Make features and annotations conform to NSSecureCoding #6559

frederoni · 2016-10-03T11:52:48Z

This PR fixes #6200

All feature and annotation classes now conform to NSSecureCoding which means they can be serialized and deserialized. This is useful for UIStateRestoring among other things.

@incanus @1ec5 👀

mention-bot · 2016-10-03T11:52:50Z

@frederoni, thanks for your PR! By analyzing the history of the files in this pull request, we identified @incanus, @1ec5 and @boundsj to be potential reviewers.

1ec5 · 2016-10-03T16:48:07Z

platform/darwin/src/MGLPointAnnotation.m

+    {
+        return NO;
+    }
+    if (other == self)


This should be the first check for better performance.

1ec5 · 2016-10-03T16:48:44Z

platform/darwin/src/MGLPointAnnotation.m

+    [coder encodeDouble:coordinate.longitude forKey:@"coordinateLongitude"];
+}
+
+- (BOOL)isEqual:(id)other


If you override -isEqual:, you also need to override -hash.

1ec5 · 2016-10-03T16:51:46Z

Individual shape subclasses need custom implementations of these methods, because they can't be represented by only one coordinate. The MGLFeature classes also need overrides to check for attribute dictionary equality.

boundsj · 2016-10-13T22:53:56Z

Individual shape subclasses need custom implementations of these methods, because they can't be represented by only one coordinate. The MGLFeature classes also need overrides to check for attribute dictionary equality.

The geoJSONGeometryDictionary method added to MGLShape and geoJSONFeatureDictionary method added to MGLFeature coming in #6524 may help.

frederoni · 2016-12-08T15:49:16Z

MGLFeature is going to be quite bloated. Might split it up into one file per class.

frederoni · 2016-12-08T16:04:17Z

platform/darwin/src/MGLFeature.mm

+    if (self)
+    {
+        NSSet<Class> *identifierClasses = [NSSet setWithArray:@[NSString.class, NSNumber.class]];
+        identifier = [decoder decodeTopLevelObjectOfClasses:identifierClasses forKey:@"identifier" error:nil];


Only available on iOS 9

frederoni · 2016-12-09T16:13:44Z

platform/darwin/src/MGLMultiPoint.mm

@@ -5,6 +5,37 @@
 #include <mbgl/util/geo.hpp>
 #include <mbgl/util/optional.hpp>

+bool compareByCoordinate(CLLocationCoordinate2D lhs, CLLocationCoordinate2D rhs)
+{
+    return lhs.latitude+lhs.longitude < rhs.latitude+rhs.longitude;


This means coordinate{0,10} is the same as {10,0} but it's only used for sorting and as long as lhs and rhs is sorted the same way it shouldn't matter.

1ec5 · 2016-12-10T01:18:49Z

platform/darwin/src/MGLTypes.h

+
+#define MGLIsEqual(lhs,rhs) ((!lhs && !rhs) || [lhs isEqual:rhs])
+#define MGLIsEqualToString(lhs,rhs) ((!lhs && !rhs) || [lhs isEqualToString:rhs])
+#define MGLIsEqualToDictionary(lhs,rhs) ((!lhs && !rhs) || [lhs isEqualToDictionary:rhs])


This SDK should not get into the business of providing general-purpose Foundation equality macros to developers.

1ec5 · 2016-12-10T01:22:01Z

platform/darwin/src/MGLTypes.h

@@ -18,6 +18,12 @@
 #define MGLColor NSColor
 #endif

+#define TARGET_OS_MACOS (TARGET_OS_MAC && !(TARGET_OS_IOS || TARGET_OS_TV || TARGET_OS_WATCH))


Let’s stick to a simple !TARGET_OS_IPHONE check at each call site for now. By the time we introduce support for tvOS and/or watchOS, we’ll’ve dropped support for macOS 10.10 and will be able to use the standard TARGET_OS_OSX macro anyways.

1ec5 · 2016-12-10T01:26:04Z

platform/darwin/src/MGLMultiPoint.mm

+{
+    if (self == other) return YES;
+
+    MGLMultiPoint *otherMultipoint = other;


What if other is an instance of some other class that implements coordinates as an NSSet?

