/taglets/LinkTaglet.java line 38:
> 
>> 36: import jdk.javadoc.internal.doclets.toolkit.Content;
>> 37: 
>> 38: public abstract class LinkTaglet extends BaseTaglet {
> 
> `{@link}` has finally got its own taglet. However, I note that `{@code}` has lost its taglet.
> 
> In the markup sense, `{@link}` is to `{@code}` as `{@linkplain}` is to `{@literal}`. If you agree with this, then I think we should leave `{@code}`, rather than `{@literal}`.

I agree, but may do that separately, or as part of subsequent cleanup

> src/jdk.javadoc/share/classes/jdk/javadoc/internal/doclets/toolkit/taglets/TagletManager.java line 190:
> 
>> 188:         this.utils = configuration.utils;
>> 189:         this.tagletPath = options.tagletPath();
>> 190: //        initStandardTaglets();
> 
> A leftover?

oops yes.

> src/jdk.javadoc/share/classes/jdk/javadoc/internal/doclets/toolkit/taglets/TagletWriter.java line 107:
> 
>> 105:      * The context in which to generate the output for a series of {@code DocTree} nodes.
>> 106:      */
>> 107:     public final Context context;
> 
> I dislike it when a variable is accessible both directly and through the getter. Nonencapsulated fields like that is harder to debug and maintain. That said, I realise that it's somewhat of a pattern in jdk.javadoc. So, no need to change it at this time; I just make an observation.

Observation noted. For my part, I don't mind non-encapsulated fields if they are final.  I prefer that style to the "noisy" parens of getter methods:

var x = utils.something;
var x = utils().something;

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/14793#discussion_r1258476623
PR Review Comment: https://git.openjdk.org/jdk/pull/14793#discussion_r1258476968
PR Review Comment: https://git.openjdk.org/jdk/pull/14793#discussion_r1258479832