nim-ffi/ffi/codegen/rust_native.nim
Ivan FB 99fdcfdff7
feat(codegen): native Rust generator handles seq / Option / nested
Extends the native Rust marshalling to sequences and optionals (the Rust
counterpart of the cpp_native increment): a field maps to Vec<T> / Option<T>,
the repr(C) mirror gains the matching `{ ptr, len }` / `{ present, value }`
fields, and `to_c` now returns a holder owning the CStrings + C-array backing
(Vec/CString live on the heap, so the C struct's raw pointers survive the move
and the call). `from_c` reads seq/Option back out via slice + present-flag.

Unblocks the timer's complex (seq-of-structs / seq-of-strings / two optionals)
and schedule (three struct params). Verified end-to-end — the demo round-trips
typed ComplexResponse / ScheduleResult.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-05-31 18:59:57 +02:00

408 lines
18 KiB
Nim

## Native (zero-serialization) Rust binding generator.
##
## Emits a `<lib>_native` crate that wraps the *native* C ABI (the `<name>`
## entry points + flat C-POD structs) — no CBOR. Each `{.ffi.}` type is a
## `#[repr(C)]` mirror (`ffi`) plus an idiomatic Rust struct with `to_c`/`from_c`
## (`types`), and `<Lib>Node` marshals typed args in / reads typed struct
## returns out (`api`). Counterpart of the CBOR generator in `rust_cbor.nim`, and
## the Rust analogue of `cpp_native.nim`.
##
## Commit 1 covers scalar / string / bool / float / nested-struct fields and the
## procs that use only those (the timer's create / version / echo). Sequences,
## optionals and native events are layered on next.
import std/[os, strutils]
import ./meta, ./string_helpers
proc isSeqT(t: string): bool =
t.strip().startsWith("seq[") and t.strip().endsWith("]")
proc isOptT(t: string): bool =
let s = t.strip()
(s.startsWith("Option[") or s.startsWith("Maybe[")) and s.endsWith("]")
proc isStringT(t: string): bool =
t.strip() in ["string", "cstring"]
proc isStructT(t: string, types: seq[FFITypeMeta]): bool =
for ty in types:
if ty.name == t.strip():
return true
false
proc seqElem(t: string): string =
t.strip()["seq[".len .. ^2].strip()
proc optElem(t: string): string =
let s = t.strip()
let p = if s.startsWith("Maybe["): "Maybe[".len else: "Option[".len
s[p .. ^2].strip()
proc rustScalar(t: string): string =
case t.strip()
of "int", "int64", "clong": "i64"
of "int32", "cint": "i32"
of "int16": "i16"
of "int8": "i8"
of "uint", "uint64", "csize_t": "u64"
of "uint32", "cuint": "u32"
of "uint16": "u16"
of "uint8", "byte": "u8"
of "bool": "bool"
of "float", "float32": "f32"
of "float64": "f64"
else: capitalizeFirstLetter(t.strip()) # nested struct -> its idiomatic name
proc rustIdiomatic(t: string): string =
let s = t.strip()
if isSeqT(s): "Vec<" & rustIdiomatic(seqElem(s)) & ">"
elif isOptT(s): "Option<" & rustIdiomatic(optElem(s)) & ">"
elif isStringT(s): "String"
else: rustScalar(s)
proc rustCElem(t: string, types: seq[FFITypeMeta], qualify = false): string =
## repr(C) type for one seq element / option payload (matches emitCStructs).
## `qualify` prefixes struct names with `ffi::` (needed outside the ffi module).
let s = t.strip()
if isStringT(s): "*const c_char"
elif s == "bool": "c_int"
elif isStructT(s, types): (if qualify: "ffi::" else: "") & capitalizeFirstLetter(s)
else: rustScalar(s)
proc procSimple(p: FFIProcMeta, types: seq[FFITypeMeta]): bool =
## All scalar / string / struct params are supported (structs may carry
## seq/Option fields). Bare seq/Option top-level params and raw pointers are
## not (the native ABI doesn't expose them either).
for ep in p.extraParams:
let t = ep.typeName.strip()
if ep.isPtr or isSeqT(t) or isOptT(t):
return false
true
# ── ffi.rs ──────────────────────────────────────────────────────────────────
proc emitFfiRs(
procs: seq[FFIProcMeta], types: seq[FFITypeMeta], libName: string
): string =
var L: seq[string] = @[]
L.add("// Generated by nim-ffi native Rust codegen. Do not edit by hand.")