1ec5 · 2016-12-10T01:27:38Z

platform/darwin/src/MGLMultiPoint.mm

+    }
+
+    return ([super isEqual:otherMultipoint]
+            && _coordinates == otherCoords);


This could be written as _coordinates == otherCoords->_coordinates to avoid having to build a copy of the vector in the first place.

1ec5 · 2016-12-10T01:29:22Z

platform/darwin/src/MGLMultiPoint.mm

+    MGLMultiPoint *otherMultipoint = other;
+
+    CLLocationCoordinate2D *otherCoordinates = [otherMultipoint coordinates];
+    std::vector<CLLocationCoordinate2D> otherCoords;


This can be written as otherCoords = { otherCoordinates, [otherMultiPoint pointCount] } to copy the array with the same efficiency as a memcpy().

1ec5 · 2016-12-10T01:30:42Z

platform/darwin/src/MGLMultiPoint.mm

+
+    NSMutableArray *coordinates = [NSMutableArray array];
+    for (auto coord : _coordinates) {
+        [coordinates addObject:@[@(coord.latitude), @(coord.longitude)]];


We should never store a coordinate pair as an array if we can help it. Instead, represent it as a dictionary so that the order is never ambiguous.

Better yet, stuff the CLLocationCoordinate2D in an NSValue, which already conforms to NSSecureCoding.

@encode CLLocationCoordinate2D would work as long as we don't use it in combination with NSKeyed(Un)Archiver. It only gets the layout of the struct but no information about the name of the fields which is causing the archiver to raise an exception when trying to archive structs. Will make it unambiguous by representing it as a dictionary instead.

CLLocationCoordinate2D does work with keyed archivers and unarchivers, as long as you use NSValue to wrap it. Archiving a series of CLLocationCoordinate2Ds would be more efficient than archiving a series of dictionaries, and I don't think there's any ambiguity with that approach because we don't define an MGLSwappedCoordinate2D struct anywhere.

1ec5 · 2016-12-12T16:49:06Z

platform/darwin/src/MGLShape.mm

+{
+    if (lhs.size() != rhs.size()) return false;
+
+    std::sort(lhs.begin(), lhs.end(), compareByCoordinate);


Order matters. Two coincident lines, one of which goes in the opposite direction as the other, are not equal.

1ec5 · 2016-12-14T16:47:09Z

platform/darwin/src/MGLPointCollection.mm

-            bounds.extend(mbgl::LatLng(coordinate.latitude, coordinate.longitude));
-        }
-        _overlayBounds = MGLCoordinateBoundsFromLatLngBounds(bounds);
+        [self setupBounds];


It would be nice if we could compute these bounds lazily the first time overlayBounds gets called, as in MGLMultiPoint.

1ec5 · 2016-12-14T16:49:53Z

platform/darwin/src/MGLFeature.mm

+
+    MGLPolygonFeature *otherPointFeature = other;
+    return ([super isEqual:other]
+            && ((!self.geoJSONDictionary && !otherPointFeature.geoJSONDictionary) || [self.geoJSONDictionary isEqualToDictionary:otherPointFeature.geoJSONDictionary]));


Have you considered implementing feature and geometry equality as operators on mapbox::geometry classes in geometry.hpp or mbgl? That way other platforms could take advantage of your logic, and we could avoid roundtripping to Objective-C and incurring Objective-C's dynamic dispatch overhead.

The equality operator on geometry is already implemented 🙈

1ec5 · 2016-12-14T16:54:00Z

platform/darwin/src/MGLShape.mm

+    return lhs.latitude != rhs.latitude || lhs.longitude != rhs.longitude;
+}
+
+bool operator==(std::vector<CLLocationCoordinate2D>& lhs, std::vector<CLLocationCoordinate2D>& rhs)


Is this operator even necessary? As long as the == operator is defined on CLLocationCoordinate2D, == should just work on a vector of them.

1ec5 · 2016-12-14T16:55:07Z

platform/darwin/src/MGLShape_Private.h

@@ -1,6 +1,11 @@
 #import "MGLShape.h"

 #import <mbgl/util/geometry.hpp>
+#import <mbgl/util/geo.hpp>
+
+bool operator!=(const CLLocationCoordinate2D lhs, const CLLocationCoordinate2D rhs);


It's customary to put operators like this entirely in a header and inline them.

