Improve the deserializer compile time
I noticed that using rmp_serde to deserialize large structs has a considerable effect on the compile time. It also has twice the impact of comparable crates like serde_json when generating deserialization code.
The solution here is to take as much code as possible out of generic methods that are instantiated for each type. This reduces the LLVM IR lines from 3652241 to 3238664 (-11.3%) and the time to compile rmp_serde::from_slice::<MyBigStruct>(&[]) from 52s to 33s (-36%). For reference, doing the same with serde_json takes about 25s to compile.
Here are the LLVM stats of the worst offenders before and after the changes. any_inner is still pretty high up and with some more elbow grease it might be possible to reduce it more, but I don't feel comfortable going any further in this PR.
before
Lines Copies Function name
----- ------ -------------
3652241 71401 (TOTAL)
350636 (9.6%, 9.6%) 331 (0.5%, 0.5%) rmp_serde::decode::Deserializer<R,C>::any_inner
256153 (7.0%, 16.6%) 339 (0.5%, 0.9%) rmp_serde::decode::any_num
253829 (6.9%, 23.6%) 302 (0.4%, 1.4%) <serde_value::de::ValueDeserializer<E> as serde::de::Deserializer>::deserialize_any
174191 (4.8%, 28.3%) 935 (1.3%, 2.7%) serde::de::Visitor::visit_byte_buf
150146 (4.1%, 32.4%) 263 (0.4%, 3.0%) <serde_value::de::ValueDeserializer<E> as serde::de::Deserializer>::deserialize_struct
102226 (2.8%, 35.2%) 399 (0.6%, 3.6%) <serde_value::de::ValueDeserializer<E> as serde::de::Deserializer>::deserialize_identifier
69688 (1.9%, 37.2%) 1218 (1.7%, 5.3%) core::result::Result<T,E>::map_err
63733 (1.7%, 38.9%) 331 (0.5%, 5.8%) rmp_serde::decode::read_str_data
55013 (1.5%, 40.4%) 755 (1.1%, 6.8%) <serde::de::value::SeqDeserializer<I,E> as serde::de::SeqAccess>::next_element_seed
34399 (0.9%, 41.3%) 400 (0.6%, 7.4%) <serde::de::value::MapDeserializer<I,E> as serde::de::MapAccess>::next_key_seed
30805 (0.8%, 42.2%) 906 (1.3%, 8.7%) <T as erased_serde::ser::Serialize>::do_erased_serialize
27173 (0.7%, 42.9%) 1268 (1.8%, 10.4%) serde::de::Visitor::visit_newtype_struct
after
Lines Copies Function name
----- ------ -------------
3238664 71407 (TOTAL)
253829 (7.8%, 7.8%) 302 (0.4%, 0.4%) <serde_value::de::ValueDeserializer<E> as serde::de::Deserializer>::deserialize_any
175446 (5.4%, 13.3%) 331 (0.5%, 0.9%) rmp_serde::decode::Deserializer<R,C>::any_inner
174191 (5.4%, 18.6%) 935 (1.3%, 2.2%) serde::de::Visitor::visit_byte_buf
150146 (4.6%, 23.3%) 263 (0.4%, 2.6%) <serde_value::de::ValueDeserializer<E> as serde::de::Deserializer>::deserialize_struct
102226 (3.2%, 26.4%) 399 (0.6%, 3.1%) <serde_value::de::ValueDeserializer<E> as serde::de::Deserializer>::deserialize_identifier
69558 (2.1%, 28.6%) 1214 (1.7%, 4.8%) core::result::Result<T,E>::map_err
55013 (1.7%, 30.3%) 755 (1.1%, 5.9%) <serde::de::value::SeqDeserializer<I,E> as serde::de::SeqAccess>::next_element_seed
53908 (1.7%, 31.9%) 339 (0.5%, 6.4%) rmp_serde::decode::any_num
34399 (1.1%, 33.0%) 400 (0.6%, 6.9%) <serde::de::value::MapDeserializer<I,E> as serde::de::MapAccess>::next_key_seed
32777 (1.0%, 34.0%) 331 (0.5%, 7.4%) rmp_serde::decode::visit_str_data
30805 (1.0%, 35.0%) 906 (1.3%, 8.6%) <T as erased_serde::ser::Serialize>::do_erased_serialize
27173 (0.8%, 35.8%) 1268 (1.8%, 10.4%) serde::de::Visitor::visit_newtype_struct