L.add("#![allow(non_snake_case, dead_code)]")
L.add("use std::os::raw::{c_char, c_int, c_void};")
L.add("")
L.add("pub type FFICallback =")
L.add(" unsafe extern \"C\" fn(ret: c_int, msg: *const c_char, len: usize, user_data: *mut c_void);")
L.add("")
# repr(C) POD mirrors — field layout mirrors codegen/c.emitCStructs.
for t in types:
L.add("#[repr(C)]")
L.add("#[derive(Clone, Copy)]")
L.add("pub struct " & capitalizeFirstLetter(t.name) & " {")
for f in t.fields:
let snake = camelToSnakeCase(f.name)
let ft = f.typeName.strip()
if isSeqT(ft):
L.add(" pub " & snake & ": *const " & rustCElem(seqElem(ft), types) & ",")
L.add(" pub " & snake & "_len: usize,")
elif isOptT(ft):
L.add(" pub " & snake & "_present: c_int,")
L.add(" pub " & snake & ": " & rustCElem(optElem(ft), types) & ",")
else:
L.add(" pub " & snake & ": " & rustCElem(ft, types) & ",")
L.add("}")
L.add("")
L.add("#[link(name = \"" & libName & "\")]")
L.add("extern \"C\" {")
for p in procs:
case p.kind
of FFIKind.CTOR:
var ps: seq[string] = @[]
for ep in p.extraParams:
ps.add(camelToSnakeCase(ep.name) & ": " & rustCElem(ep.typeName, types))
ps.add("callback: FFICallback")
ps.add("user_data: *mut c_void")
L.add(" pub fn " & p.procName & "(" & ps.join(", ") & ") -> *mut c_void;")
of FFIKind.FFI:
if not procSimple(p, types):
continue
var ps = @["ctx: *mut c_void", "callback: FFICallback", "user_data: *mut c_void"]
for ep in p.extraParams:
ps.add(camelToSnakeCase(ep.name) & ": " & rustCElem(ep.typeName, types))
L.add(" pub fn " & p.procName & "(" & ps.join(", ") & ") -> c_int;")
of FFIKind.DTOR:
L.add(" pub fn " & p.procName & "(ctx: *mut c_void) -> c_int;")
L.add("}")
return L.join("\n")
# ── types.rs (idiomatic structs + to_c / from_c) ─────────────────────────────
proc emitTypesRs(types: seq[FFITypeMeta]): string =
var L: seq[string] = @[]
L.add("// Generated by nim-ffi native Rust codegen. Do not edit by hand.")
L.add("#![allow(dead_code)]")
L.add("use std::ffi::{CStr, CString};")
L.add("use std::os::raw::c_char;")
L.add("use super::ffi;")
L.add("")
L.add("fn cstr(p: *const c_char) -> String {")
L.add(" if p.is_null() { String::new() }")
L.add(" else { unsafe { CStr::from_ptr(p) }.to_string_lossy().into_owned() }")
L.add("}")
L.add("")
for t in types:
let rn = capitalizeFirstLetter(t.name)
let hn = rn & "C"
# idiomatic struct
L.add("#[derive(Clone, Debug, Default)]")
L.add("pub struct " & rn & " {")
for f in t.fields:
L.add(" pub " & camelToSnakeCase(f.name) & ": " & rustIdiomatic(f.typeName) & ",")
L.add("}")
L.add("")