1ec5 · 2016-12-14T16:59:08Z

platform/ios/src/MGLAnnotationImage.m

+- (void)encodeWithCoder:(NSCoder *)coder {
+    [coder encodeObject:_image forKey:@"image"];
+    [coder encodeObject:_reuseIdentifier forKey:@"reuseIdentifier"];
+    [coder encodeObject:_styleIconIdentifier forKey:@"styleIconIdentifier"];


This identifier is generated by MGLMapView and mbgl, so I'm not sure that it makes sense to encode it.

1ec5 · 2016-12-14T17:07:47Z

platform/darwin/src/NSArray+MGLAdditions.mm

@@ -23,4 +23,29 @@ @implementation NSArray (MGLAdditions)
    return vector;
 }

+ (NSArray *)mgl_coordinatesFromCoordinates:(std::vector<CLLocationCoordinate2D>)coords {


Let's categorize NSCoder to have -encodeMGLLocationCoordinate2D:forKey: and -decodeMGLLocationCoordinate2DForKey:, as has been done for other structs like CGSize and UIEdgeInsets.

1ec5 · 2016-12-15T16:51:14Z

platform/darwin/src/MGLFeature.mm

+- (void)encodeWithCoder:(NSCoder *)coder {
+    [super encodeWithCoder:coder];
+    [coder encodeObject:identifier forKey:@"identifier"];
+    [coder encodeObject:attributes forKey:@"attributes"];


We can reduce the repetitiveness of implementing -initWithCoder:, -encodeWithCoder:, and -isEqual: across all the MGLFeature-conforming classes using a macro.

The macro will be obsolete after #7454 but let's use it in the meantime.

1ec5 · 2016-12-15T17:00:35Z

platform/darwin/src/MGLMultiPoint.mm

+{
+    NSUInteger hash = [super hash];
+    for (auto coord : _coordinates) {
+        hash += coord.latitude+coord.longitude;


Normally we box doubles in NSNumber and take its hash. The assumption is that NSNumber’s hash is more collision-proof than the raw number itself. Unfortunately, that could be less performant than just adding the coordinate. I wonder if we could take the hash of an NSValue of the coordinate instead of the NSNumbers of the latitude and longitude separately, to halve the cost.

1ec5 · 2016-12-15T18:04:07Z

platform/darwin/src/NSCoder+MGLAdditions.h

+
+@interface NSCoder (MGLAdditions)
+
+- (void)mgl_encodeLocationCoordinates2D:(std::vector<CLLocationCoordinate2D>)coordinates forKey:(NSString *)key;


I think it would be useful to publicly expose a method that encodes a single CLLocationCoordinate2D. It would be called -encodeMGLCoordinate:, for consistency with existing category methods on NSCoder. (See also -[NSValue valueWithMGLCoordinate:].)

1ec5 · 2016-12-15T18:07:33Z

platform/darwin/src/NSCoder+MGLAdditions.mm

+@implementation NSCoder (MGLAdditions)
+
+- (void)mgl_encodeLocationCoordinates2D:(std::vector<CLLocationCoordinate2D>)coordinates forKey:(NSString *)key {
+    [self encodeObject:[NSArray mgl_coordinatesFromCoordinates:coordinates] forKey:key];


Now that we have dedicated methods for working with CLLocationCoordinate2Ds, let’s encode the result of -[NSValue(MGLAdditions) valueWithMGLCoordinate:] instead of a dictionary.

I use NSKeyedArchiver for testing NSCoding which can't encode structs, not even structs wrapped in NSValue encoded with @encode(CLLocationCoordinate2D). ([NSKeyedArchiver encodeValueOfObjCType:at:]: this archiver cannot encode structs) Any other idea how to solve this? UIStateRestoring will probably still work if we switch to NSValue but a solution that works with archiving as well would be preferred.

I stand corrected. Apparently the root cause is that CLLocationCoordinate2D is a typedef to an anonymous struct, so @encode can't do anything useful.

We could hard-code the type string {dd}, as many Swift developers have had to do, or we could define our own non-anonymous struct based on CLLocationCoordinate2D solely for the purpose of archiving, or we could eat the performance cost and use dictionaries here. I haven't profiled this code, so I don't know at what point performance becomes a problem or whether we're OK with performance degrading after a certain point.

A non-anonymous struct sounded like the best option but still no luck trying to archive it.

