Source code

Revision control

Copy as Markdown

Other Tools

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
use anyhow::Result;
use askama::Template;
use heck::{ToShoutySnakeCase, ToSnakeCase, ToUpperCamelCase};
use serde::{Deserialize, Serialize};
use std::borrow::Borrow;
use crate::interface::*;
const RESERVED_WORDS: &[&str] = &[
"alias", "and", "BEGIN", "begin", "break", "case", "class", "def", "defined?", "do", "else",
"elsif", "END", "end", "ensure", "false", "for", "if", "module", "next", "nil", "not", "or",
"redo", "rescue", "retry", "return", "self", "super", "then", "true", "undef", "unless",
"until", "when", "while", "yield", "__FILE__", "__LINE__",
];
fn is_reserved_word(word: &str) -> bool {
RESERVED_WORDS.contains(&word)
}
/// Get the canonical, unique-within-this-component name for a type.
///
/// When generating helper code for foreign language bindings, it's sometimes useful to be
/// able to name a particular type in order to e.g. call a helper function that is specific
/// to that type. We support this by defining a naming convention where each type gets a
/// unique canonical name, constructed recursively from the names of its component types (if any).
pub fn canonical_name(t: &Type) -> String {
match t {
// Builtin primitive types, with plain old names.
Type::Int8 => "i8".into(),
Type::UInt8 => "u8".into(),
Type::Int16 => "i16".into(),
Type::UInt16 => "u16".into(),
Type::Int32 => "i32".into(),
Type::UInt32 => "u32".into(),
Type::Int64 => "i64".into(),
Type::UInt64 => "u64".into(),
Type::Float32 => "f32".into(),
Type::Float64 => "f64".into(),
Type::String => "string".into(),
Type::Bytes => "bytes".into(),
Type::Boolean => "bool".into(),
// API defined types.
// Note that these all get unique names, and the parser ensures that the names do not
// conflict with a builtin type. We add a prefix to the name to guard against pathological
// cases like a record named `SequenceRecord` interfering with `sequence<Record>`.
// However, types that support importing all end up with the same prefix of "Type", so
// that the import handling code knows how to find the remote reference.
Type::Object { name, .. } => format!("Type{name}"),
Type::Enum { name, .. } => format!("Type{name}"),
Type::Record { name, .. } => format!("Type{name}"),
Type::CallbackInterface { name, .. } => format!("CallbackInterface{name}"),
Type::Timestamp => "Timestamp".into(),
Type::Duration => "Duration".into(),
// Recursive types.
// These add a prefix to the name of the underlying type.
// The component API definition cannot give names to recursive types, so as long as the
// prefixes we add here are all unique amongst themselves, then we have no chance of
// acccidentally generating name collisions.
Type::Optional { inner_type } => format!("Optional{}", canonical_name(inner_type)),
Type::Sequence { inner_type } => format!("Sequence{}", canonical_name(inner_type)),
Type::Map {
key_type,
value_type,
} => format!(
"Map{}{}",
canonical_name(key_type).to_upper_camel_case(),
canonical_name(value_type).to_upper_camel_case()
),
// A type that exists externally.
Type::External { name, .. } | Type::Custom { name, .. } => format!("Type{name}"),
}
}
// Some config options for it the caller wants to customize the generated ruby.
// Note that this can only be used to control details of the ruby *that do not affect the underlying component*,
// since the details of the underlying component are entirely determined by the `ComponentInterface`.
#[derive(Debug, Clone, Default, Serialize, Deserialize)]
pub struct Config {
pub(super) cdylib_name: Option<String>,
cdylib_path: Option<String>,
}
impl Config {
pub fn cdylib_name(&self) -> String {
self.cdylib_name
.clone()
.unwrap_or_else(|| "uniffi".to_string())
}
pub fn custom_cdylib_path(&self) -> bool {
self.cdylib_path.is_some()
}
pub fn cdylib_path(&self) -> String {
self.cdylib_path.clone().unwrap_or_default()
}
}
#[derive(Template)]
#[template(syntax = "rb", escape = "none", path = "wrapper.rb")]
pub struct RubyWrapper<'a> {
config: Config,
ci: &'a ComponentInterface,
canonical_name: &'a dyn Fn(&Type) -> String,
}
impl<'a> RubyWrapper<'a> {
pub fn new(config: Config, ci: &'a ComponentInterface) -> Self {
Self {
config,
ci,
canonical_name: &canonical_name,
}
}
}
mod filters {
use super::*;
pub use crate::backend::filters::*;
pub fn type_ffi(type_: &FfiType) -> Result<String, askama::Error> {
Ok(match type_ {
FfiType::Int8 => ":int8".to_string(),
FfiType::UInt8 => ":uint8".to_string(),
FfiType::Int16 => ":int16".to_string(),
FfiType::UInt16 => ":uint16".to_string(),
FfiType::Int32 => ":int32".to_string(),
FfiType::UInt32 => ":uint32".to_string(),
FfiType::Int64 => ":int64".to_string(),
FfiType::UInt64 => ":uint64".to_string(),
FfiType::Float32 => ":float".to_string(),
FfiType::Float64 => ":double".to_string(),
FfiType::Handle => ":uint64".to_string(),
FfiType::RustArcPtr(_) => ":pointer".to_string(),
FfiType::RustBuffer(_) => "RustBuffer.by_value".to_string(),
FfiType::RustCallStatus => "RustCallStatus".to_string(),
FfiType::ForeignBytes => "ForeignBytes".to_string(),
FfiType::Callback(_) => unimplemented!("FFI Callbacks not implemented"),
// Note: this can't just be `unimplemented!()` because some of the FFI function
// definitions use references. Those FFI functions aren't actually used, so we just
// pick something that runs and makes some sense. Revisit this once the references
// are actually implemented.