# holder: owns the C struct + all heap backing (CStrings, arrays, nested
# holders). Returned by value; the backing lives on the heap, so the C
# struct's raw pointers stay valid across the move and for the call.
L.add("pub struct " & hn & " {")
L.add(" pub c: ffi::" & rn & ",")
L.add(" _s: Vec<CString>,")
for f in t.fields:
let s = camelToSnakeCase(f.name)
let ft = f.typeName.strip()
if isSeqT(ft):
L.add(" _a_" & s & ": Vec<" & rustCElem(seqElem(ft), types, qualify = true) & ">,")
if isStructT(seqElem(ft), types):
L.add(" _h_" & s & ": Vec<" & capitalizeFirstLetter(seqElem(ft)) & "C>,")
elif isOptT(ft) and isStructT(optElem(ft), types):
L.add(" _h_" & s & ": Vec<" & capitalizeFirstLetter(optElem(ft)) & "C>,")
elif isStructT(ft, types):
L.add(" _h_" & s & ": Vec<" & capitalizeFirstLetter(ft) & "C>,")
L.add("}")
L.add("")
L.add("impl " & rn & " {")
L.add(" pub fn to_c(&self) -> " & hn & " {")
L.add(" #[allow(unused_mut)]")
L.add(" let mut h = " & hn & " { c: unsafe { std::mem::zeroed() }, _s: Vec::new(),")
for f in t.fields:
let s = camelToSnakeCase(f.name)
let ft = f.typeName.strip()
if isSeqT(ft):
L.add(" _a_" & s & ": Vec::new(),")
if isStructT(seqElem(ft), types):
L.add(" _h_" & s & ": Vec::new(),")
elif (isOptT(ft) and isStructT(optElem(ft), types)) or isStructT(ft, types):
L.add(" _h_" & s & ": Vec::new(),")
L.add(" };")
for f in t.fields:
let s = camelToSnakeCase(f.name)
let ft = f.typeName.strip()
if isSeqT(ft):
let e = seqElem(ft)
L.add(" for x in &self." & s & " {")
if isStringT(e):
L.add(" let cs = CString::new(x.as_str()).unwrap_or_default();")
L.add(" h._a_" & s & ".push(cs.as_ptr());")
L.add(" h._s.push(cs);")
elif isStructT(e, types):
L.add(" let eh = x.to_c();")
L.add(" h._a_" & s & ".push(eh.c);")
L.add(" h._h_" & s & ".push(eh);")
elif e == "bool":
L.add(" h._a_" & s & ".push(if *x { 1 } else { 0 });")
else:
L.add(" h._a_" & s & ".push(*x);")
L.add(" }")
L.add(" h.c." & s & " = if h._a_" & s & ".is_empty() { std::ptr::null() } else { h._a_" & s & ".as_ptr() };")
L.add(" h.c." & s & "_len = h._a_" & s & ".len();")
elif isOptT(ft):
let e = optElem(ft)
if isStringT(e):
L.add(" if let Some(x) = &self." & s & " {")
L.add(" let cs = CString::new(x.as_str()).unwrap_or_default();")
L.add(" h.c." & s & " = cs.as_ptr(); h.c." & s & "_present = 1; h._s.push(cs);")
L.add(" }")
elif isStructT(e, types):
L.add(" if let Some(x) = &self." & s & " {")
L.add(" let eh = x.to_c(); h.c." & s & " = eh.c; h.c." & s & "_present = 1; h._h_" & s & ".push(eh);")
L.add(" }")
elif e == "bool":
L.add(" if let Some(v) = self." & s & " { h.c." & s & "_present = 1; h.c." & s & " = if v { 1 } else { 0 }; }")
else:
L.add(" if let Some(v) = self." & s & " { h.c." & s & "_present = 1; h.c." & s & " = v; }")
elif isStringT(ft):
L.add(" { let cs = CString::new(self." & s & ".as_str()).unwrap_or_default(); h.c." & s & " = cs.as_ptr(); h._s.push(cs); }")
elif ft == "bool":
L.add(" h.c." & s & " = if self." & s & " { 1 } else { 0 };")
elif isStructT(ft, types):
L.add(" h._h_" & s & ".push(self." & s & ".to_c()); h.c." & s & " = h._h_" & s & "[0].c;")
else:
L.add(" h.c." & s & " = self." & s & ";")
L.add(" h")
L.add(" }")
# from_c
L.add(" pub fn from_c(c: &ffi::" & rn & ") -> Self {")
L.add(" " & rn & " {")
for f in t.fields:
let s = camelToSnakeCase(f.name)
let ft = f.typeName.strip()
if isSeqT(ft):
let e = seqElem(ft)
var conv = "|&v| v"
if isStringT(e): conv = "|&p| cstr(p)"
elif isStructT(e, types): conv = "|e| " & capitalizeFirstLetter(e) & "::from_c(e)"
elif e == "bool": conv = "|&v| v != 0"
L.add(" " & s & ": if c." & s & ".is_null() { Vec::new() } else { unsafe { std::slice::from_raw_parts(c." & s & ", c." & s & "_len) }.iter().map(" & conv & ").collect() },")
