From 8ecaa338bc2485baddb7bf8ddc054e1682c4f0d6 Mon Sep 17 00:00:00 2001
From: q66 <daniel@octaforge.org>
Date: Sat, 4 Mar 2017 18:25:09 +0100
Subject: [PATCH] add initial coroutine module (slow ucontext_t, POSIX only,
 WiP)

---
 ostd/coroutine.hh | 201 ++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 201 insertions(+)
 create mode 100644 ostd/coroutine.hh

diff --git a/ostd/coroutine.hh b/ostd/coroutine.hh
new file mode 100644
index 0000000..2b3c9f7
--- /dev/null
+++ b/ostd/coroutine.hh
@@ -0,0 +1,201 @@
+/* Coroutines for OctaSTD.
+ *
+ * This file is part of OctaSTD. See COPYING.md for further information.
+ */
+
+#ifndef OSTD_COROUTINE_HH
+#define OSTD_COROUTINE_HH
+
+/* currently there is only POSIX support using obsolete ucontext stuff...
+ * we will want to implement Windows support using its fibers and also
+ * lightweight custom context switching with handwritten asm where we
+ * want this to run, but ucontext will stay as a fallback
+ */
+
+#include <signal.h>
+#include <ucontext.h>
+
+#include <memory>
+#include <exception>
+#include <stdexcept>
+#include <utility>
+#include <tuple>
+
+#include "ostd/types.hh"
+
+namespace ostd {
+
+constexpr size_t COROUTINE_DEFAULT_STACK_SIZE = SIGSTKSZ;
+
+struct coroutine_error: std::runtime_error {
+    using std::runtime_error::runtime_error;
+};
+
+struct coroutine_context {
+    coroutine_context(size_t ss, void (*callp)(void *), void *data):
+        p_stack(new byte[ss])
+    {
+        getcontext(&p_coro);
+        p_coro.uc_link = &p_orig;
+        p_coro.uc_stack.ss_sp = p_stack.get();
+        p_coro.uc_stack.ss_size = ss;
+        using mcfp = void (*)();
+        using cpfp = void (*)(void *);
+        if constexpr(sizeof(void *) > sizeof(int)) {
+            union intu {
+                struct { int p1, p2; };
+                void *p;
+                cpfp fp;
+            };
+            intu ud, uf;
+            ud.p = data;
+            uf.fp = callp;
+            using amcfp = void (*)(int, int, int, int);
+            amcfp mcf = [](int f1, int f2, int d1, int d2) -> void {
+                intu ud2, uf2;
+                uf2.p1 = f1, uf2.p2 = f2;
+                ud2.p1 = d1, ud2.p2 = d2;
+                (uf2.fp)(ud2.p);
+            };
+            makecontext(
+                &p_coro, reinterpret_cast<mcfp>(mcf), 4,
+                uf.p1, uf.p2, ud.p1, ud.p2
+            );
+        } else {
+            using amcfp = void (*)(int, int);
+            amcfp mcf = [](int f1, int d1) {
+                reinterpret_cast<cpfp>(f1)(reinterpret_cast<void *>(d1));
+            };
+            makecontext(&p_coro, reinterpret_cast<mcfp>(mcf), 2, callp, data);
+        }
+    }
+
+    void call() {
+        if (p_finished) {
+            throw coroutine_error{"dead coroutine"};
+        }
+        swapcontext(&p_orig, &p_coro);
+        if (p_except) {
+            std::rethrow_exception(std::move(p_except));
+        }
+    }
+
+    void yield_swap() {
+        swapcontext(&p_coro, &p_orig);
+    }
+
+    void set_eh() {
+        p_except = std::current_exception();
+    }
+
+    void set_done() {
+        p_finished = true;
+    }
+
+    bool is_done() const {
+        return p_finished;
+    }
+
+private:
+    /* TODO: new'ing the stack is sub-optimal */
+    std::unique_ptr<byte[]> p_stack;
+    ucontext_t p_coro;
+    ucontext_t p_orig;
+    std::exception_ptr p_except;
+    bool p_finished = false;
+};
+
+template<typename T>
+struct coroutine;
+
+namespace detail {
+    /* we need this because yield is specialized based on result */
+    template<typename R, typename ...A>
+    struct coro_base {
+        coro_base(void (*callp)(void *), size_t ss):
+            p_ctx(ss, callp, this)
+        {}
+
+        std::tuple<A &&...> yield(R &&ret) {
+            p_result = std::forward<R>(ret);
+            p_ctx.yield_swap();
+            return std::move(p_args);
+        }
+
+    protected:
+        R ctx_call(A ...args) {
+            p_args = std::forward_as_tuple(std::forward<A>(args)...);
+            p_ctx.call();
+            return std::forward<R>(p_result);
+        }
+
+        std::tuple<A...> p_args;
+        R p_result;
+        coroutine_context p_ctx;
+    };
+
+    template<typename ...A>
+    struct coro_base<void, A...> {
+        coro_base(void (*callp)(void *), size_t ss):
+            p_ctx(ss, callp, this)
+        {}
+
+        std::tuple<A &&...> yield() {
+            p_ctx.yield_swap();
+            return std::move(p_args);
+        }
+
+    protected:
+        void ctx_call(A ...args) {
+            p_args = std::forward_as_tuple(std::forward<A>(args)...);
+            p_ctx.call();
+        }
+
+        std::tuple<A...> p_args;
+        coroutine_context p_ctx;
+    };
+} /* namespace detail */
+
+template<typename R, typename ...A>
+struct coroutine<R(A...)>: detail::coro_base<R, A...> {
+    coroutine(
+        std::function<R(coroutine<R(A...)> &, A...)> func,
+        size_t ss = COROUTINE_DEFAULT_STACK_SIZE
+    ):
+        detail::coro_base<R, A...>(&ctx_func, ss), p_func(std::move(func))
+    {}
+
+    operator bool() const {
+        return this->p_ctx.is_done();
+    }
+
+    R operator()(A ...args) {
+        return this->ctx_call(std::forward<A>(args)...);
+    }
+private:
+    template<size_t ...I>
+    R call(std::index_sequence<I...>) {
+        return p_func(*this, std::forward<A>(std::get<I>(this->p_args))...);
+    }
+
+    static void ctx_func(void *data) {
+        coroutine &self = *(static_cast<coroutine *>(data));
+        try {
+            using indices = std::index_sequence_for<A...>;
+            if constexpr(std::is_same_v<R, void>) {
+                self.call(indices{});
+            } else {
+                self.p_result = self.call(indices{});
+            }
+        } catch (...) {
+            self.p_ctx.set_eh();
+        }
+        self.p_ctx.set_done();
+    }
+
+    std::function<R(coroutine<R(A...)> &, A...)> p_func;
+};
+
+} /* namespace ostd */
+
+#endif