package streamable
A collection of types suitable for incremental serialization
Install
Dune Dependency
Authors
Maintainers
Sources
v0.16.1.tar.gz
md5=c410b847f5a0f0be3c67b3403af04282
sha512=62cdeea4d38110bc6bd318e564e8277355928156c947400ee3e023998b60009a9d20073e8f7efedbe3bbada739bea56c6ff7d1eaba26898f6675ab690f03aadb
Description
A collection of types suitable for incremental serialization.
Published: 08 Aug 2023
README
README.mkd
# Streamable library The Streamable library contains a signature for types supporting incremental serialization. It acts as the "runtime library" for [`ppx_streamable`](https://github.com/janestreet/streamable/tree/master/ppx/doc), which may be easier to use than directly invoking the functors exposed by this library. This is useful for types that need to be serializable but may be too large to serialize in a single async cycle. Signature ========= The core of the Streamable is the signature of incrementally serializable types, found in the `Module_type` module. At a high-level, the signature includes * a type `t`, that you want to serialize * a type `part`, which is `bin_io`-able * a function `t -> part Sequence.t` * a set of functions/types for incrementally putting together a sequence of `part`s back into the original `t` It is expected that `t`s round-trip through this process; that is, turning a value into parts and then putting them back together should give an equal value. More in-depth documentation can be found alongside the type signature definition. Functors ======== The goal of the functors are to avoid the need to manually fulfill the streamable interface. In most cases they should be sufficient. The functors all fulfill the round-trip identity law, as long as their arguments do the same. All of the functors are available in `Stable` forms. A Stable module is guaranteed to not change over time. If we ever want to change how we serialize a particular type, we'll do so by adding a new stable version so that client code can upgrade on their own time. Of_atomic --------- The simplest provided functor is `Of_atomic`, which takes a `bin_io`-able type and returns the streamable signature on the same type. This is done in the trivial way, meaning that no benefit is gained in switching from `bin_io` to streamable if the streamable interface is satisfied with `Of_atomic`. Of_streamable ------------- The `Of_streamable` functor takes a streamable type, a non-streamable analagous type, and conversion functions between the two, and returns a streamable instance of the latter type. This is most commonly used to make a record streamable, by going through one of the `Of_tupleN` functors: ``` type t = {x : int; y : string} include Streamable.Remove_t (Streamable.Of_streamable (Streamable.Of_tuple2 (Streamable.Of_atomic (Int)) (Streamable.Of_atomic (String))) (struct type nonrec t = t let to_streamable {x; y} = (* *) (x, y) let of_streamable (x, y) = (* *) {x; y} end)) ``` Or, similarly, to make a variant streamable by going through one of the `Of_variantN` functors. Remove_t ---------- The `Remove_t` functor in the previous example is a helper that just removes the `type t` from the returned module, since there is already a `type t` in scope. It's often useful. For container types ------------------- Most of the remaining functors are used to lift one or more streamable types into a more complex streamable type by applying some type constructor. For example, a `string list` could be made streamable in the following way: ``` Streamable.Of_list (Streamable.Of_atomic (String)) ``` Almost all of the module signatures encountered in these functors are just `Streamable.S` with various types `t`. The exceptions are the module for the key type in `Of_map` and the module for the type in `Of_set`. Both of these only expect the module signature `Stable` (the module for the _key_, not the corresponding map/set); that is, they must be `bin_io`-able, and will be streamed atomically. This may not be what you want if values of those types can get too large to `bin_io` atomically. Fixpoint -------- The `Fixpoint` functor can be used to make recursive types streamable. See the `btree.ml` in the `streamable_examples` library. Packed ------ The `Packed` functor takes a streamable type and returns the same type, but with a new implementation of the streamable interface. It attempts to decrease the number of parts by collecting multiple parts from the original streamable into a single part until it reaches a size threshold (currently `2^17` bytes). Many of the functors for container types are already wrapped in `Packed`, so it is usually unnecessary to use this one outside of the library. This is most often useful if you are manually implementing the `Streamable` interface but the most natural way of doing so generates many small parts.
Dependencies (9)
-
ppxlib
>= "0.28.0"
-
dune
>= "2.0.0"
-
ppx_jane
>= "v0.16" & < "v0.17"
-
core_kernel
>= "v0.16" & < "v0.17"
-
core
>= "v0.16" & < "v0.17"
-
base
>= "v0.16" & < "v0.17"
-
async_rpc_kernel
>= "v0.16" & < "v0.17"
-
async_kernel
>= "v0.16" & < "v0.17"
-
ocaml
>= "4.14.0"
Dev Dependencies
None
Conflicts
None
sectionYPositions = computeSectionYPositions($el), 10)"
x-init="setTimeout(() => sectionYPositions = computeSectionYPositions($el), 10)"
>
On This Page