Multithreaded cryptography #850
Conversation
this yields more efficient througput. probably.
| @@ -39,7 +39,7 @@ struct DemoCall : public abyss::http::IRPCClientHandler | |||
| bool HandleResponse(abyss::http::RPC_Response) override | |||
| { | |||
| llarp::LogInfo("response get"); | |||
| m_Logic->queue_func(m_Callback); | |||
| m_Logic->queue_func([=]() { m_Callback(); }); | |||
There was a problem hiding this comment.
I don't understand what this change achieves?
There was a problem hiding this comment.
storing a snapshot of the scope? I see this type of hack in JS a lot
There was a problem hiding this comment.
js is pass by reference, C++ is pass by value (unless explicitly noted otherwise)
| std::bit_xor< size_t >()), | ||
| std::bit_xor< size_t >()); | ||
| } | ||
| }; |
There was a problem hiding this comment.
you could use abseil hashing instead:
template <typename H>
H AbslHashValue(H h, const Path& p) {
return H::combine(std::move(h), p.hops[0].upstream, p.hops[0].rxID, p.hops[0].txID);
}(providing the sub-types were also hashable)
llarp/util/thread/logic.cpp
Outdated
| @@ -97,7 +109,7 @@ namespace llarp | |||
| bool | |||
| Logic::can_flush() const | |||
| { | |||
| return ourID == std::this_thread::get_id(); | |||
| return id.has_value() && id.value() == std::this_thread::get_id(); | |||
There was a problem hiding this comment.
this can just be id == std::this_thread::get_id()
| @@ -433,7 +433,8 @@ extern "C" | |||
| void | |||
| llarp_main_signal(struct llarp_main *ptr, int sig) | |||
| { | |||
| ptr->ctx->HandleSignal(sig); | |||
| ptr->ctx->logic->queue_func( | |||
| std::bind(&llarp::Context::HandleSignal, ptr->ctx.get(), sig)); | |||
There was a problem hiding this comment.
this is probably clearer as:
ptr->ctx->logic->queue_func([ctx=ptr->ctx.get(),sig]() { ctx->HandleSignal(sig); });
There was a problem hiding this comment.
for that matter, as long as queue_func takes an r-value reference it could be even cleaner to do:
auto func = <make the lambda>; logic->queue_func(func);
as 2 separate lines, but either way it's a style consideration.
this makes most of the cryptography done in worker threads.
includes iwp multiack and public key pinning because separating them would effectively mean rewriting that part so eh why not include it. this DOES break the protocol handshake but we're going to do that anyways.