RFR: 8236935: Improve UX of the search control

[Jonathan Gibbons]

Very nice!  thanks

-- Jon

On 7/10/20 12:34 PM, Hannes Wallnoefer wrote:
> Jon, I’m uploading the API docs without delay here:
>
> http://cr.openjdk.java.net/~hannesw/8236935/api.nodelay.01/
>
> Hannes
>
>> Am 10.07.2020 um 17:40 schrieb Jonathan Gibbons <jonathan.gibbons at oracle.com>:
>>
>> Hannes,
>>
>> Code looks OK.
>>
>> Can you upload API docs without the built-in delay?
>>
>> -- Jon
>>
>> On 7/10/20 6:40 AM, Hannes Wallnoefer wrote:
>>> Ping. I realized I hadn’t uploaded the webrev, just the rendered API docs with the delayed index loading, so here is the new webrev:
>>>
>>> https://cr.openjdk.java.net/~hannesw/8236935/webrev.01/
>>>
>>> We need to decide whether we want to get this into 15 or defer to 16.
>>>
>>> The most significant change outside of the UX fix is that all index files are generated all the time, even if they contain no items.
>>>
>>> Hannes
>>>
>>>> Am 29.06.2020 um 16:36 schrieb Hannes Wallnoefer <hannes.wallnoefer at oracle.com>:
>>>>
>>>> I uploaded a new set of API docs that keeps the „loading…“ message at the top of the result list until all the index files have been loaded.
>>>>
>>>> http://cr.openjdk.java.net/~hannesw/8236935/api.01/
>>>>
>>>> I also tweaked the delay for loading index files for testing purposes to 3 seconds for modules, packages, and search tags, 6 seconds for types, and 9 seconds for members.
>>>>
>>>> Try searchign for „logg“ to generate search results in all categories and see the effect of results being updated.
>>>>
>>>> I actually think this is better than my previous solution. If everybody agrees I’ll upload a new webrev with the changed behaviour.
>>>>
>>>> Hannes
>>>>
>>>>> Am 18.06.2020 um 22:25 schrieb Jonathan Gibbons <jonathan.gibbons at oracle.com>:
>>>>>
>>>>> Hannes,
>>>>>
>>>>> I played with the posted APIs again. It seems to be the case that the "loading" message only appeared briefly, for only as long as no search results are found. Once some results are found, even if not all have been found, the message goes away.   After the message goes away, the search results may still get updated when other files have been loaded.   It is reasonable to continue updates, but I think it would be nicer if the 'loading' message persisted until all the files have been loaded.
>>>>>
>>>>> The behavior I observe seems to be confirmed by this extract from the code:
>>>>>
>>>>> 311         if (result.length === 0) {
>>>>> 312             result.push(loading);
>>>>> 313         }
>>>>> Does it work to just remove the `if (expr)` ? It is already the case that this code is protected inside `if (!indexFilesLoaded()) ...`
>>>>>
>>>>> -- Jon
>>>>>
>>>>> On 6/10/20 2:44 PM, Hannes Wallnoefer wrote:
>>>>>> Please review a fix to display a „Loading search index…“ message in javadoc search while the search index files are loading and updating the results when loading is completed.
>>>>>>
>>>>>> JBS:
>>>>>> https://bugs.openjdk.java.net/browse/JDK-8236935
>>>>>>
>>>>>> Webrev:
>>>>>> http://cr.openjdk.java.net/~hannesw/8236935/webrev.00/
>>>>>>
>>>>>>
>>>>>> This introduces an updateSearchResults function in script.js that is empty by default. When a search is performed before the index files have been fully loaded, the updateSearchResults function is replaced with a version that will update the search results when an index file is loaded and evaluated.
>>>>>>
>>>>>> An invocation of updateSearchResults() is added at the end of each search index file. Also, as a consequence of this change all search index files are generated regardless of whether it is empty or not (provided the index option is enabled). This means a bit of added overhead for docs without modules or search tags, but the cost should be neglectible for a mostly empty script file.
>>>>>>
>>>>>> JDK API docs generated with this patch at the URL below. These docs also feature an extra 5 second delay for loading the search index files to make it easier to test the new feature.
>>>>>>
>>>>>>
>>>>>> http://cr.openjdk.java.net/~hannesw/8236935/api.00/
>>>>>>
>>>>>>
>>>>>> One thing you might notice is that sometimes  search results are updated multiple times - once for each search index file that is loaded. This is a bit annoying, and it also reminded me that loading each index in a separate file probably adds quite a bit of network and evaluation overhead.  As an experiment, I created a version that combines all search indices in one single file. I’m not proposing that for 15 since it’s a bigger change that shouldn’t be rushed, but I’m including it here as a possible next step.
>>>>>>
>>>>>> Single search index file (just for evaluation, not proposed for JDK 15):
>>>>>> Webrev:
>>>>>> http://cr.openjdk.java.net/~hannesw/8236935/webrev.singleindex.00/
>>>>>>
>>>>>> API docs:
>>>>>> http://cr.openjdk.java.net/~hannesw/8236935/api.singleindex.00/
>>>>>>
>>>>>>
>>>>>> Hannes
>>>>>>