1391266 - Feature: Tracking protection: verify startup and tab init perf is minimally affected

Status: RESOLVED WORKSFORME

(Firefox for iOS :: General, defect)

Description

[:garvan]

follow up to bug 1390148

Areas of potential performance hits:
- loading new JSON files into the rule store and 'compiling' and saving into the rule store
- rule store has all rules compiled and saved, verify loading from disk at app start is performant
- verify opening new tab and applying rule set changes (which should not require hitting the disk) is zero perf impact

Comment 1

[:garvan]

On an iPhone SE, to load the Strict blocklist set, the startup costs in the profiler are:

- Fresh install (parsing required): 340ms on a dedicated background thread. The function to look for is WebCore::ContentExtensions:compileRuleList()
-- overall startup time is not affected, we are main-thread bound startup. 

- Boot after parsed (WebKit loads the pre-parsed rules): I get no matches in the profiler for `ContentExtensions` during startup, or when creating a new tab that has content blockers. A simple profile filter for all `WebKit` symbols shows no significant difference between installing or not installing a content blocker, either on startup or the time to create a new tab. I am not concerned enough to investigate load-testing this by synthesizing a heavy list of rules for testing.

Lastly,
for additional interest, here is the startup time of 3 sessions in sequence with DYLD lib profiling (and I added a log message for when the web view begins loading, t0 for that timer is when main() is hit):

1) ContentBlocker disabled in code:
Total pre-main time: 1.2 seconds (100.0%)
         dylib loading time: 995.84 milliseconds (81.0%)
        rebase/binding time:  67.10 milliseconds (5.4%)
            ObjC setup time:  53.66 milliseconds (4.3%)
           initializer time: 111.86 milliseconds (9.1%)
           slowest intializers :
             libSystem.B.dylib :  10.23 milliseconds (0.8%)
    libMainThreadChecker.dylib :  31.56 milliseconds (2.5%)
app did finish launch time: 0.412301003932953 sec
app did start first navigation: 0.775330007076263 sec

2) ContentBlocker enabled, parse required:
Total pre-main time: 1.1 seconds (100.0%)
         dylib loading time: 1.0 seconds (87.6%)
        rebase/binding time:  34.89 milliseconds (3.0%)
            ObjC setup time:  32.93 milliseconds (2.8%)
           initializer time:  74.20 milliseconds (6.4%)
           slowest intializers :
             libSystem.B.dylib :   8.55 milliseconds (0.7%)
app did finish launch time: 0.334550023078918 sec
app did start first navigation: 0.63961797952652 sec

3) same as #1:
Total pre-main time: 1.0 seconds (100.0%)
         dylib loading time: 963.63 milliseconds (87.8%)
        rebase/binding time:  27.99 milliseconds (2.5%)
            ObjC setup time:  32.28 milliseconds (2.9%)
           initializer time:  72.90 milliseconds (6.6%)
           slowest intializers :
             libSystem.B.dylib :   8.58 milliseconds (0.7%)
app did finish launch time: 0.331308007240295 sec
app did start first navigation: 0.673924028873444 sec

Comment 2

[:garvan]

Closing bug, we have no facility to automate this testing. In future, as we add whole-app startup profiling to regression test we can use that test to watch for startup time regression with new features.