tmt96 / rs-object-space Goto Github PK

Agent is our main method of providing simple multithreading. Each Agent should run on a separate thread & be tied to one ObjectSpace. Ideally, it should be a thin wrapper around ObjectSpace functionality.

Agent API is currently unfinished, mostly due to trouble specifying different associated types for Agent, AgentRange, and AgentKey. We should finalize before alpha preview.

After #7 , performance regresses by about 30%. We want to investigate how to revert this performance regression

Currently the naming of the API is unclear (e.g: ObjectSpaceRange, ObjectSpaceKey,...) We should replace it with more descriptive naming (e.g: RangedLookupSpace)

Currently we have two versions for single object retrieval methods (read and take): one is non-blocking and returns Option<T> (try_read). The other is blocking until an object satisfying condition is found (read). We could add a third version that block until a timeout and returns Option<T> (e.g: read_timeout): if timeout elapsed and no object satisfying the condition is found we returns None

Float in Rust does not implement PartialOrd, causing problems for implementing Range functions for TreeobjectSpace, which relies on BTreeMap. Converting float Range to Range<NotNaN<f64>> should solve the problem.

Currently upon storing objects TreeObjectSpace flattens the result Value from serde_json serialization to support field range & value lookup.

e.g: Given this struct

#[derive(Serialize, Deserialize)]
struct Pagination {
    limit: u64,
    offset: u64,
    total: u64,
}

#[derive(Serialize, Deserialize)]
struct Users {
    users: Vec<User>,
    pagination: Pagination,
}

The resulting Value resembles:

Value {
    "users": Value(Array),
    "pagination.limit": Value(Number),
    "pagination.offset": Value(Number),
    "pagination.total": Value(Number)
}

A recent release from serde includes a flatten flag. using the flag for Pagination gives:

    "users": Value(Array),
    "limit": Value(Number),
    "offset": Value(Number),
    "total": Value(Number)

We may investigate how we could use this flag to simplify our serialization process.

Questions:

• What happens if a struct has two child struct, each as a field of the same name, i.e count?
• How could we apply this flag for each field in a struct?

Currently TreeObjectSpace does not support Enum. As Enum is an integral part of Rust (e.g: Option, Result, etc), supporting for Enum is crucial.

This might require rethinking field name checking in Entry as different subtypes in an enum may contain different field names. Refer https://serde.rs/json.html for more information on how enum is represented in serde-json