Builds TinyCC Cli and Library For C Scripting in R
Rtinycc is an R interface to TinyCC, providing both CLI access and a libtcc-backed in-memory compiler. It includes an FFI inspired by Bun’s FFI for binding C symbols with predictable type conversions and pointer utilities. The package works on unix-alikes and Windows and focuses on embedding TinyCC and enabling JIT-compiled bindings directly from R. Combined with treesitter.c, which provides C header parsers, it can be used to rapidly generate declarative bindings.
When you call tcc_compile(), Rtinycc generates C wrapper functions
whose signature follows the .Call convention (SEXP in, SEXP out).
These wrappers convert R types to C, call the target function, and
convert the result back. TCC compiles them in-memory – no shared library
is written to disk and no R_init_* registration is needed.
After tcc_relocate(), wrapper pointers are retrieved via
tcc_get_symbol(), which internally calls RC_libtcc_get_symbol().
That function converts TCC’s raw void* into a DL_FUNC wrapped with
R_MakeExternalPtrFn (tagged "native symbol"). On the R side,
make_callable() creates a closure that passes this external
pointer to .Call (aliased as .RtinyccCall to keep R CMD check
happy).
The design follows CFFI’s API-mode
pattern: instead of computing struct layouts and calling conventions in
R (ABI-mode, like Python’s ctypes), the generated C code lets TCC handle
sizeof, offsetof, and argument passing. Rtinycc never replicates
platform-specific layout rules. The wrappers can also link against
external shared libraries whose symbols TCC resolves at relocation time.
For background on how this compares to a libffi approach, see the
RSimpleFFI
README.
On macOS the configure script strips -flat_namespace from TCC’s build
to avoid BUS ERROR
issues. Without
it, TCC cannot resolve host symbols (e.g. RC_free_finalizer) through
the dynamic linker. Rtinycc works around this with
RC_libtcc_add_host_symbols(), which registers package-internal C
functions via tcc_add_symbol() before relocation. Any new C function
referenced by generated TCC code must be added there.
On Windows, the configure.win script generates a UCRT-backed
msvcrt.def so TinyCC resolves CRT symbols against ucrtbase.dll (R
4.2+ uses UCRT).
Ownership semantics are explicit. Pointers from tcc_malloc() are
tagged rtinycc_owned and can be released with tcc_free() (or by
their R finalizer). Generated struct constructors use a struct-specific
tag (struct_<name>) with an RC_free_finalizer; free them with
struct_<name>_free(), not tcc_free(). Pointers from tcc_data_ptr()
are tagged rtinycc_borrowed and are never freed by Rtinycc. Array
returns are copied into a fresh R vector; set free = TRUE only when
the C function returns a malloc-owned buffer.
install.packages(
'Rtinycc',
repos = c('https://sounkou-bioinfo.r-universe.dev',
'https://cloud.r-project.org')
)The CLI interface compiles C source files to standalone executables using the bundled TinyCC toolchain.
library(Rtinycc)
src <- system.file("c_examples", "forty_two.c", package = "Rtinycc")
exe <- tempfile()
tcc_run_cli(c(
"-B", tcc_prefix(),
paste0("-I", tcc_include_paths()),
paste0("-L", tcc_lib_paths()),
src, "-o", exe
))
#> [1] 0
Sys.chmod(exe, mode = "0755")
system2(exe, stdout = TRUE)
#> [1] "42"For in-memory workflows, prefer libtcc instead.
We can compile and call C functions entirely in memory. This is the simplest path for quick JIT compilation.
state <- tcc_state(output = "memory")
tcc_compile_string(state, "int forty_two(){ return 42; }")
#> [1] 0
tcc_relocate(state)
#> [1] 0
tcc_call_symbol(state, "forty_two", return = "int")
#> [1] 42The lower-level API gives full control over include paths, libraries,
and the R C API. Using #define _Complex as a workaround for TCC’s lack
of complex type
support,
we can link against R’s headers and call into libR.