Yeah, looks like you’re right. Unfortunately we’ll have to stick to an array of dictionaries. If performance does turn out to be a major issue, we can flatten the array, at the cost of triggering another debate about latitude-longitude order versus longitude-latitude order (or is it longitude-latitude order versus latitude-longitude order?).

1ec5 · 2016-12-21T11:34:41Z

platform/darwin/src/MGLFeature_Private.h

@@ -37,3 +37,31 @@ mbgl::Feature mbglFeature(mbgl::Feature feature, id identifier, NSDictionary *at
 NS_DICTIONARY_OF(NSString *, id) *NSDictionaryFeatureForGeometry(NSDictionary *geometry, NSDictionary *attributes, id identifier);

 NS_ASSUME_NONNULL_END
+
+#define MGLFeatureInitCoding() \


Macros should be named in SCREAMING_SNAKE_CASE. 📣🐍 In this case, since this macro is defining a method, it should be named with a verb, like MGL_DEFINE_FEATURE_INIT_WITH_CODER().

1ec5 · 2016-12-21T11:35:48Z

platform/darwin/src/MGLFeature_Private.h

+
+#define MGLFeatureInitCoding() \
+- (instancetype)initWithCoder:(NSCoder *)decoder { \
+if (self = [super initWithCoder:decoder]) { \


Indent the macro to aid with readability. This line should start with two tabs.

1ec5 · 2016-12-21T11:39:04Z

platform/darwin/src/MGLFeature_Private.h

+- (BOOL)isEqual:(id)other { \
+if (other == self) return YES; \
+if (![other isKindOfClass:[self class]]) return NO; \
+klass *otherFeature = other; \


typeof(self) also works in place of klass *, since we’ve already asserted that other is of the same kind as self.

1ec5 · 2016-12-21T11:40:59Z

platform/darwin/src/NSCoder+MGLAdditions.h

+
+@interface NSCoder (MGLAdditions)
+
+- (void)encodeMGLCoordinate:(CLLocationCoordinate2D)coordinate forKey:(NSString *)key;


Publicly vending these methods would’ve made more sense if we could come up with a way to encode and decode the coordinate as a struct. But with the dictionary-based implementation is dependent on the keys we choose, so unfortunately it isn’t portable enough to expose publicly. MGLCode+MGLAdditions_Private.h can be folded into this header, which can be marked with Project visibility. Sorry for the detour.

1ec5

Looks good. Just one nit regarding indentation.

1ec5 · 2016-12-21T16:20:23Z

platform/darwin/src/MGLFeature_Private.h

@@ -37,3 +37,31 @@ mbgl::Feature mbglFeature(mbgl::Feature feature, id identifier, NSDictionary *at
 NS_DICTIONARY_OF(NSString *, id) *NSDictionaryFeatureForGeometry(NSDictionary *geometry, NSDictionary *attributes, id identifier);

 NS_ASSUME_NONNULL_END
+
+#define MGL_DEFINE_FEATURE_INIT_WITH_CODER() \
+- (instancetype)initWithCoder:(NSCoder *)decoder { \


Also indent this line and the closing brace.

frederoni added iOS Mapbox Maps SDK for iOS macOS Mapbox Maps SDK for macOS labels Oct 3, 2016

1ec5 suggested changes Oct 3, 2016

View reviewed changes

frederoni changed the title ~~[ios] WIP Conform to NSSecureCoding~~ [ios] MGLPointAnnotation should conform to NSSecureCoding Oct 4, 2016

1ec5 added bug annotations Annotations on iOS and macOS or markers on Android labels Oct 4, 2016

boundsj mentioned this pull request Oct 13, 2016

MGLPointAnnotation Hashable conformance #6677

Closed

1ec5 added this to the ios-3.4.1 milestone Dec 5, 2016

1ec5 changed the title ~~[ios] MGLPointAnnotation should conform to NSSecureCoding~~ Make MGLPointAnnotation conform to NSSecureCoding Dec 5, 2016

