RFR: [S] 8211407: Bad links to non-existent entries on serialized-form page

[Jan Lahoda]

Looks good to me.

Jan

On 3.10.2018 23:48, Jonathan Gibbons wrote:
> Doh! I forgot to give a link to the updated webrev.
>
>   Updated webrev: http://cr.openjdk.java.net/~jjg/8211407/webrev.01/
>
> -- Jon
>
> On 10/03/2018 02:39 PM, Jonathan Gibbons wrote:
>> Jan,
>>
>> Thanks, but ... regrettably, the fix was a bit too clever and failed
>> one of the javadoc tests
>> that had annotations on a package declaration. It failed because javac
>> ended up trying
>> to add the annotations to the package twice, caused by calling
>> `.complete()` too soon.
>>
>> So, this version is less aggressive, and just completes the
>> package-info symbols
>> on the included packages, after all other setup processing has been done.
>>
>> This new test is unchanged, and this time all the javadoc tests pass.
>>
>> -- Jon
>>
>>
>> On 10/03/2018 12:07 PM, Jan Lahoda wrote:
>>> Looks good to me.
>>>
>>> Jan
>>>
>>> On 3.10.2018 02:24, Jonathan Gibbons wrote:
>>>> Please review a small fix, and medium test, that fixes hundreds(!) of
>>>> broken
>>>> links in the JDK API documentation.
>>>>
>>>> The broken links in question are generally all in the javax.swing.**
>>>> packages,
>>>> and are the "See Serialized Form" links in class documentation pages
>>>> to the
>>>> serialized-form page. The package-info files for those directories are
>>>> not read
>>>> in a timely manner, meaning that the "@serial exclude" tag in those
>>>> directories is
>>>> effectively ignored at the time when links to the serialized form page
>>>> are being
>>>> generated.  The package-info files are eventually lazily read, before
>>>> the serialized
>>>> form page is generated, and so the target of the (incorrect) link is
>>>> (correctly)
>>>> not generated, thus causing the broken link.
>>>>
>>>> Although it's a javadoc problem and fix, the fix is way down in the
>>>> javadoc
>>>> machinery that interacts with javac, in the JavadocEnter class. The fix
>>>> is to ensure that we call .complete() for the Symbol for the
>>>> package-info
>>>> for any package for which we read a source file. Completing that symbol
>>>> is enough to ensure that the package-info.java file is read in a timely
>>>> manner.
>>>>
>>>> A small combo test is written. The test is verified to fail without
>>>> the fix
>>>> and to succeed with the fix. The regression is specific to the use of
>>>> packages in modules, containing `@serial exclude` although the test
>>>> covers additional test cases.
>>>>
>>>> JBS: https://bugs.openjdk.java.net/browse/JDK-8211407
>>>> Webrev: http://cr.openjdk.java.net/~jjg/8211407/webrev.00/
>>>>
>>>> -- Jon
>>>>
>>>>
>>>>
>>>>
>>
>