state <- tcc_state(output = "memory")
tcc_add_include_path(state, R.home("include"))
#> [1] 0
tcc_add_library_path(state, R.home("lib"))
#> [1] 0
code <- '
#define _Complex
#include <R.h>
#include <Rinternals.h>
double call_r_sqrt(void) {
SEXP fn = PROTECT(Rf_findFun(Rf_install("sqrt"), R_BaseEnv));
SEXP val = PROTECT(Rf_ScalarReal(16.0));
SEXP call = PROTECT(Rf_lang2(fn, val));
SEXP out = PROTECT(Rf_eval(call, R_GlobalEnv));
double res = REAL(out)[0];
UNPROTECT(4);
return res;
}
'
tcc_compile_string(state, code)
#> [1] 0
tcc_relocate(state)
#> [1] 0
tcc_call_symbol(state, "call_r_sqrt", return = "double")
#> [1] 4Rtinycc ships a set of typed memory access functions similar to what the
ctypesio package offers,
but designed around our FFI pointer model. Every scalar C type has a
corresponding tcc_read_* / tcc_write_* pair that operates at a byte
offset into any external pointer, so you can walk structs, arrays, and
output parameters without writing C helpers.
ptr <- tcc_cstring("hello")
tcc_read_cstring(ptr)
#> [1] "hello"
tcc_read_bytes(ptr, 5)
#> [1] 68 65 6c 6c 6f
tcc_ptr_addr(ptr, hex = TRUE)
#> [1] "0x5ed89220dc50"
tcc_ptr_is_null(ptr)
#> [1] FALSE
tcc_free(ptr)
#> NULLTyped reads and writes cover the full scalar range (i8/u8,
i16/u16, i32/u32, i64/u64, f32/f64) plus pointer
dereferencing via tcc_read_ptr / tcc_write_ptr. All operations use a
byte offset and memcpy internally for alignment safety.
buf <- tcc_malloc(32)
tcc_write_i32(buf, 0L, 42L)
tcc_write_f64(buf, 8L, pi)
tcc_read_i32(buf, offset = 0L)
#> [1] 42
tcc_read_f64(buf, offset = 8L)
#> [1] 3.141593
tcc_free(buf)
#> NULLPointer-to-pointer workflows are supported for C APIs that return values through output parameters.
ptr_ref <- tcc_malloc(.Machine$sizeof.pointer %||% 8L)
target <- tcc_malloc(8)
tcc_ptr_set(ptr_ref, target)
#> <pointer: 0x5ed8952bacf0>
tcc_data_ptr(ptr_ref)
#> <pointer: 0x5ed897497ec0>
tcc_ptr_set(ptr_ref, tcc_null_ptr())
#> <pointer: 0x5ed8952bacf0>
tcc_free(target)
#> NULL
tcc_free(ptr_ref)
#> NULLA declarative interface inspired by Bun’s FFI sits on top of the lower-level API. We define types explicitly and Rtinycc generates the binding code, compiling it in memory with TCC.
The FFI exposes a small set of type mappings between R and C. Conversions are explicit and predictable so callers know when data is shared versus copied.
Scalar types map one-to-one: i8, i16, i32, i64 (integers); u8,
u16, u32, u64 (unsigned); f32, f64 (floats); bool (logical);
cstring (NUL-terminated string).
Array arguments pass R vectors to C with zero copy: raw maps to
uint8_t*, integer_array to int32_t*, numeric_array to double*.
Pointer types include ptr (opaque external pointer), sexp (pass a
SEXP directly), and callback signatures like
callback:double(double).
Variadic functions are supported in two forms: typed prefix tails
(varargs) and bounded dynamic tails (varargs_types +
varargs_min/varargs_max). Prefix mode is the cheaper runtime path
because dispatch is by tail arity only; bounded dynamic mode adds
per-call scalar type inference to select a compatible wrapper. For hot
loops, prefer fixed arity first, then prefix variadics with a tight
maximum tail size.
Array returns use
returns = list(type = "integer_array", length_arg = 2, free = TRUE) to
copy the result into a new R vector. The length_arg is the 1-based
index of the C argument that carries the array length. Set free = TRUE
when the C function returns a malloc-owned buffer.
ffi <- tcc_ffi() |>
tcc_source("
int add(int a, int b) { return a + b; }
") |>
tcc_bind(add = list(args = list("i32", "i32"), returns = "i32")) |>
tcc_compile()
ffi$add(5L, 3L)
#> [1] 8Rtinycc supports two ways to bind variadic tails. The legacy approach
uses varargs as a typed prefix tail, while the bounded dynamic
approach uses varargs_types together with varargs_min and
varargs_max. In the bounded mode, wrappers are generated across the
allowed arity and type combinations, and runtime dispatch selects the
matching wrapper from the scalar tail values provided at call time.
ffi_var <- tcc_ffi() |>
tcc_header("#include <R_ext/Print.h>") |>
tcc_source('
#include <stdarg.h>
int sum_fmt(int n, ...) {
va_list ap;
va_start(ap, n);
int s = 0;
for (int i = 0; i < n; i++) s += va_arg(ap, int);
va_end(ap);
Rprintf("sum_fmt(%d) = %d\\n", n, s);
return s;
}
') |>
tcc_bind(
Rprintf = list(
args = list("cstring"),
variadic = TRUE,
varargs_types = list("i32"),
varargs_min = 0L,
varargs_max = 4L,
returns = "void"
),
sum_fmt = list(
args = list("i32"),
variadic = TRUE,
varargs_types = list("i32"),
varargs_min = 0L,
varargs_max = 4L,
returns = "i32"
)
) |>
tcc_compile()
ffi_var$Rprintf("Rprintf via bind: %d + %d = %d\n", 2L, 3L, 5L)
#> Rprintf via bind: 2 + 3 = 5
#> NULL
ffi_var$sum_fmt(0L)
#> sum_fmt(0) = 0
#> [1] 0
ffi_var$sum_fmt(2L, 10L, 20L)
#> sum_fmt(2) = 30
#> [1] 30
ffi_var$sum_fmt(4L, 1L, 2L, 3L, 4L)
#> sum_fmt(4) = 10
#> [1] 10We can bind directly to symbols in shared libraries. Here we link
against libm.
math <- tcc_ffi() |>
tcc_library("m") |>
tcc_bind(
sqrt = list(args = list("f64"), returns = "f64"),
sin = list(args = list("f64"), returns = "f64"),
floor = list(args = list("f64"), returns = "f64")
) |>
tcc_compile()
math$sqrt(16.0)
#> [1] 4
math$sin(pi / 2)
#> [1] 1
math$floor(3.7)
#> [1] 3Use tcc_options() to pass raw TinyCC options in the high-level FFI
pipeline. For low-level states, use tcc_set_options() directly.
ffi_opt_off <- tcc_ffi() |>
tcc_options("-O0") |>
tcc_source('
int opt_macro() {
#ifdef __OPTIMIZE__
return 1;
#else
return 0;
#endif
}
') |>
tcc_bind(opt_macro = list(args = list(), returns = "i32")) |>
tcc_compile()
ffi_opt_on <- tcc_ffi() |>
tcc_options(c("-Wall", "-O2")) |>
tcc_source('
int opt_macro() {
#ifdef __OPTIMIZE__
return 1;
#else
return 0;
#endif
}
') |>
tcc_bind(opt_macro = list(args = list(), returns = "i32")) |>
tcc_compile()
ffi_opt_off$opt_macro()
#> [1] 0
ffi_opt_on$opt_macro()
#> [1] 1R vectors are passed to C with zero copy. Mutations in C are visible in R.
ffi <- tcc_ffi() |>
tcc_source("
#include <stdlib.h>
#include <string.h>
int64_t sum_array(int32_t* arr, int32_t n) {
int64_t s = 0;
for (int i = 0; i < n; i++) s += arr[i];
return s;
}
void bump_first(int32_t* arr) { arr[0] += 10; }
int32_t* dup_array(int32_t* arr, int32_t n) {
int32_t* out = malloc(sizeof(int32_t) * n);
memcpy(out, arr, sizeof(int32_t) * n);
return out;
}
") |>
tcc_bind(
sum_array = list(args = list("integer_array", "i32"), returns = "i64"),
bump_first = list(args = list("integer_array"), returns = "void"),
dup_array = list(
args = list("integer_array", "i32"),
returns = list(type = "integer_array", length_arg = 2, free = TRUE)
)
) |>
tcc_compile()
x <- as.integer(1:100) # to avoid ALTREP
.Internal(inspect(x))
#> @5ed8955ac528 13 INTSXP g0c0 [REF(65535)] 1 : 100 (compact)
ffi$sum_array(x, length(x))
#> [1] 5050
# Zero-copy: C mutation reflects in R
ffi$bump_first(x)
#> NULL
x[1]
#> [1] 11
# Array return: copied into a new R vector, C buffer freed
y <- ffi$dup_array(x, length(x))
y[1]
#> [1] 11
.Internal(inspect(x))
#> @5ed8955ac528 13 INTSXP g0c0 [REF(65535)] 11 : 110 (expanded)Complex C types are supported declaratively. Use tcc_struct() to
generate allocation and accessor helpers. Free instances when done.
