moka is too polymorphic (cargo-llvm-lines) #203
Comments
|
looks like I was a bit overeager. Apart from the one I found, most of the things are either due explicit |
|
I’m been playing a bit with this, trying to create a stresstest aimed at getting a hold of the compile times. use moka::future::Cache;
macro_rules! generate {
($($n:ident: $a:ty => $b: ty,)+) => {
struct State {
$($n: Cache<$a, $b>,)+
}
impl State {
fn new() -> Self {
Self {
$($n: Cache::builder().build(),)+
}
}
$(async fn $n(&self, k: $a) -> $b {
self.$n.get_with(k, async { 1 }).await
})+
}
fn main() {
let state = State::new();
let sum = futures_executor::block_on(async {
let mut sum = 0;
$(sum += state.$n(0).await as u64;)+
sum
});
println!("{sum}");
}
};
}
generate!(
cache_u8_u8: u8 => u8,
cache_u8_u16: u8 => u16,
cache_u8_u32: u8 => u32,
cache_u8_u64: u8 => u64,
cache_u16_u8: u16 => u8,
cache_u16_u16: u16 => u16,
cache_u16_u32: u16 => u32,
cache_u16_u64: u16 => u64,
cache_u32_u8: u32 => u8,
cache_u32_u16: u32 => u16,
cache_u32_u32: u32 => u32,
cache_u32_u64: u32 => u64,
cache_u64_u8: u64 => u8,
cache_u64_u16: u64 => u16,
cache_u64_u32: u64 => u32,
cache_u64_u64: u64 => u64,
);This testcase will generate up to 16 different instantiations of On my Windows system, I see a compile-time ( A run of Using I think most of the methods actually do need to have 16 different variants, though internally it might be a good idea to deduplicate some things. |
|
A quick check with my #265 PR: A debug build with 16 versions take ~9s, and a release build ~21s, so not that much of a difference.
Interestingly, |
|
great work |
I was playing around with https://github.com/dtolnay/cargo-llvm-lines recently, and found out that
mokadoes generate a ton of llvm IR due to too much internal generics.This is suboptimal for a couple of reasons:
Running
cargo-llvm-lineson a small test program:gives me this these functions on the top:
You can see that some functions are duplicated as often as 21 times.
I do have some ideas to reduce this bloat, so watch out for some patches ;-)
The text was updated successfully, but these errors were encountered: