Optimize NSRegularExpression performance - ios

Question

My app has a feed with posts that can contain URL's, #hashtags and #mentions. I display them using a pod called ActiveLabel. This pod does a good job, but my feed slightly lags when scrolling. My feed is a UICollectionView, and the cell generation is slightly lagging. I profiled my app when scrolling, and analysed the lag spikes. The lag is almost unnoticeable, but it annoys me.
As you can see, the main offender is the NSRegularExpression search.
I tried to optimize this slightly by disabling data detection when there is no instances of the data type, using .contains(). This made it marginally faster, but the lag spikes remains.
let enabledTypes:[ActiveType] = {
var types = [ActiveType]()
if ad.caption.current.string.contains("#") { types.append(.hashtag) }
if ad.caption.current.string.contains("#") { types.append(.mention) }
if ad.caption.current.string.contains("://") { types.append(.url) }
if ad.caption.canExpand { types.append(seeMore) }
return types
}()
label.enabledTypes = enabledTypes
I also followed every step in this article, which slighty helped, but not enough. So I need to fix the regex.
The regex statements ActiveLabel uses is
static let hashtagPattern = "(?:^|\\s|$)#[\\p{L}0-9_]*"
static let mentionPattern = "(?:^|\\s|$|[.])#[\\p{L}0-9_]*"
static let urlPattern = "(^|[\\s.:;?\\-\\]<\\(])" + "((https?://|www\\.|pic\\.)[-\\w;/?:#&=+$\\|\\_.!~*\\|'()\\[\\]%#,☺]+[\\w/#](\\(\\))?)" + "(?=$|[\\s',\\|\\(\\).:;?\\-\\[\\]>\\)])"
and it uses them with
static func getElements(from text: String, with pattern: String, range: NSRange) -> [NSTextCheckingResult]{
guard let elementRegex = try? NSRegularExpression(pattern: pattern, options: [.caseInsensitive]) else { return [] }
return elementRegex.matches(in: text, options: [], range: range)
}
Is searched around for other regexes to detect hashtags and mentions, but I didn't find anything that made a difference.
I tried to layout the label on a background thread, but that obviously crashed because UI doesn't like being done on a background thread. I could rewrite ActiveLabel to work mainly on background threads where I can and use callbacks instead of return types, but I'd like to avoid that.
Some samples of strings that I detect data on:
"Arnie says, Aspen. Str. Small. Varm og god jakke. Veldig fin på! Fremstår ubrukt. Kun brukt et par ganger, rett og slett fordi jeg har alt for mange jakker🙈 #urban #arnie #says #aspen #ubrukt"
"Skjorte pent brukt i organisk bomull fra tom tailor originalpris 300kr #organisk #bomullsjorte #bomull #flower #floral"
"Jean Paul genser i 100% ull, pent brukt✨ er i str.m, men veldig liten, passer xs-s! \n #jeanpaul #genser #classy #litebrukt #brun #ull"
As you can see, our users mainly hashtag stuff, so that one is the most important one.
Is there any way I can improve either NSRegularExpression or the regex statements to avoid the performance hit?

Answer 1

As #raidfive suggests, most likely your best course of action here is to create one or more NSRegularExpression instances ahead of time and reuse them whenever needed.
Note that since it's the creation/compiling of regexes that makes the biggest difference in your time profile (at least, in as much of the time profile as you've shared), caching regexes may win you back enough performance that you no longer need your intermediate optimization of enabling only the detection elements you need. In that case, you need only one regex (representing detection of all possible element types), so caching/reuse is easy.
Note furthermore that your intermediate "optimization" may not actually improve performance to begin with — it might even harm performance. Matching a regex, however complicated, requires searching the entire string in its entirety (roughly) once. Trying to decide which element types to detect means searching the string multiple times — once for each contains("#") (etc) test, then once more to evaluate the string against the regex. Repeated string searches might well cost more than the compilation of a single regex.
If you find after implementing the single cached universal regex that you're still (somehow) hamstrung on regex performance, you can cache multiple regexes, one for each search scenario you're processing. The combinatorics presumably work out such that you still have far fewer different regexes than you have strings to process, so if you compile them all before the user even starts scrolling, you're not paying the time cost of compiling them during scrolling. Per the previous paragraph, though, this only makes sense if you have a cheap (i.e. not string search) way of detecting which regex you need for each string.

Answer 2

You could try creating and storing the NSRegularExpression instances in a variable (class or instance) so you're only creating them once.