ffi <- tcc_ffi() |>
tcc_source('
#include <math.h>
struct point { double x; double y; };
double distance(struct point* a, struct point* b) {
double dx = a->x - b->x, dy = a->y - b->y;
return sqrt(dx * dx + dy * dy);
}
') |>
tcc_library("m") |>
tcc_struct("point", accessors = c(x = "f64", y = "f64")) |>
tcc_bind(distance = list(args = list("ptr", "ptr"), returns = "f64")) |>
tcc_compile()
p1 <- ffi$struct_point_new()
ffi$struct_point_set_x(p1, 0.0)
#> <pointer: 0x5ed893f08110>
ffi$struct_point_set_y(p1, 0.0)
#> <pointer: 0x5ed893f08110>
p2 <- ffi$struct_point_new()
ffi$struct_point_set_x(p2, 3.0)
#> <pointer: 0x5ed896fd09b0>
ffi$struct_point_set_y(p2, 4.0)
#> <pointer: 0x5ed896fd09b0>
ffi$distance(p1, p2)
#> [1] 5
ffi$struct_point_free(p1)
#> NULL
ffi$struct_point_free(p2)
#> NULLEnums are exposed as helper functions that return integer constants.
ffi <- tcc_ffi() |>
tcc_source("enum color { RED = 0, GREEN = 1, BLUE = 2 };") |>
tcc_enum("color", constants = c("RED", "GREEN", "BLUE")) |>
tcc_compile()
ffi$enum_color_RED()
#> [1] 0
ffi$enum_color_BLUE()
#> [1] 2Bitfields are handled by TCC. Accessors read and write them like normal fields.
ffi <- tcc_ffi() |>
tcc_source("
struct flags {
unsigned int active : 1;
unsigned int level : 4;
};
") |>
tcc_struct("flags", accessors = c(active = "u8", level = "u8")) |>
tcc_compile()
s <- ffi$struct_flags_new()
ffi$struct_flags_set_active(s, 1L)
#> <pointer: 0x5ed893aed890>
ffi$struct_flags_set_level(s, 9L)
#> <pointer: 0x5ed893aed890>
ffi$struct_flags_get_active(s)
#> [1] 1
ffi$struct_flags_get_level(s)
#> [1] 9
ffi$struct_flags_free(s)
#> NULLC globals can be exposed with explicit getter/setter helpers.
ffi <- tcc_ffi() |>
tcc_source("
int counter = 7;
double pi_approx = 3.14159;
") |>
tcc_global("counter", "i32") |>
tcc_global("pi_approx", "f64") |>
tcc_compile()
ffi$global_counter_get()
#> [1] 7
ffi$global_pi_approx_get()
#> [1] 3.14159
ffi$global_counter_set(42L)
#> [1] 42
ffi$global_counter_get()
#> [1] 42R functions can be registered as C function pointers via
tcc_callback() and passed to compiled code. Specify a
callback:<signature> argument in tcc_bind() so the trampoline is
generated automatically. Call tcc_callback_close() when you want
deterministic invalidation and earlier release of the preserved R
function.
cb <- tcc_callback(function(x) x * x, signature = "double (*)(double)")
code <- '
double apply_fn(double (*fn)(void* ctx, double), void* ctx, double x) {
return fn(ctx, x);
}
'
ffi <- tcc_ffi() |>
tcc_source(code) |>
tcc_bind(
apply_fn = list(
args = list("callback:double(double)", "ptr", "f64"),
returns = "f64"
)
) |>
tcc_compile()
ffi$apply_fn(cb, tcc_callback_ptr(cb), 7.0)
#> [1] 49
tcc_callback_close(cb)If a callback throws an R error, the trampoline catches it, emits a
warning, and returns a type-appropriate sentinel instead of unwinding
through C. In practice this means NA-like numeric or integer values,
NA logical, NA string, or a null external pointer depending on the
declared return type.
cb_err <- tcc_callback(
function(x) stop("boom"),
signature = "double (*)(double)"
)
ffi_err <- tcc_ffi() |>
tcc_source('
double call_cb_err(double (*cb)(void* ctx, double), void* ctx, double x) {
return cb(ctx, x);
}
') |>
tcc_bind(
call_cb_err = list(
args = list("callback:double(double)", "ptr", "f64"),
returns = "f64"
)
) |>
tcc_compile()
warned <- FALSE
res <- withCallingHandlers(
ffi_err$call_cb_err(cb_err, tcc_callback_ptr(cb_err), 1.0),
warning = function(w) {
warned <<- TRUE
invokeRestart("muffleWarning")
}
)
list(warned = warned, result = res)
#> $warned
#> [1] TRUE
#>
#> $result
#> [1] NA
tcc_callback_close(cb_err)For thread-safe scheduling from worker threads, use
callback_async:<signature> in tcc_bind(). The async callback queue
is initialized automatically at package load.
