2 unstable releases

new 0.2.1 Nov 4, 2024
0.2.0 Nov 4, 2024
0.1.0 Feb 15, 2024

#40 in Value formatting

Download history 638/week @ 2024-07-17 1204/week @ 2024-07-24 946/week @ 2024-07-31 850/week @ 2024-08-07 2071/week @ 2024-08-14 831/week @ 2024-08-21 1373/week @ 2024-08-28 1220/week @ 2024-09-04 922/week @ 2024-09-11 1275/week @ 2024-09-18 504/week @ 2024-09-25 316/week @ 2024-10-02 1556/week @ 2024-10-09 1628/week @ 2024-10-16 194/week @ 2024-10-23 1526/week @ 2024-10-30

4,948 downloads per month
Used in 13 crates (via strict_types)

Apache-2.0

15KB
100 lines

Vesper: declarative human-readable structural language, better than XML

Build Tests Lints codecov

crates.io Docs Apache-2 licensed

Overview

Sometimes we need to represent multidimensional data hierarchies in a very laconic style, which will be easily-readable for humans. At the same time, we do not want to lose precision, and have this representation formal and deterministic.

Up to day, the choice had to be made between JSON, YAML and XML. JSON lacks multidimensionality and, while being flexible, lacks proper efficient way of defining data formal structure (schema). Yes, JSON schemata exists, but are very clunky - like XML. XML has more tools to formalize the data structure, but is quite hard to visually parse. YAML is the most parsable, it also supports more structure than JSON - but still less than XML, and again, lacks efficient schema languages.

Thus, we created Vesper: formal language which is visually clean, and at the same time has the same power as XML:

Transaction rec
    Version enum U8 V1=1 V2=2
    Inputs list 0..MAX64
        PrevOut rec
            Txid bytes 32
            Vout as U32
        Sequence as U32
        ScriptSig bytes 0..MAX64
        Witness list 0..MAX64
            ByteStr bytes 0..MAX64
    Outputs list 0..MAX64
        Value as U64
        ScriptPubkey bytes 0..MAX64
    LockTime as U32

The above is the full representation of both data type hierarchy, semantic structure and memory layout for a Bitcoin transaction: this is how clean it can be. Written in JSON, it would take at least twice more text, plenty of quotation marks, brackets and braces. XML equivalent is also much more verbose and less readable:

<rec name="Transaction">
    <enum name="Version" type="U8">
        <variant tag="V1" value="1"/>
        <variant tag="V2" value="2"/>
    </enum>
    <list name="Inputs" min-len="0" max-len="MAX64">
        <rec name="PrevOut">
            <bytes name="Txid" len="32"/>
            <field name="Vout" type="U32"/>
        </rec>
        <field name="Sequence" type="U32"/>
        <bytes name="ScriptSig" min-len="0" max-len="MAX64"/>
        <list name="Witness" min-len="0" max-len="MAX64">
            <bytes name="ByteStr" min-len="0" max-len="MAX64"/>
        </list>
    </list>
    <list name="Outputs" min-len="0" max-len="MAX64">
        <field name="Value" type="U64"/>
        <bytes name="ScriptPubkey" min-len="0" max-len="MAX64"/>
    </list>
    <field name="LockTime" type="U32"/>
</rec>

Vesper uses new clause notation, called T-expressions. You may be aware of [S-expressions] and [M-expressions], and T-expressions are the new guest in the town, which follows a specific visually-clean pattern to represent semantic constructs: subject predicate attributes \ content, where subject-predicate-attributes go in one line, forming a kind of sentence, and content is nested below and can span multiple lines. Each line of the content is in fact another T-expression, and we end up with a tree (here is why the expression is called "T", i.e. "Tree-expression").

The full grammar of Vesper is so simple, that its formal definition can fit just nine lines (here we use our custom BNF-styled notation):

t-expr => subject predicate `{` ( attr )* `}`
            ( t-expr )\*

subject => ident
predicate => ident
attr => simple | named
simple => ident | expr
named => ident `=` ( ident | expr )

ident => (\w_)[\w\d_-]*
expr => [\S]+ -- all ASCII printable symbols except of whitespace

Vesper comes with its own schema language (named "Vesper schema"), which allows creating domain-specific sublanguages, like we did for the data type and memory layout above:

rec := - (*)
tuple := - (*)
as := \type (-)
enum := \type? \ident+=\int (-)
union := \type? (*)
bytes := \range (-)
list := \range (+)
char := - (-)
str := \range (-)

As you see, it is also ultra-concise: the whole definition for an arbitrary data type hierarchy and memory layout fits in just nine lines!

Contributing

CONTRIBUTING.md

License

The libraries are distributed on the terms of Apache 2.0 license.

Dependencies

~2.5MB
~55K SLoC