| SHA1 | ||
|---|---|---|
Test: TH Bug: 175662146 Change-Id: Ib7e2e61a987def59f8e8fc841d07f37ae894573d 
junyulai
commited |
0c813cc91e | |
Test: atest NetworkStaticLibTests Bug: 174123988 Change-Id: Id918227716367264042e6b674ce29c18e2d901e4
junyulai
commited |
148bdcb48b | |
There doesn't seem to be a way to get any code at all to run in a generic way on the thread that will execute a test in android JUnit. JUnit can do it, but the android harness imposes a thread hop. This is inconvenent with ReadHead, because ReadHead can't be used in multithread for semantic reasons : it would make no sense at all. ReadHead enforces this today in a relatively strong fashion. But technically, the only guarantee that is needed is that poll() is executed only on one thread in slabs of time defined between a read barrier and a write barrier, where the write barrier of the previous slab happen-before the read barrier of the next slab. This in generally speaking a bit difficult to achieve, but the tests run satisfy this condition by starting the thread after executing the @Before methods and joining it before executing the @After methods. So in fact, it's safe to use ReadHead during the befores and afters, if a bit uncomfortable, as long as only one thread uses it during the before, one thread during the test, and one thread during the afters. This patch relaxes the enforcement on ReadHead to accept the usage above. If any slab of time contains usage of poll() on multiple threads, this will instead crash with ConcurrentModificationException in almost all cases. It's no longer possible to detect every possible misuse of poll() under these conditions, but with this code it's going to be very unlikely at each test run that the code is incorrect and still doesn't crash, so incorrect use should still cause at least very severe flakiness. The main drawback, aside from the very slight performance penalty, is that some use can be accidentally correct and this code will now accept it. So it's possible that code that is only accidentally correct is accepted, and a seemingly unrelated change later breaks that correctness, causing seemingly unrelated tests to fail. Still, this is going to be an improvement in usability of ReadHead. We'll no longer need to have lazy creation of it, and it will be possible to check remaining events after a tests runs. Test: FrameworksNetTests NetworkStackTests Change-Id: If87eef1686b235e53cb444a32f9985b4c2a0a901 
Chalard Jean
commited |
460e78931a | |
* changes: Add utilities to test tap interface networks Add PacketFilter utilities for DNS testing 
Remi NGUYEN VAN
authored and
 Gerrit Code Review
commited |
edb36498eb | |
Add packet filters that are useful for filtering / testing DNS queries and ARP requests. This change is a prerequisite for tests verifying DNS and validation metrics. Test: tests using these utilities Bug: 160617623 Bug: 160656765 Change-Id: I4216f4ec60d1a427f60fad6eb8017e36b082665f
Remi NGUYEN VAN
commited |
2f2a3b467c | |
Nobody cares what language this is written in Test: builds Change-Id: I3ed3bce276479fd503e716a06fd4912da8c49eda
Chalard Jean
commited |
b1856fe06b | |
Test: builds Change-Id: I41979cc8096a682c6844df0825ee56ab00a828b5
Chalard Jean
commited |
e06c5c8fc0 | |
These files used to be in the network stack directory, but the libs
directory is a much more suitable place for them.
Also fix a typo : ConcurrentIntepreter → ConcurrentInterpreter
Also move {FdEvents,Packet}Reader to internal annotations. That's
what they should have been using in the first place anyway.
Note that this does not fix preupload issues reported by
checkstyle to make review easier. The fixes are in a followup
patch to this one.
Test: checkbuild
Change-Id: I675077fd42cbb092c0e6bd56571f2fc022e582fd
Chalard Jean
commited |
48c6c7d838 |