When a bound function has any callback_async: argument, the generated
wrapper automatically runs your C function on a new thread while
draining callbacks on the main R thread. Your C code doesn’t need to
know about draining at all — just call the callback as normal and the
wrapper handles the rest.
Void return (fire-and-forget): the callback is enqueued from any
thread and executed on the main R thread automatically — on Windows via
R’s message pump, on Linux/macOS via R’s event loop addInputHandler.
Non-void return (synchronous): the worker thread blocks until the
main R thread executes the callback and returns the real result.
Supported return types: integer variants (int, int32_t, i8, i16,
u8, u16), floating-point (double, float), bool/logical, and
pointer (void*, T*).
# Fire-and-forget: void callback accumulated from 100 worker threads
hits <- 0L
cb_async <- tcc_callback(
function(x) { hits <<- hits + x; NULL },
signature = "void (*)(int)"
)
code_async <- '
struct task { void (*cb)(void* ctx, int); void* ctx; int value; };
#ifdef _WIN32
#include <windows.h>
static DWORD WINAPI worker(LPVOID data) {
struct task* t = (struct task*) data;
t->cb(t->ctx, t->value);
return 0;
}
int spawn_async(void (*cb)(void* ctx, int), void* ctx, int value) {
if (!cb || !ctx) return -1;
struct task t;
t.cb = cb;
t.ctx = ctx;
t.value = value;
HANDLE th = CreateThread(NULL, 0, worker, &t, 0, NULL);
if (!th) return -2;
WaitForSingleObject(th, INFINITE);
CloseHandle(th);
return 0;
}
#else
#include <pthread.h>
static void* worker(void* data) {
struct task* t = (struct task*) data;
t->cb(t->ctx, t->value);
return NULL;
}
int spawn_async(void (*cb)(void* ctx, int), void* ctx, int value) {
if (!cb || !ctx) return -1;
const int n = 100;
struct task tasks[100];
pthread_t th[100];
for (int i = 0; i < n; i++) {
tasks[i].cb = cb;
tasks[i].ctx = ctx;
tasks[i].value = value;
if (pthread_create(&th[i], NULL, worker, &tasks[i]) != 0) {
for (int j = 0; j < i; j++) pthread_join(th[j], NULL);
return -2;
}
}
for (int i = 0; i < n; i++) pthread_join(th[i], NULL);
return 0;
}
#endif
'
ffi_async <- tcc_ffi() |>
tcc_source(code_async)
if (.Platform$OS.type != "windows") {
ffi_async <- tcc_library(ffi_async, "pthread")
}
ffi_async <- ffi_async |>
tcc_bind(
spawn_async = list(
args = list("callback_async:void(int)", "ptr", "i32"),
returns = "i32"
)
) |>
tcc_compile()
rc <- ffi_async$spawn_async(cb_async, tcc_callback_ptr(cb_async), 2L)
hits
#> [1] 200
tcc_callback_close(cb_async)Non-void return works the same way — the generated wrapper handles the drain loop transparently:
cb_triple <- tcc_callback(
function(x) x * 3L,
signature = "int (*)(int)"
)
# Pure C: the worker calls the sync callback and returns its result.
# No drain logic needed — the generated wrapper handles it.
code_sync <- '
#ifdef _WIN32
#include <windows.h>
struct itask { int (*cb)(void*,int); void* ctx; int in; int out; };
static DWORD WINAPI iworker(LPVOID p) {
struct itask* t = (struct itask*)p;
t->out = t->cb(t->ctx, t->in);
return 0;
}
int run_worker(int (*cb)(void*,int), void* ctx, int x) {
struct itask t;
t.cb = cb; t.ctx = ctx; t.in = x; t.out = -1;
HANDLE th = CreateThread(NULL, 0, iworker, &t, 0, NULL);
if (!th) return -1;
WaitForSingleObject(th, INFINITE);
CloseHandle(th);
return t.out;
}
#else
#include <pthread.h>
struct itask { int (*cb)(void*,int); void* ctx; int in; int out; };
static void* iworker(void* p) {
struct itask* t = (struct itask*)p;
t->out = t->cb(t->ctx, t->in);
return NULL;
}
int run_worker(int (*cb)(void*,int), void* ctx, int x) {
struct itask t;
t.cb = cb; t.ctx = ctx; t.in = x; t.out = -1;
pthread_t th;
if (pthread_create(&th, NULL, iworker, &t) != 0) return -1;
pthread_join(th, NULL);
return t.out;
}
#endif
'
ffi_sync <- tcc_ffi() |>
tcc_source(code_sync)
if (.Platform$OS.type != "windows") {
ffi_sync <- tcc_library(ffi_sync, "pthread")
}
ffi_sync <- ffi_sync |>
tcc_bind(
run_worker = list(args = list("callback_async:int(int)", "ptr", "i32"), returns = "i32")
) |>
tcc_compile()
ffi_sync$run_worker(cb_triple, tcc_callback_ptr(cb_triple), 7L) # 21
#> [1] 21
tcc_callback_close(cb_triple)This example ties together external library linking, callbacks, and
pointer dereferencing. We open an in-memory SQLite database, execute
queries, and collect rows through an R callback that reads char**
arrays using tcc_read_ptr and tcc_read_cstring.
