libutil: begin porting serialization to generators

generators are a better basis for serializers than streaming into sinks
as we do currently for many reasons, such as being usable as sources if
one wishes to (without requiring an intermediate sink to serialize full
data sets into memory, or boost coroutines to turn sinks into sources),
composing more naturally (as one can just yield a sub-generator instead
of being forced to wrap entire substreams into clunky functions or even
more clunky custom types to implement operator<< on), allowing wrappers
to transform data with clear ownership semantics (removing the need for
explicit memory allocations and Source wrappers), and many other things

Change-Id: I361d89ff556354f6930d9204f55117565f2f7f20

5 files changed, 174 insertions, 90 deletions

diff --git a/src/libstore/daemon.cc b/src/libstore/daemon.cc
index 6d64644d1..cdc9c09c7 100644
--- a/src/libstore/daemon.cc
+++ b/src/libstore/daemon.cc
@@ -160,8 +160,7 @@ struct TunnelSink : Sink
     TunnelSink(Sink & to) : to(to) { }
     void operator () (std::string_view data)
     {
-        to << STDERR_WRITE;
-        writeString(data, to);
+        to << STDERR_WRITE << data;
     }
 };
 
diff --git a/src/libstore/remote-store.cc b/src/libstore/remote-store.cc
index b2f8a285d..55a71f502 100644
--- a/src/libstore/remote-store.cc
+++ b/src/libstore/remote-store.cc
@@ -897,7 +897,7 @@ std::exception_ptr RemoteStore::Connection::processStderr(Sink * sink, Source *
             if (!source) throw Error("no source");
             size_t len = readNum<size_t>(from);
             auto buf = std::make_unique<char[]>(len);
-            writeString({(const char *) buf.get(), source->read(buf.get(), len)}, to);
+            to << std::string_view((const char *) buf.get(), source->read(buf.get(), len));
             to.flush();
         }
 
diff --git a/src/libutil/serialise.cc b/src/libutil/serialise.cc
index 80b111f08..11bc183cc 100644
--- a/src/libutil/serialise.cc
+++ b/src/libutil/serialise.cc
@@ -330,55 +330,40 @@ void writePadding(size_t len, Sink & sink)
 }
 
 
-void writeString(std::string_view data, Sink & sink)
+WireFormatGenerator SerializingTransform::operator()(std::string_view s)
 {
-    sink << data.size();
-    sink(data);
-    writePadding(data.size(), sink);
+    co_yield s.size();
+    co_yield Bytes(s.begin(), s.size());
+    co_yield SerializingTransform::padding(s.size());
 }
 
-
-Sink & operator << (Sink & sink, std::string_view s)
-{
-    writeString(s, sink);
-    return sink;
-}
-
-
-template<class T> void writeStrings(const T & ss, Sink & sink)
-{
-    sink << ss.size();
-    for (auto & i : ss)
-        sink << i;
-}
-
-Sink & operator << (Sink & sink, const Strings & s)
+WireFormatGenerator SerializingTransform::operator()(const Strings & ss)
 {
-    writeStrings(s, sink);
-    return sink;
+    co_yield ss.size();
+    for (const auto & s : ss)
+        co_yield std::string_view(s);
 }
 
-Sink & operator << (Sink & sink, const StringSet & s)
+WireFormatGenerator SerializingTransform::operator()(const StringSet & ss)
 {
-    writeStrings(s, sink);
-    return sink;
+    co_yield ss.size();
+    for (const auto & s : ss)
+        co_yield std::string_view(s);
 }
 
-Sink & operator << (Sink & sink, const Error & ex)
+WireFormatGenerator SerializingTransform::operator()(const Error & ex)
 {
     auto & info = ex.info();
-    sink
-        << "Error"
-        << info.level
-        << "Error" // removed
-        << info.msg.str()
-        << 0 // FIXME: info.errPos
-        << info.traces.size();
+    co_yield "Error";
+    co_yield info.level;
+    co_yield "Error"; // removed
+    co_yield info.msg.str();
+    co_yield 0; // FIXME: info.errPos
+    co_yield info.traces.size();
     for (auto & trace : info.traces) {
-        sink << 0; // FIXME: trace.pos
-        sink << trace.hint.str();
+        co_yield 0; // FIXME: trace.pos
+        co_yield trace.hint.str();
     }
-    return sink;
 }
 
 
diff --git a/src/libutil/serialise.hh b/src/libutil/serialise.hh
index 491b1987d..2651ec979 100644
--- a/src/libutil/serialise.hh
+++ b/src/libutil/serialise.hh
@@ -350,33 +350,82 @@ inline Sink & operator<<(Sink & sink, Generator<Bytes> && g)
     return sink;
 }
 