elif isOptT(ft):
let e = optElem(ft)
var val = "c." & s
if isStringT(e): val = "cstr(c." & s & ")"
elif isStructT(e, types): val = capitalizeFirstLetter(e) & "::from_c(&c." & s & ")"
elif e == "bool": val = "c." & s & " != 0"
L.add(" " & s & ": if c." & s & "_present != 0 { Some(" & val & ") } else { None },")
elif isStringT(ft):
L.add(" " & s & ": cstr(c." & s & "),")
elif ft == "bool":
L.add(" " & s & ": c." & s & " != 0,")
elif isStructT(ft, types):
L.add(" " & s & ": " & capitalizeFirstLetter(ft) & "::from_c(&c." & s & "),")
else:
L.add(" " & s & ": c." & s & ",")
L.add(" }")
L.add(" }")
L.add("}")
L.add("")
return L.join("\n")
# ── api.rs (Node + blocking calls) ───────────────────────────────────────────
proc rustMethod(procName, libName: string): string =
let prefix = libName & "_"
let bare =
if procName.startsWith(prefix): procName[prefix.len .. ^1] else: procName
camelToSnakeCase(bare)
proc emitApiRs(
procs: seq[FFIProcMeta], types: seq[FFITypeMeta], libName: string
): string =
let nodeT = snakeToPascalCase(libName) & "Node" # idiomatic Rust: MyTimerNode
var L: seq[string] = @[]
L.add("// Generated by nim-ffi native Rust codegen. Do not edit by hand.")
L.add("#![allow(dead_code)]")
L.add("use std::os::raw::{c_char, c_int, c_void};")
L.add("use std::sync::mpsc::{sync_channel, SyncSender};")
L.add("use super::ffi;")
L.add("use super::types::*;")
L.add("")
L.add("const RET_OK: c_int = 0;")
L.add("")
L.add("unsafe fn err_text(msg: *const c_char, len: usize) -> String {")
L.add(" if msg.is_null() || len == 0 { return String::from(\"error\"); }")
L.add(" String::from_utf8_lossy(std::slice::from_raw_parts(msg as *const u8, len)).into_owned()")
L.add("}")
L.add("")
L.add("unsafe extern \"C\" fn ack_cb(ret: c_int, msg: *const c_char, len: usize, ud: *mut c_void) {")
L.add(" let tx = Box::from_raw(ud as *mut SyncSender<Result<(), String>>);")
L.add(" let _ = tx.send(if ret == RET_OK { Ok(()) } else { Err(err_text(msg, len)) });")
L.add("}")
L.add("unsafe extern \"C\" fn str_cb(ret: c_int, msg: *const c_char, len: usize, ud: *mut c_void) {")
L.add(" let tx = Box::from_raw(ud as *mut SyncSender<Result<String, String>>);")
L.add(" let r = if ret == RET_OK {")
L.add(" let b = if msg.is_null() || len == 0 { Vec::new() } else { std::slice::from_raw_parts(msg as *const u8, len).to_vec() };")
L.add(" Ok(String::from_utf8_lossy(&b).into_owned())")
L.add(" } else { Err(err_text(msg, len)) };")
L.add(" let _ = tx.send(r);")
L.add("}")
# Per-struct-return callbacks.
for p in procs:
if p.kind != FFIKind.FFI or not procSimple(p, types): continue
if not isStructT(p.returnTypeName, types): continue
let rt = capitalizeFirstLetter(p.returnTypeName)
L.add("unsafe extern \"C\" fn cb_" & p.procName & "(ret: c_int, msg: *const c_char, len: usize, ud: *mut c_void) {")
L.add(" let tx = Box::from_raw(ud as *mut SyncSender<Result<" & rt & ", String>>);")
L.add(" let r = if ret == RET_OK && !msg.is_null() {")
L.add(" Ok(" & rt & "::from_c(&*(msg as *const ffi::" & rt & ")))")
L.add(" } else { Err(err_text(msg, len)) };")
L.add(" let _ = tx.send(r);")
L.add("}")
L.add("")
L.add("pub struct " & nodeT & " { ctx: *mut c_void }")
L.add("unsafe impl Send for " & nodeT & " {}")
L.add("unsafe impl Sync for " & nodeT & " {}")
L.add("")
L.add("impl " & nodeT & " {")
# ctor
for p in procs:
if p.kind != FFIKind.CTOR: continue
var params: seq[string] = @[]
for ep in p.extraParams:
params.add(camelToSnakeCase(ep.name) & ": " & capitalizeFirstLetter(ep.typeName))
L.add(" pub fn new(" & params.join(", ") & ") -> Result<Self, String> {")
var args: seq[string] = @[]
for ep in p.extraParams:
