[ios] Crash fix on render with view-backed annotations

[fabian-guerra]

Fixes #8163 while this is only a nil check, we need to continue investigating why annotationContext.annotation is nil when calling visibleAnnotationsInRect

[mention-bot]

@fabian-guerra, thanks for your PR! By analyzing this pull request, we identified @1ec5, @incanus and @boundsj to be potential reviewers.

[1ec5]

We should audit the MGLMapView implementation for places where we use subscript notation ([]) on _annotationContextsByAnnotationTag. The tricky thing about std::map is that subscripting with a nonexistent key will create a bogus value to go with the key. That could be how the context ends up with a nil annotation. In most places, we’re using std::map::at(), which throws an exception, and checking for existence (with count()) before calling at().

[fabian-guerra]

@1ec5 even before I added a nil check we did checked with count()

            if (!_annotationContextsByAnnotationTag.count(annotationTag))
            {
                continue;
            }
            MGLAnnotationContext annotationContext = _annotationContextsByAnnotationTag.at(annotationTag);
            if (annotationContext.annotation)
            {
                [annotations addObject:annotationContext.annotation];
            }

[1ec5]

even before I added a nil check we did checked with count()

Yes, but there are several other places in the code that use subscripting notation on _annotationContextsByAnnotationTag – could one of these places be responsible for the nil annotation? This PR is fine as a stopgap fix, but unless we also hunt down the root cause, we may well end up with bizarre behavior elsewhere in the annotation code and have even less of an idea what’s causing it.