+struct SerializingTransform;
+using WireFormatGenerator = Generator<Bytes, SerializingTransform>;
+
+struct SerializingTransform
+{
+    std::array<unsigned char, 8> buf;
+
+    Bytes operator()(uint64_t n)
+    {
+        buf[0] = n & 0xff;
+        buf[1] = (n >> 8) & 0xff;
+        buf[2] = (n >> 16) & 0xff;
+        buf[3] = (n >> 24) & 0xff;
+        buf[4] = (n >> 32) & 0xff;
+        buf[5] = (n >> 40) & 0xff;
+        buf[6] = (n >> 48) & 0xff;
+        buf[7] = (unsigned char) (n >> 56) & 0xff;
+        return {reinterpret_cast<const char *>(buf.begin()), 8};
+    }
+
+    static Bytes padding(size_t unpadded)
+    {
+        return Bytes("\0\0\0\0\0\0\0", unpadded % 8 ? 8 - unpadded % 8 : 0);
+    }
+
+    // opt in to generator chaining. without this co_yielding
+    // another generator of any type will cause a type error.
+    auto operator()(Generator<Bytes> && g)
+    {
+        return std::move(g);
+    }
+
+    // only choose this for *exactly* char spans, do not allow implicit
+    // conversions. this would cause ambiguities with strings literals,
+    // and resolving those with more string-like overloads needs a lot.
+    template<typename Span>
+        requires std::same_as<Span, std::span<char>> || std::same_as<Span, std::span<const char>>
+    Bytes operator()(Span s)
+    {
+        return s;
+    }
+    WireFormatGenerator operator()(std::string_view s);
+    WireFormatGenerator operator()(const Strings & s);
+    WireFormatGenerator operator()(const StringSet & s);
+    WireFormatGenerator operator()(const Error & s);
+};
+
 void writePadding(size_t len, Sink & sink);
-void writeString(std::string_view s, Sink & sink);
 
-inline Sink & operator << (Sink & sink, uint64_t n)
+inline Sink & operator<<(Sink & sink, uint64_t u)
 {
-    unsigned char buf[8];
-    buf[0] = n & 0xff;
-    buf[1] = (n >> 8) & 0xff;
-    buf[2] = (n >> 16) & 0xff;
-    buf[3] = (n >> 24) & 0xff;
-    buf[4] = (n >> 32) & 0xff;
-    buf[5] = (n >> 40) & 0xff;
-    buf[6] = (n >> 48) & 0xff;
-    buf[7] = (unsigned char) (n >> 56) & 0xff;
-    sink({(char *) buf, sizeof(buf)});
-    return sink;
+    return sink << [&]() -> WireFormatGenerator { co_yield u; }();
+}
+
+inline Sink & operator<<(Sink & sink, std::string_view s)
+{
+    return sink << [&]() -> WireFormatGenerator { co_yield s; }();
+}
+
+inline Sink & operator<<(Sink & sink, const Strings & s)
+{
+    return sink << [&]() -> WireFormatGenerator { co_yield s; }();
 }
 
-Sink & operator << (Sink & in, const Error & ex);
-Sink & operator << (Sink & sink, std::string_view s);
-Sink & operator << (Sink & sink, const Strings & s);
-Sink & operator << (Sink & sink, const StringSet & s);
+inline Sink & operator<<(Sink & sink, const StringSet & s)
+{
+    return sink << [&]() -> WireFormatGenerator { co_yield s; }();
+}
 
+inline Sink & operator<<(Sink & sink, const Error & ex)
+{
+    return sink << [&]() -> WireFormatGenerator { co_yield ex; }();
+}
 
 MakeError(SerialisationError, Error);
 
-
 template<typename T>
 T readNum(Source & source)
 {
diff --git a/tests/unit/libutil/serialise.cc b/tests/unit/libutil/serialise.cc
index 95ae43115..78882ad2c 100644
--- a/tests/unit/libutil/serialise.cc
+++ b/tests/unit/libutil/serialise.cc
@@ -2,30 +2,47 @@
 #include "error.hh"
 #include "fmt.hh"
 #include "pos-table.hh"
+#include "generator.hh"
 #include "ref.hh"
 #include "types.hh"
 
+#include <concepts>
+#include <cstdint>
+#include <initializer_list>
 #include <limits.h>
 #include <gtest/gtest.h>
 