ptr_size <- .Machine$sizeof.pointer
read_string_array <- function(ptr, n) {
vapply(seq_len(n), function(i) {
tcc_read_cstring(tcc_read_ptr(ptr, (i - 1L) * ptr_size))
}, "")
}
cb <- tcc_callback(
function(argc, argv, cols) {
values <- read_string_array(argv, argc)
names <- read_string_array(cols, argc)
cat(paste(names, values, sep = " = ", collapse = ", "), "\n")
0L
},
signature = "int (*)(int, char **, char **)"
)
sqlite <- tcc_ffi() |>
tcc_header("#include <sqlite3.h>") |>
tcc_library("sqlite3") |>
tcc_source('
void* open_db() {
sqlite3* db = NULL;
sqlite3_open(":memory:", &db);
return db;
}
int close_db(void* db) {
return sqlite3_close((sqlite3*)db);
}
') |>
tcc_bind(
open_db = list(args = list(), returns = "ptr"),
close_db = list(args = list("ptr"), returns = "i32"),
sqlite3_libversion = list(args = list(), returns = "cstring"),
sqlite3_exec = list(
args = list("ptr", "cstring", "callback:int(int, char **, char **)", "ptr", "ptr"),
returns = "i32"
)
) |>
tcc_compile()
sqlite$sqlite3_libversion()
#> [1] "3.45.1"
db <- sqlite$open_db()
sqlite$sqlite3_exec(db, "CREATE TABLE t (id INTEGER, name TEXT);", cb, tcc_callback_ptr(cb), tcc_null_ptr())
#> [1] 0
sqlite$sqlite3_exec(db, "INSERT INTO t VALUES (1, 'hello'), (2, 'world');", cb, tcc_callback_ptr(cb), tcc_null_ptr())
#> [1] 0
sqlite$sqlite3_exec(db, "SELECT * FROM t;", cb, tcc_callback_ptr(cb), tcc_null_ptr())
#> id = 1, name = hello
#> id = 2, name = world
#> [1] 0
sqlite$close_db(db)
#> [1] 0
tcc_callback_close(cb)For header-driven bindings, we use treesitter.c to parse function
signatures and generate binding specifications automatically. For
struct, enum, and global helpers, tcc_generate_bindings() handles the
code generation.
The default mapper is conservative for pointers: char* is treated as
ptr because C does not guarantee NUL-terminated strings. If you know a
parameter is a C string, provide a custom mapper that returns cstring
for that type.
header <- '
double sqrt(double x);
double sin(double x);
struct point { double x; double y; };
enum status { OK = 0, ERROR = 1 };
int global_counter;
'
tcc_treesitter_functions(header)
#> capture_name text start_line start_col params return_type
#> 1 decl_name sqrt 2 8 double double
#> 2 decl_name sin 3 8 double double
tcc_treesitter_structs(header)
#> capture_name text start_line
#> 1 struct_name point 4
tcc_treesitter_enums(header)
#> capture_name text start_line
#> 1 enum_name status 5
tcc_treesitter_globals(header)
#> capture_name text start_line
#> 1 global_name global_counter 6
# Bind parsed functions to libm
symbols <- tcc_treesitter_bindings(header)
math <- tcc_link("m", symbols = symbols)
math$sqrt(16.0)
#> [1] 4
# Generate struct/enum/global helpers
ffi <- tcc_ffi() |>
tcc_source(header) |>
tcc_generate_bindings(
header,
functions = FALSE, structs = TRUE,
enums = TRUE, globals = TRUE
) |>
tcc_compile()
ffi$struct_point_new()
#> <pointer: 0x5ed895123b00>
ffi$enum_status_OK()
#> [1] 0
ffi$global_global_counter_get()
#> [1] 0CSV parser using io_uring
on linux
if (Sys.info()[["sysname"]] == "Linux") {
c_file <- system.file("c_examples", "io_uring_csv.c", package = "Rtinycc")
