0
188
ViewsInconsistent behavior of kerning in NSAttributedString
I need to use Kern attribute of NSAttributedString. As I can see in the documentation, default value of that attribute is 0.0. But I faced with strange behaviour for phrase Hello, world (for phrase "Hello" all fine):
NSDictionary<NSString*, id>* attributes = @{NSFontAttributeName: [UIFont systemFontOfSize:12]};
NSString* text = @"Hello, World";
NSAttributedString* string = [[NSAttributedString alloc] initWithString:text attributes:attributes];
CGSize size1 = [string size];
NSMutableDictionary<NSString*, id>* attributesWithKernel = [attributes mutableCopy];
attributesWithKernel[NSKernAttributeName] = @(0.0);
NSAttributedString* stringWithKern = [[NSAttributedString alloc] initWithString:text attributes:attributesWithKernel];
CGSize size2 = [stringWithKern size];
XCTAssertTrue(CGSizeEqualToSize(size1, size2)); //here test falls
//size1 = size1 = (width = 68.8125, height = 14.3203125)
//size2 = (width = 69.515625, height = 14.3203125)
To make size1 and size2 equal, kerning should be equal -7.105427357601002e-15 , I know that this is very close to 0.0, but it is strange, because this changes the width almost a pixel.
NSAttributedString has same behaviour in Objective-C and in Swift, example for swift:
let text = "Hello, World"
let attributes : [NSAttributedString.Key : Any] = [NSAttributedString.Key.font: UIFont.systemFont(ofSize: UIFont.systemFontSize)]
let str = NSAttributedString(string: text, attributes: attributes)
let size = str.size()
var attributesWithKern = attributes
attributesWithKern[NSAttributedString.Key.kern] = NSNumber(value: 0.0)
let strWithKern = NSAttributedString(string: text, attributes: attributesWithKern)
let sizeWithKern = strWithKern.size()
XCTAssertTrue(size == sizeWithKern)
How I can fix this behaviour?
P.S.
Now I just remove NSKernAttributeKey from an attribute string if the key equal to 0.0, but I do not think that it is a good solution.
2 answers
Answer question0
I believe the docs here are wrong, and it is worth opening a radar about that. When no value is set, it is interpreted as "normal kerning." When 0 is set, it is interpreted as "disable kerning," which is why the width is a little wider (kerning typically is slightly negative, bringing kern-pair character, like "W" and "o" in this font, a little closer). I don't think there's any way to explicitly request "default kerning" without removing the attribute.
For your purposes, I believe you're doing the right thing by removing the value when it's zero, because you want default kerning, not to disable kerning.
The reason your tiny negative value is working is because it's not zero, so it's not disabling kerning, but it's so small that the behavior is very, very close to the default, and you're running into the precision of Double in the intermediate calculations (or possibly the precision of Float, depending on how it's implemented internally). You should find that your test passes for any value smaller (closer to zero) than this one, not just that value. In my tests, positive 7e-15 also works, for example.
0
I did some tests based on your code and I confirm this behaviour which looks like a bug. The point is for some letters strings are equal, for some not. For instance strings with "Hello Mo" or "Hello Oo" are equal, but for instance "Hello WoWoWo" differ a lot. So we see here some kerning being added for "W" without any reason. It can also depend on chosen font, I did not test it though. I see here the only solution that you've used - removing NSKernAttributeKey if it equals to 0.