1ec5 mentioned this pull request Dec 6, 2016

Annotation classes don’t conform to NSCoding #6200

Closed

frederoni changed the base branch from master to release-ios-v3.4.0 December 8, 2016 12:37

frederoni force-pushed the 6200-nssecurecoding branch from 139b6d9 to 1530ef5 Compare December 8, 2016 12:37

frederoni added the ⚠️ DO NOT MERGE Work in progress, proof of concept, or on hold label Dec 8, 2016

frederoni changed the title ~~Make MGLPointAnnotation conform to NSSecureCoding~~ Make features and annotations conform to NSSecureCoding Dec 8, 2016

frederoni force-pushed the 6200-nssecurecoding branch from fdcfa8b to a781215 Compare December 8, 2016 16:03

frederoni commented Dec 8, 2016

View reviewed changes

frederoni force-pushed the 6200-nssecurecoding branch 3 times, most recently from 2ffda25 to ec1ea25 Compare December 9, 2016 16:07

frederoni commented Dec 9, 2016

View reviewed changes

frederoni force-pushed the 6200-nssecurecoding branch from ec1ea25 to 7ed7782 Compare December 9, 2016 16:18

1ec5 suggested changes Dec 10, 2016

View reviewed changes

frederoni force-pushed the 6200-nssecurecoding branch 4 times, most recently from 7ad498e to 8dfaa0b Compare December 12, 2016 15:31

1ec5 suggested changes Dec 12, 2016

View reviewed changes

frederoni force-pushed the 6200-nssecurecoding branch 5 times, most recently from 8ac1a08 to 9b37141 Compare December 14, 2016 14:08

1ec5 suggested changes Dec 14, 2016

View reviewed changes

frederoni force-pushed the 6200-nssecurecoding branch from fc9e1b4 to 1b60d6a Compare December 15, 2016 15:17

1ec5 suggested changes Dec 15, 2016

View reviewed changes

frederoni force-pushed the 6200-nssecurecoding branch from f0d4f70 to e7fe451 Compare December 19, 2016 13:43

1ec5 suggested changes Dec 21, 2016

View reviewed changes

1ec5 approved these changes Dec 21, 2016

View reviewed changes

[ios, macos] features and annotations now conforms to NSSecureCoding

0625d4b

frederoni force-pushed the 6200-nssecurecoding branch from 6c5c826 to 0625d4b Compare December 22, 2016 12:31

frederoni merged commit 2c7569b into release-ios-v3.4.0 Dec 22, 2016

frederoni deleted the 6200-nssecurecoding branch December 22, 2016 12:58

friedbunny assigned frederoni Dec 22, 2016

friedbunny removed the ⚠️ DO NOT MERGE Work in progress, proof of concept, or on hold label Dec 22, 2016

1ec5 modified the milestones: ios-v3.4.0, ios-3.4.1 Dec 22, 2016

1ec5 mentioned this pull request Jan 11, 2017

Fix miscellaneous static analyzer warnings #7670

Merged

boundsj mentioned this pull request Jan 17, 2017

When instantiated from nib, MGLAnnotationView doesn't call the delegate methods when tapped #7747

Closed

This was referenced Mar 10, 2017

Two annotations with identical coordinates but different titles are considered equal #8351

Closed

Second polyline with identical coordinates not rendered #7811

Closed

1ec5 mentioned this pull request Feb 9, 2018

RFC: Feature identity for interactivity mapbox/mapbox-gl-js#6019

Closed


		@interface NSCoder (MGLAdditions)

		- (void)mgl_encodeLocationCoordinates2D:(std::vector<CLLocationCoordinate2D>)coordinates forKey:(NSString *)key;


		@interface NSCoder (MGLAdditions)

		- (void)encodeMGLCoordinate:(CLLocationCoordinate2D)coordinate forKey:(NSString *)key;

Make features and annotations conform to NSSecureCoding #6559

Make features and annotations conform to NSSecureCoding #6559

Conversation

frederoni commented Oct 3, 2016 • edited Loading

mention-bot commented Oct 3, 2016

Choose a reason for hiding this comment

Choose a reason for hiding this comment

1ec5 commented Oct 3, 2016

boundsj commented Oct 13, 2016

frederoni commented Dec 8, 2016 • edited Loading

frederoni Dec 8, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

1ec5 Dec 16, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

1ec5 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

frederoni commented Oct 3, 2016 •

edited

Loading

frederoni commented Dec 8, 2016 •

edited

Loading

frederoni Dec 8, 2016 •

edited

Loading

1ec5 Dec 16, 2016 •

edited

Loading