n_rows <- 20000L
n_cols <- 8L
block_size <- 1024L * 1024L
set.seed(42)
tmp_csv <- tempfile("rtinycc_io_uring_readme_", fileext = ".csv")
on.exit(unlink(tmp_csv), add = TRUE)
mat <- matrix(runif(n_rows * n_cols), ncol = n_cols)
df <- as.data.frame(mat)
names(df) <- paste0("V", seq_len(n_cols))
utils::write.table(df, file = tmp_csv, sep = ",", row.names = FALSE, col.names = TRUE, quote = FALSE)
csv_size_mb <- as.double(file.info(tmp_csv)$size) / 1024^2
message(sprintf("CSV size: %.2f MB", csv_size_mb))
io_uring_src <- paste(readLines(c_file, warn = FALSE), collapse = "\n")
ffi <- tcc_ffi() |>
tcc_source(io_uring_src) |>
tcc_bind(
csv_table_read = list(
args = list("cstring", "i32", "i32"),
returns = "sexp"
),
csv_table_io_uring = list(
args = list("cstring", "i32", "i32"),
returns = "sexp"
)
) |>
tcc_compile()
baseline <- utils::read.table(tmp_csv, sep = ",", header = TRUE)
c_tbl <- ffi$csv_table_read(tmp_csv, block_size, n_cols)
uring_tbl <- ffi$csv_table_io_uring(tmp_csv, block_size, n_cols)
vroom_tbl <- vroom::vroom(
tmp_csv,
delim = ",",
altrep = FALSE,
col_types = vroom::cols(.default = "d"),
progress = FALSE,
show_col_types = FALSE
)
stopifnot(
identical(dim(c_tbl), dim(baseline)),
identical(dim(uring_tbl), dim(baseline)),
identical(dim(vroom_tbl), dim(baseline)),
isTRUE(all.equal(c_tbl, baseline, tolerance = 1e-8, check.attributes = FALSE)),
isTRUE(all.equal(uring_tbl, baseline, tolerance = 1e-8, check.attributes = FALSE)),
isTRUE(all.equal(vroom_tbl, baseline, tolerance = 1e-8, check.attributes = FALSE))
)
timings <- bench::mark(
read_table_df = {
x <- utils::read.table(tmp_csv, sep = ",", header = TRUE)
nrow(x)
},
vroom_df_altrep_false = {
x <- vroom::vroom(
tmp_csv,
delim = ",",
altrep = FALSE,
col_types = vroom::cols(.default = "d"),
progress = FALSE,
show_col_types = FALSE
)
nrow(x)
},
vroom_df_altrep_false_mat = {
x <- vroom::vroom(
tmp_csv,
delim = ",",
altrep = FALSE,
col_types = vroom::cols(.default = "d"),
progress = FALSE,
show_col_types = FALSE
)
x <- as.matrix(x)
nrow(x)
},
c_read_df = {
x <- ffi$csv_table_read(tmp_csv, block_size, n_cols)
nrow(x)
},
io_uring_df = {
x <- ffi$csv_table_io_uring(tmp_csv, block_size, n_cols)
nrow(x)
},
iterations = 2,
memory = TRUE
)
print(timings)
plot(timings, type = "boxplot") + bench::scale_x_bench_time(base = NULL)
}
#> CSV size: 2.75 MB
#> # A tibble: 5 × 13
#> expression min median `itr/sec` mem_alloc `gc/sec` n_itr n_gc total_time
#> <bch:expr> <bch:t> <bch:t> <dbl> <bch:byt> <dbl> <int> <dbl> <bch:tm>
#> 1 read_tabl… 48.33ms 49.71ms 20.1 6.33MB 0 2 0 99.4ms
#> 2 vroom_df_… 6.22ms 6.38ms 157. 1.22MB 0 2 0 12.8ms
#> 3 vroom_df_… 6.99ms 7.52ms 133. 2.44MB 0 2 0 15ms
#> 4 c_read_df 21.43ms 21.43ms 46.7 1.22MB 46.7 1 1 21.4ms
#> 5 io_uring_… 20.4ms 20.49ms 48.8 1.22MB 0 2 0 41ms
#> # ℹ 4 more variables: result <list>, memory <list>, time <list>, gc <list>
TCC does not support C99 _Complex types. Generated code works around
this with #define _Complex, which suppresses the keyword. Apply the
same workaround in your own tcc_source() code when headers pull in
complex types.
