Crates.io | skrillax-serde |
lib.rs | skrillax-serde |
version | 0.2.0 |
source | src |
created_at | 2024-04-21 17:02:15.568563 |
updated_at | 2024-10-27 10:50:37.952636 |
description | A serialization/deserialization library for Silkroad Online packets. |
homepage | |
repository | https://git.eternalwings.de/tim/skrillax-network |
max_upload_size | |
id | 1215527 |
size | 20,646 |
serde is a well known and respected serialization and deserialization framework for Rust. And
while we could make it work for our context, Silkroad is unfortunately a little annoying in a few respects, which
would make working with serde
a little cumbersome. Thus, this provides a more focused approach to
serialization/deserialization in the Silkroad context.
Silkroad does a few things that make it hard to generally serialize the packets:
For example, there are three types if list representations in Silkroad: Length based, Break based, or Continue based.
The first is quite obvious; a length is prepended and then all elements follow without any separator. "Break based"
and "Continue based" are similar in that there's a separator in between each element which differs on the last
element. Both use 1
as the separator, but will use 2
or 0
as the end element, respectively. Why do you need
multiple ways to encode a list of values? I don't know. Other weirdness includes some strings using wide character
(2 byte) encoding or the content completely changing depending on what type of item it belongs to.
This is not impossible to implement with serde, in fact, the implementation would be easy, but would be more
exhausting to use given you'd have to constantly reach for deserialize_with
and remembering the exact function that
does what you want. Especially once you have to serialize/deserialize a field that depends on a previous field,
you're going to have a hard time (Example: a ChatMessage
packet has an optional field that depends on the type of
message which has been encoded in a previous, non-adjacent field). To make it easier to work with and make these quirks
"implementable", this custom 'serde' exists.
Most of the serialization/deserialization logic is implemented in the -derive
crate. This crate only includes ser/de definitions for primitive values, such as time, u8
and the like. However,
I'd like to explain the basics of how data is serialized - in general - in the Silkroad world.
Let's begin with the basics: Silkroad encodes data (mostly) in little endian byte order, so a packet length of 256
,
which is a u16
, will show up as 0x00 0x01
instead of 0x01 0x00
. There are no separators between data fields,
so values are present directly adjacent to each other. If you have three fields of varying size, the serialized size
will always match the total space the fields would use up in memory. Arrays of constant length also encode only the
data it contains (the number of elements is know and so is the size of each element). Varying length lists (or in
Rust terms: Vec
s) have, as previously mentioned, three different ways of being encoded. Either the length is
prepended, which is usually of size u8
, but may also be u16
, and then all elements will follow without any
further separators. Alternatively, a u8
is used to show if there's more (with a value of 1
) followed by the next
element or if it's the end with either 0
or a 2
, depending on where it's used. Enum-like elements (for example,
error codes or, generally, different variants of something) often contain a "variant id" denoted as a u8
, followed
by the specific fields for that variant. As such, variants that don't contain any data would only have the variant
id and nothing else. Lastly, there are optional fields, which contain a u8
specifying if data is present (value of
1
) or no data (value of 0
). Everything else uses any combination of the previously defined atoms.