[core][ios][osx][android] fix icons with non-integer width/height

[ansis]

ref #3031
ref #2198

For example, an icon that has:

• a pixel width of 10
• a pixel ratio of 3
• a scaled with of 3.333

is now supported.

SpriteImage is now constructed with the pixel width of the image instead of the width.

this works in android with @tobrun's https://github.com/mapbox/mapbox-gl-native/tree/3031-test-branch-android
@1ec5 I think this should fix #2198 - what's the best way to test that?

[jfirebaugh]

Approach here looks sound. I would like to continue by changing the SpriteImage constructor signature to:

SpriteImage(PremultipliedImage&&, float pixelRatio, bool sdf = false)

And replacing these SpriteImage members:

    const uint16_t width;
    const uint16_t height;
    const uint16_t pixelWidth;
    const uint16_t pixelHeight;
    const std::string data;

With:

const PremultipliedImage image;

[1ec5]

I think this should fix #2198 - what's the best way to test that?

Change this line to:

CGRect rect = CGRectMake(0, 0, 20, 15.5);

[1ec5]

Can you add a line to the Android and iOS sections of the changelog, too? Lots of folks are eager to see this fixed. 👏

[bleege]

Tagging myself here so that we can cherry pick this over to release-android-3.1.0 branch when this lands.

[ansis]

@jfirebaugh SpriteImage and PremultipliedImage changes implemented in 9cef5a7

[jfirebaugh]

This line needs to be moved to after the std::copy. (Actually I'm reading now that GetByteArrayRegion is a better API to use: http://developer.android.com/training/articles/perf-jni.html.)

[ansis]

@jfirebaugh thanks for finding all the bugs, they should be fixed now

[jfirebaugh]

👍

[bleege]

Should this be cherry picked into the release-android-v3.1.0 branch?

[ansis]

@bleege I think so. at least the first commit (26faa6a).

[tobrun]

I verified that this fixed the initial issue of having a rectangle image as marker icon

[bleege]

Cool! @tobrun confirmed that he tested from master and both 26faa6a and d34f8eb made it in there. I'll cherry pick both to release-android-v3.1.0 for today's release at #3537 .

Thanks everyone!

[bleege]

FYI... I fudged the cherry pick process for the first commit 26faa6a and ended up creating a new commit after resolving the changes instead of doing a git cherry-pick --continue hence 52fd9d7b505559972654a9f7af05d6be9c244f73. I did it the proper way for the second commit however. Discussed with @zugaldia this morning and we decided to leave this as is (and not deal with force pushes and other crazy git commands to be named later) and deal with the pain when the merge of release-android-v3.1.0 into master after release.

[danswick]

Huge! Our mobile SDK users will be pumped when this lands. 👏

[bleege]

@danswick FYI... we had to pull this from Android 3.1.0 today due to C++ Pain. It's now scheduled for 4.0.0.

[afathman]

When is the expected release of 4.0.0?

[bleege]

@afathman We're still planning the specifics around 4.0.0, but in the interim the fix for this issues is available immediately for use via the nighlty 4.0.0-SNAPSHOT builds.