R represents i64 and u64 values as double, which loses precision
beyond
sprintf("2^53: %.0f", 2^53)
#> [1] "2^53: 9007199254740992"
sprintf("2^53 + 1: %.0f", 2^53 + 1)
#> [1] "2^53 + 1: 9007199254740992"
identical(2^53, 2^53 + 1)
#> [1] TRUEFor exact 64-bit arithmetic, keep values in C-allocated storage and manipulate them through pointers.
Named nested struct fields can now be declared explicitly with
struct:<name>. Getters return borrowed nested views and setters copy
bytes from another struct object of the declared nested type.
ffi <- tcc_ffi() |>
tcc_source('
struct inner { int a; };
struct outer { struct inner in; };
') |>
tcc_struct("inner", accessors = c(a = "i32")) |>
tcc_struct("outer", accessors = list(`in` = "struct:inner")) |>
tcc_compile()
outer <- ffi$struct_outer_new()
inner <- ffi$struct_inner_new()
inner <- ffi$struct_inner_set_a(inner, 42L)
outer <- ffi$struct_outer_set_in(outer, inner)
inner_view <- ffi$struct_outer_get_in(outer)
ffi$struct_inner_get_a(inner_view)
#> [1] 42
ffi$struct_inner_free(inner)
#> NULL
ffi$struct_outer_free(outer)
#> NULLFor treesitter-generated bindings, nested struct fields inside structs
still fall back to ptr-like accessors. If you want a borrowed nested
view plus a copy-in setter, declare the nested field explicitly with
struct:<name>.
Bitfields are separate from ordinary addressable fields. They use scalar
helper accessors, but field_addr() and container_of() reject
bitfield members.
ffi <- tcc_ffi() |>
tcc_source('struct flags { unsigned int flag : 1; };') |>
tcc_struct(
"flags",
accessors = list(flag = list(type = "u8", bitfield = TRUE, width = 1))
)
tcc_field_addr(ffi, "flags", "flag")
#> Error:
#> ! field_addr does not support bitfield members
tcc_container_of(ffi, "flags", "flag")
#> Error:
#> ! container_of does not support bitfield membersArray fields require the list(type = ..., size = N, array = TRUE)
syntax in tcc_struct(), which generates element-wise accessors.
ffi <- tcc_ffi() |>
tcc_source('struct buf { unsigned char data[16]; };') |>
tcc_struct("buf", accessors = list(
data = list(type = "u8", size = 16, array = TRUE)
)) |>
tcc_compile()
b <- ffi$struct_buf_new()
ffi$struct_buf_set_data_elt(b, 0L, 0xCAL)
#> <pointer: 0x5ed89da7e280>
ffi$struct_buf_set_data_elt(b, 1L, 0xFEL)
#> <pointer: 0x5ed89da7e280>
ffi$struct_buf_get_data_elt(b, 0L)
#> [1] 202
ffi$struct_buf_get_data_elt(b, 1L)
#> [1] 254
ffi$struct_buf_free(b)
#> NULLCompiled FFI objects are fork-safe: parallel::mclapply() and other
fork()-based parallelism work out of the box because TCC’s compiled
code lives in memory mappings that survive fork() via copy-on-write.
Serialization is also supported. Each tcc_compiled object stores its
FFI recipe internally, so after saveRDS() / readRDS() (or
serialize() / unserialize()), the first $ access detects the dead
TCC state pointer and recompiles transparently.
ffi <- tcc_ffi() |>
tcc_source("int square(int x) { return x * x; }") |>
tcc_bind(square = list(args = list("i32"), returns = "i32")) |>
tcc_compile()
ffi$square(7L)
#> [1] 49
tmp <- tempfile(fileext = ".rds")
saveRDS(ffi, tmp)
ffi2 <- readRDS(tmp)
unlink(tmp)
# Auto-recompiles on first access
ffi2$square(7L)
#> [Rtinycc] Recompiling FFI bindings after deserialization
#> [1] 49For explicit control, use tcc_recompile(). Note that raw tcc_state
objects and bare pointers from tcc_malloc() do not carry a recipe and
remain dead after deserialization.
Rtinycc is optimized for fast in-process compilation and convenient
FFI workflows, not for winning every microbenchmark against a
conventional precompiled .Call() shared library.
In practice, the usual pattern is:
Rtinycccompiles tiny modules very quickly- a regular
.Call()module can have lower minimal per-call overhead - array-oriented zero-copy inputs are a much better fit than many tiny scalar crossings
- return paths that copy native buffers back into fresh R vectors make that copy cost visible
If you want the benchmark details rather than the high-level summary,
see the Compilation and Call Overhead vignette.
GPL-3