FfiType::Reference(_) => ":pointer".to_string(),
FfiType::VoidPointer => ":pointer".to_string(),
FfiType::Struct(_) => {
unimplemented!("Structs are not implemented")
}
})
}
pub fn literal_rb(literal: &Literal) -> Result<String, askama::Error> {
Ok(match literal {
Literal::Boolean(v) => {
if *v {
"true".into()
} else {
"false".into()
}
}
// use the double-quote form to match with the other languages, and quote escapes.
Literal::String(s) => format!("\"{s}\""),
Literal::None => "nil".into(),
Literal::Some { inner } => literal_rb(inner)?,
Literal::EmptySequence => "[]".into(),
Literal::EmptyMap => "{}".into(),
Literal::Enum(v, type_) => match type_ {
Type::Enum { name, .. } => {
format!("{}::{}", class_name_rb(name)?, enum_name_rb(v)?)
}
_ => panic!("Unexpected type in enum literal: {type_:?}"),
},
Literal::Int(i, radix, _) => match radix {
Radix::Octal => format!("0o{i:o}"),
Radix::Decimal => format!("{i}"),
Radix::Hexadecimal => format!("{i:#x}"),
},
Literal::UInt(i, radix, _) => match radix {
Radix::Octal => format!("0o{i:o}"),
Radix::Decimal => format!("{i}"),
Radix::Hexadecimal => format!("{i:#x}"),
},
Literal::Float(string, _type_) => string.clone(),
})
}
pub fn class_name_rb(nm: &str) -> Result<String, askama::Error> {
Ok(nm.to_string().to_upper_camel_case())
}
pub fn fn_name_rb(nm: &str) -> Result<String, askama::Error> {
Ok(nm.to_string().to_snake_case())
}
pub fn var_name_rb(nm: &str) -> Result<String, askama::Error> {
let nm = nm.to_string();
let prefix = if is_reserved_word(&nm) { "_" } else { "" };
Ok(format!("{prefix}{}", nm.to_snake_case()))
}
pub fn enum_name_rb(nm: &str) -> Result<String, askama::Error> {
Ok(nm.to_string().to_shouty_snake_case())
}
pub fn coerce_rb(nm: &str, ns: &str, type_: &Type) -> Result<String, askama::Error> {
Ok(match type_ {
Type::Int8 => format!("{ns}::uniffi_in_range({nm}, \"i8\", -2**7, 2**7)"),
Type::Int16 => format!("{ns}::uniffi_in_range({nm}, \"i16\", -2**15, 2**15)"),
Type::Int32 => format!("{ns}::uniffi_in_range({nm}, \"i32\", -2**31, 2**31)"),
Type::Int64 => format!("{ns}::uniffi_in_range({nm}, \"i64\", -2**63, 2**63)"),
Type::UInt8 => format!("{ns}::uniffi_in_range({nm}, \"u8\", 0, 2**8)"),
Type::UInt16 => format!("{ns}::uniffi_in_range({nm}, \"u16\", 0, 2**16)"),
Type::UInt32 => format!("{ns}::uniffi_in_range({nm}, \"u32\", 0, 2**32)"),
Type::UInt64 => format!("{ns}::uniffi_in_range({nm}, \"u64\", 0, 2**64)"),
Type::Float32 | Type::Float64 => nm.to_string(),
Type::Boolean => format!("{nm} ? true : false"),
Type::Object { .. } | Type::Enum { .. } | Type::Record { .. } => nm.to_string(),
Type::String => format!("{ns}::uniffi_utf8({nm})"),
Type::Bytes => format!("{ns}::uniffi_bytes({nm})"),
Type::Timestamp | Type::Duration => nm.to_string(),
Type::CallbackInterface { .. } => {
panic!("No support for coercing callback interfaces yet")
}
Type::Optional { inner_type: t } => format!("({nm} ? {} : nil)", coerce_rb(nm, ns, t)?),
Type::Sequence { inner_type: t } => {
let coerce_code = coerce_rb("v", ns, t)?;
if coerce_code == "v" {
nm.to_string()
} else {
format!("{nm}.map {{ |v| {coerce_code} }}")
}
}
Type::Map { value_type: t, .. } => {
let k_coerce_code = coerce_rb("k", ns, &Type::String)?;
let v_coerce_code = coerce_rb("v", ns, t)?;
if k_coerce_code == "k" && v_coerce_code == "v" {
nm.to_string()
} else {
format!(
"{nm}.each.with_object({{}}) {{ |(k, v), res| res[{k_coerce_code}] = {v_coerce_code} }}"
)
}
}
Type::External { .. } => panic!("No support for external types, yet"),
Type::Custom { .. } => panic!("No support for custom types, yet"),
})
}
pub fn check_lower_rb(nm: &str, type_: &Type) -> Result<String, askama::Error> {
Ok(match type_ {
Type::Object { name, .. } => {
format!("({}.uniffi_check_lower {nm})", class_name_rb(name)?)
