Always report skipped feature hits for the original layer

[ahocevar]

In the process of getting to the bottom of #4822, it occurred to me why so many iterations of patching the select interaction fixed one issue, only to create another: hit detection of features should return the same features and layers, no matter be features selected or not.

I think this fix makes the affected code easier to read and understand, and turns the select interaction into one that could almost be loved.

Fixes #4822.

[marcjansen]

Nice branch name 😉

[tschaub]

Nice branch name 😉

Agreed.

Other things I like:

• Net 7 lines removed!
• And that is with something like 15 test lines added!

Things that make me scared:

• This code always breaks.
• This changes a test in a way that makes it seem like applications would break, but I'm so scared of the select interaction that I don't know if this is expected behavior or not.

In summary:

• Andreas is awesome for his perseverance with the select interaction.
• I hope that someone else who uses the select interaction can review this.
• It would be really nice to have a clear understanding of what was broken before and what is fixed now.

[ahocevar]

@tschaub I wrote a blog post about the whole story: http://ahocevar.net/openlayers/2016/02/17/trilogy-of-error.html. I hope it explains the whole evolution of these problems.

[ahocevar]

@tschaub:

This changes a test in a way that makes it seem like applications would break, but I'm so scared of the select interaction that I don't know if this is expected behavior or not.

There is no real need to make this change, so I changed it back to the way it was.

The key change of this pull request is that skipped features are now hit-detected the same way as if they had never been changed on their original layer. This means that they will be hit-detected on their original layer only, even when duplicated on one or more unmanaged layers.

[ahocevar]

It would be really nice to have a clear understanding of what was broken before and what is fixed now.

When a feature is selected, the Select interaction calls ol.Map#skipFeature() with the selected feature, and copies the feature to its internal unmanaged layer. Before this change, an already selected feature was reported by ol.Map#forEachFeatureAtPixel to come from a null layer. After this change, it is still reported to come from the layer that it was selected from. The Select interaction had a lot of code to deal with this change of layers for a selected feature, which could now be removed.

For user created unmanaged layers, nothing changes, because a user cannot call ol.Map#skipFeature().

[ischas]

I wrote a blog post about the whole story: http://ahocevar.net/openlayers/2015/02/17/trilogy-of-error.html.

@ahocevar So if you knew that all one year before (see publishing date), why didn't you fix it earlier? ;-)

[ahocevar]

Thanks @ischas - I shouldn't be writing blog posts at 2am. Updated blog post URL: http://ahocevar.net/openlayers/2016/02/17/trilogy-of-error.html.

[ischas]

@ahocevar 404! First link was good, but the publishing date was wrong! Maybe you must fix the blog post, too.

[ahocevar]

@ischas Took a few seconds to update on the server. Now it's live.

[jonataswalker]

Andreas is awesome for his perseverance with the select interaction.

No doubt. All of us are benefited at his cost.

[bjornharrtell]

Nice work @ahocevar, did some testing with an application of mine that makes good use of the Select interaction here and there and it works as expected.

This also makes #3453 easier to fix so I've rebased that on this PR.

[ahocevar]

Anyone brave enough to review this? @fredj or @tsauerwein maybe?

[fredj]

LGTM, thanks Andreas