feat: add method unswizzling #4647
Conversation
🚨 Detected changes in high risk code 🚨High-risk code can easily blow up and is hard to test. We had severe bugs in the past. Be extra careful when changing these files, and have an extra careful look at these:
|
|
🚨 Detected changes in high risk code 🚨High-risk code can easily blow up and is hard to test. We had severe bugs in the past. Be extra careful when changing these files, and have an extra careful look at these:
|
philprime
requested review from
philipphofmann,
brustolin and
armcknight
as code owners
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #4647 +/- ##
=============================================
+ Coverage 90.584% 90.633% +0.048%
=============================================
Files 621 621
Lines 71120 71201 +81
Branches 25306 25936 +630
=============================================
+ Hits 64424 64532 +108
+ Misses 6602 6573 -29
- Partials 94 96 +2
... and 26 files with indirect coverage changes Continue to review full report in Codecov by Sentry.
|
Performance metrics 🚀
|
| Revision | Plain | With Sentry | Diff |
|---|---|---|---|
| b9a9ffd | 1201.61 ms | 1215.06 ms | 13.45 ms |
| 282cc99 | 1238.27 ms | 1253.46 ms | 15.19 ms |
| 984eb2d | 1220.62 ms | 1235.24 ms | 14.62 ms |
| 4350d44 | 1228.75 ms | 1246.75 ms | 18.00 ms |
| 279841c | 1237.29 ms | 1254.12 ms | 16.83 ms |
| e145ca1 | 1207.19 ms | 1230.67 ms | 23.48 ms |
| 9b9f2a0 | 1233.36 ms | 1244.04 ms | 10.68 ms |
| 7b022df | 1220.53 ms | 1227.56 ms | 7.03 ms |
| aa225e2 | 1218.00 ms | 1236.45 ms | 18.45 ms |
| 9f0d9e0 | 1226.31 ms | 1242.70 ms | 16.40 ms |
App size
| Revision | Plain | With Sentry | Diff |
|---|---|---|---|
| b9a9ffd | 21.58 KiB | 418.43 KiB | 396.85 KiB |
| 282cc99 | 22.85 KiB | 414.09 KiB | 391.24 KiB |
| 984eb2d | 20.76 KiB | 425.77 KiB | 405.01 KiB |
| 4350d44 | 21.58 KiB | 629.82 KiB | 608.24 KiB |
| 279841c | 22.84 KiB | 403.19 KiB | 380.34 KiB |
| e145ca1 | 21.58 KiB | 669.70 KiB | 648.12 KiB |
| 9b9f2a0 | 21.58 KiB | 707.42 KiB | 685.84 KiB |
| 7b022df | 22.85 KiB | 412.95 KiB | 390.10 KiB |
| aa225e2 | 21.58 KiB | 694.47 KiB | 672.89 KiB |
| 9f0d9e0 | 21.58 KiB | 424.28 KiB | 402.70 KiB |
There was a problem hiding this comment.
FYI, we already tried that in the past and it always blew up the tests. We instead chose a different approach: We don't unswizzle, but we make sure the operations are NoOps when closing the SDK. I mean we can try again, but we need some good reasons. What drove you to open this PR?
|
@philipphofmann in #4634 I added unit tests initializing the SentrySDK with an experimental feature flag enabled or disabled. If the SentrySDK is enabled, and swizzling happens, the swizzling remains even after closing and re-enabling the SentrySDK without swizzling. We could also change this PR to only unswizzle in unit tests, i.e. using |
I agree with this!! |
🚨 Detected changes in high risk code 🚨High-risk code can easily blow up and is hard to test. We had severe bugs in the past. Be extra careful when changing these files, and have an extra careful look at these:
|
| @@ -164,7 +234,7 @@ + (BOOL)swizzleInstanceMethod:(SEL)selector | |||
| } | |||
| } | |||
|
|
|||
| swizzle(classToSwizzle, selector, factoryBlock); | |||
| swizzle(classToSwizzle, selector, factoryBlock, key); | |||
There was a problem hiding this comment.
m: The key could be nil or NULL here, if I'm not mistaken. We check for that below with if (key). The swizzle method would need to handle that.
| NSStringFromSelector(selector), | ||
| class_isMetaClass(classToUnswizzle) ? @"class" : @"instance", classToUnswizzle); | ||
|
|
||
| static pthread_mutex_t gLock = PTHREAD_MUTEX_INITIALIZER; |
There was a problem hiding this comment.
m: If I'm not mistaken, this is a different mutex from the one in Swizzle. Wouldn't it be better to share one mutex as they manipulate the same dict.
| NSValue *originalImplementationValue = [refsToOriginalImplementations objectForKey:keyValue]; | ||
| return (IMP)[originalImplementationValue pointerValue]; |
There was a problem hiding this comment.
m: What happens if the dict doesn't contain a value for the key?
| NSAssert(key != NULL, @"Key may not be NULL."); | ||
|
|
||
| if (key == NULL) { | ||
| NSLog(@"Key may not be NULL."); |
There was a problem hiding this comment.
l: I would prefer using SentryLog in here instead of NSLog.
| @@ -83,6 +83,21 @@ | |||
| _SentrySWWrapArg(SentrySWArguments), _SentrySWWrapArg(SentrySWReplacement), \ | |||
| SentrySwizzleMode, key) | |||
|
|
|||
| #if TEST || TESTCI | |||
| /** | |||
| * Unswizzles the instance method of the class. | |||
There was a problem hiding this comment.
m: Please also add a comment that we don't want to use this in production and only for tests explaining that this isn't safe to use in production.
|
|
||
| [SentryTestsLog clear]; | ||
|
|
||
| [SentrySwizzle unswizzleInstanceMethod:methodForUnswizzling |
There was a problem hiding this comment.
m: I would like to have some test cases for calling unswizzleInstanceMethod for non swizzled methods and calling unswizzleInstanceMethod twice.
📜 Description
I adapted the method swizzling helper to keep a reference to the original implementation, so that the
unswizzlemethods can revert swizzling.💡 Motivation and Context
We are introducing a new experimental flag in #4634 to conditionally swizzle Objective-C methods. Therefore some unit tests (e.g
testFoo) swizzle methods, while others (e.g.testBar) do not. If we do not offer a functionality to unswizzle methods, they are leaking isolation, i.e.testFooruns beforetestBartherefore methods are swizzled, even tough we don't want that.This is required for #4634 to pass tests.
💚 How did you test it?
I added a unit test to test the unswizzling.
📝 Checklist
You have to check all boxes before merging:
sendDefaultPIIis enabled.🔮 Next steps
#if TESTto not be used in production