}
Type::Enum { .. }
| Type::Record { .. }
| Type::Optional { .. }
| Type::Sequence { .. }
| Type::Map { .. } => format!(
"RustBuffer.check_lower_{}({})",
class_name_rb(&canonical_name(type_))?,
nm
),
_ => "".to_owned(),
})
}
pub fn lower_rb(nm: &str, type_: &Type) -> Result<String, askama::Error> {
Ok(match type_ {
Type::Int8
| Type::UInt8
| Type::Int16
| Type::UInt16
| Type::Int32
| Type::UInt32
| Type::Int64
| Type::UInt64
| Type::Float32
| Type::Float64 => nm.to_string(),
Type::Boolean => format!("({nm} ? 1 : 0)"),
Type::String => format!("RustBuffer.allocFromString({nm})"),
Type::Bytes => format!("RustBuffer.allocFromBytes({nm})"),
Type::Object { name, .. } => format!("({}.uniffi_lower {nm})", class_name_rb(name)?),
Type::CallbackInterface { .. } => {
panic!("No support for lowering callback interfaces yet")
}
Type::Enum { .. }
| Type::Record { .. }
| Type::Optional { .. }
| Type::Sequence { .. }
| Type::Timestamp
| Type::Duration
| Type::Map { .. } => format!(
"RustBuffer.alloc_from_{}({})",
class_name_rb(&canonical_name(type_))?,
nm
),
Type::External { .. } => panic!("No support for lowering external types, yet"),
Type::Custom { .. } => panic!("No support for lowering custom types, yet"),
})
}
pub fn lift_rb(nm: &str, type_: &Type) -> Result<String, askama::Error> {
Ok(match type_ {
Type::Int8
| Type::UInt8
| Type::Int16
| Type::UInt16
| Type::Int32
| Type::UInt32
| Type::Int64
| Type::UInt64 => format!("{nm}.to_i"),
Type::Float32 | Type::Float64 => format!("{nm}.to_f"),
Type::Boolean => format!("1 == {nm}"),
Type::String => format!("{nm}.consumeIntoString"),
Type::Bytes => format!("{nm}.consumeIntoBytes"),
Type::Object { name, .. } => format!("{}.uniffi_allocate({nm})", class_name_rb(name)?),
Type::CallbackInterface { .. } => {
panic!("No support for lifting callback interfaces, yet")
}
Type::Enum { .. } => {
format!(
"{}.consumeInto{}",
nm,
class_name_rb(&canonical_name(type_))?
)
}
Type::Record { .. }
| Type::Optional { .. }
| Type::Sequence { .. }
| Type::Timestamp
| Type::Duration
| Type::Map { .. } => format!(
"{}.consumeInto{}",
nm,
class_name_rb(&canonical_name(type_))?
),
Type::External { .. } => panic!("No support for lifting external types, yet"),
Type::Custom { .. } => panic!("No support for lifting custom types, yet"),
})
}
}
#[cfg(test)]
mod test_type {
use super::*;
#[test]
fn test_canonical_names() {
// Non-exhaustive, but gives a bit of a flavour of what we want.
assert_eq!(canonical_name(&Type::UInt8), "u8");
assert_eq!(canonical_name(&Type::String), "string");
assert_eq!(canonical_name(&Type::Bytes), "bytes");
assert_eq!(
canonical_name(&Type::Optional {
inner_type: Box::new(Type::Sequence {
inner_type: Box::new(Type::Object {
module_path: "anything".to_string(),
name: "Example".into(),
imp: ObjectImpl::Struct,
})
})
}),
"OptionalSequenceTypeExample"
);
}
}
#[cfg(test)]
mod tests;