EvenQueue: fix template functions passing UserAllocatedEvent<...> as argument. #11391
Conversation
Fix template functions passing UserAllocatedEvent<...> as argument.
|
Looks like ARMC6 has better template deduction algorithms? @kjbracey-arm |
|
@maciejbocianski, thank you for your changes. |
|
CI started |
|
@kjbracey-arm could you do a quick review ? |
Test run: FAILEDSummary: 1 of 4 test jobs failed Failed test jobs:
|
|
CI needs restarting , internal fault |
|
CI now restarted |
| template<typename... Args> | ||
| bool cancel(UserAllocatedEvent<Args...> *event) | ||
| template<typename F, typename A> | ||
| bool cancel(UserAllocatedEvent<F, A> *event) |
There was a problem hiding this comment.
Original code wasn't technically wrong, but I can see why ARMC5 might have got confused.
We're not actually instantiating anything here. UserAllocatedEvent takes two template arguments, and we're declaring something that could potentially give it variable.
There should only be a parameter mismatch if this was instantiated with Args... having anything other than 2 arguments. And that would never happen by function parameter deduction (there are no UserAllocatedEvents with anything other than 2 arguments to pass). It should only barf if explicitly called as cancel<int,int,int>(nullptr).
Presumably ARMC5 has some logic that crosschecks template arguments and parameters when a template is declared, and that's activating here, and thinking "variadic" doesn't match "2". That check may have been reasonable before variadic arguments. It catches this:
template <typename A, typename B>
class Foo;
template <typename A>
bool cancel(Foo<A> *event);
I don't actually know whether that is legal; and if it isn't, is a compiler is required to diagnose it if cancel is never instantiated?
Anyway, this is an example of the sort of ARMC5 glitch we're going to keep hitting. Its C++11 implementation is incomplete, and buggy. We'll have to keep working around it.
There was a problem hiding this comment.
Anyway, this is an example of the sort of ARMC5 glitch we're going to keep hitting. Its C++11 implementation is incomplete, and buggy. We'll have to keep working around it.
😭 👍
There was a problem hiding this comment.
I have section "11.13 C++11 supported features" from the ARM C 5.06 manual printed to 3 sheets of A4 and pinned to the side of my cubicle.
Recommended reading for anyone thinking of pushing their C++11 luck.
Test run: SUCCESSSummary: 11 of 11 test jobs passed |
Fix
EvenQueuetemplate functions passing UserAllocatedEvent<...> as argument.Description
Pull request type
Reviewers
Release Notes