# The holder must outlive the call (it owns the CStrings/arrays the C
# struct points into); the native ctor deep-copies before returning.
L.add(" let c_" & camelToSnakeCase(ep.name) & " = " & camelToSnakeCase(ep.name) & ".to_c();")
args.add("c_" & camelToSnakeCase(ep.name) & ".c")
L.add(" let (tx, rx) = sync_channel::<Result<(), String>>(1);")
L.add(" let raw = Box::into_raw(Box::new(tx)) as *mut c_void;")
let cargs = (args & @["ack_cb", "raw"]).join(", ")
L.add(" let ctx = unsafe { ffi::" & p.procName & "(" & cargs & ") };")
L.add(" let res = rx.recv().map_err(|_| String::from(\"callback channel closed\"))?;")
L.add(" res?;")
L.add(" if ctx.is_null() { return Err(String::from(\"" & p.procName & " returned null\")); }")
L.add(" Ok(" & nodeT & " { ctx })")
L.add(" }")
L.add("")
# methods
for p in procs:
if p.kind != FFIKind.FFI: continue
if not procSimple(p, types):
L.add(" // SKIPPED " & p.procName & ": seq/Option params not yet supported by native Rust codegen")
continue
let structRet = isStructT(p.returnTypeName, types)
let retT = if structRet: capitalizeFirstLetter(p.returnTypeName) else: "String"
let cbName = if structRet: "cb_" & p.procName else: "str_cb"
var params: seq[string] = @[]
var conv: seq[string] = @[]
var callArgs: seq[string] = @["self.ctx", cbName, "raw"]
for ep in p.extraParams:
params.add(camelToSnakeCase(ep.name) & ": " & capitalizeFirstLetter(ep.typeName))
conv.add(" let c_" & camelToSnakeCase(ep.name) & " = " & camelToSnakeCase(ep.name) & ".to_c();")
callArgs.add("c_" & camelToSnakeCase(ep.name) & ".c")
L.add(" pub fn " & rustMethod(p.procName, libName) & "(&self" & (if params.len > 0: ", " & params.join(", ") else: "") & ") -> Result<" & retT & ", String> {")
for c in conv: L.add(c)
L.add(" let (tx, rx) = sync_channel::<Result<" & retT & ", String>>(1);")
L.add(" let raw = Box::into_raw(Box::new(tx)) as *mut c_void;")
L.add(" let rc = unsafe { ffi::" & p.procName & "(" & callArgs.join(", ") & ") };")
L.add(" if rc != RET_OK {")
L.add(" drop(unsafe { Box::from_raw(raw as *mut SyncSender<Result<" & retT & ", String>>) });")
L.add(" return Err(String::from(\"" & p.procName & " dispatch failed\"));")
L.add(" }")
L.add(" rx.recv().map_err(|_| String::from(\"callback channel closed\"))?")
L.add(" }")
L.add("")
# dtor
for p in procs:
if p.kind == FFIKind.DTOR:
L.add("}")
L.add("")
L.add("impl Drop for " & nodeT & " {")
L.add(" fn drop(&mut self) { unsafe { ffi::" & p.procName & "(self.ctx); } }")
L.add("}")
return L.join("\n")
proc generateRustNativeCrate*(
procs: seq[FFIProcMeta],
types: seq[FFITypeMeta],
libName: string,
outputDir: string,
nimSrcRelPath: string,
events: seq[FFIEventMeta] = @[],
) =
createDir(outputDir / "src")
writeFile(outputDir / "Cargo.toml",
"[package]\nname = \"" & libName & "_native\"\nversion = \"0.1.0\"\nedition = \"2021\"\n")
writeFile(outputDir / "src" / "lib.rs",
"mod ffi;\nmod types;\nmod api;\npub use types::*;\npub use api::*;\n")
writeFile(outputDir / "src" / "ffi.rs", emitFfiRs(procs, types, libName))
writeFile(outputDir / "src" / "types.rs", emitTypesRs(types))
writeFile(outputDir / "src" / "api.rs", emitApiRs(procs, types, libName))