 #include <numeric>
+#include <stdexcept>
+#include <string_view>
+#include <type_traits>
 
 namespace nix {
 
-TEST(Sink, uint64_t)
+// don't deduce the type of `val` for added insurance.
+template<typename T>
+static std::string toWire(const std::type_identity_t<T> & val)
 {
-    StringSink s;
-    s << 42;
-    ASSERT_EQ(s.s, std::string({42, 0, 0, 0, 0, 0, 0, 0}));
+    std::string result;
+    auto g = [] (const auto & val) -> WireFormatGenerator { co_yield val; }(val);
+    while (auto buffer = g.next()) {
+        result.append(buffer->data(), buffer->size());
+    }
+    return result;
 }
 
-TEST(Sink, string_view)
+TEST(WireFormatGenerator, uint64_t)
 {
-    StringSink s;
-    s << "";
+    auto s = toWire<uint64_t>(42);
+    ASSERT_EQ(s, std::string({42, 0, 0, 0, 0, 0, 0, 0}));
+}
+
+TEST(WireFormatGenerator, string_view)
+{
+    auto s = toWire<std::string_view>("");
     // clang-format off
     ASSERT_EQ(
-        s.s,
+        s,
         std::string({
             // length
             0, 0, 0, 0, 0, 0, 0, 0,
@@ -34,11 +51,10 @@ TEST(Sink, string_view)
     );
     // clang-format on
 
-    s = {};
-    s << "test";
+    s = toWire<std::string_view>("test");
     // clang-format off
     ASSERT_EQ(
-        s.s,
+        s,
         std::string({
             // length
             4, 0, 0, 0, 0, 0, 0, 0,
@@ -50,11 +66,10 @@ TEST(Sink, string_view)
     );
     // clang-format on
 
-    s = {};
-    s << "longer string";
+    s = toWire<std::string_view>("longer string");
     // clang-format off
     ASSERT_EQ(
-        s.s,
+        s,
         std::string({
             // length
             13, 0, 0, 0, 0, 0, 0, 0,
@@ -67,13 +82,12 @@ TEST(Sink, string_view)
     // clang-format on
 }
 
-TEST(Sink, StringSet)
+TEST(WireFormatGenerator, StringSet)
 {
-    StringSink s;
-    s << StringSet{};
+    auto s = toWire<StringSet>({});
     // clang-format off
     ASSERT_EQ(
-        s.s,
+        s,
         std::string({
             // length
             0, 0, 0, 0, 0, 0, 0, 0,
@@ -82,11 +96,10 @@ TEST(Sink, StringSet)
     );
     // clang-format on
 
-    s = {};
-    s << StringSet{"a", ""};
+    s = toWire<StringSet>({"a", ""});
     // clang-format off
     ASSERT_EQ(
-        s.s,
+        s,
         std::string({
             // length
             2, 0, 0, 0, 0, 0, 0, 0,
@@ -99,13 +112,12 @@ TEST(Sink, StringSet)
     // clang-format on
 }
 
-TEST(Sink, Strings)
+TEST(WireFormatGenerator, Strings)
 {
-    StringSink s;
-    s << Strings{};
+    auto s = toWire<Strings>({});
     // clang-format off
     ASSERT_EQ(
-        s.s,
+        s,
         std::string({
             // length
             0, 0, 0, 0, 0, 0, 0, 0,
@@ -114,11 +126,10 @@ TEST(Sink, Strings)
     );
     // clang-format on
 
-    s = {};
-    s << Strings{"a", ""};
+    s = toWire<Strings>({"a", ""});
     // clang-format off
     ASSERT_EQ(
-        s.s,
+        s,
         std::string({
             // length
             2, 0, 0, 0, 0, 0, 0, 0,
@@ -131,23 +142,22 @@ TEST(Sink, Strings)
     // clang-format on
 }
 
-TEST(Sink, Error)
+TEST(WireFormatGenerator, Error)
 {
     PosTable pt;
     auto o = pt.addOrigin(Pos::String{make_ref<std::string>("test")}, 4);
 
-    StringSink s;
-    s << Error{ErrorInfo{
+    auto s = toWire<Error>(Error{ErrorInfo{
         .level = lvlInfo,
         .msg = HintFmt("foo"),
         .pos = pt[pt.add(o, 1)],
         .traces = {{.pos = pt[pt.add(o, 2)], .hint = HintFmt("b %1%", "foo")}},
-    }};
+    }});
     // NOTE position of the error and all traces are ignored
     // by the wire format
     // clang-format off
     ASSERT_EQ(
-        s.s,
+        s,
         std::string({
             5, 0, 0, 0, 0, 0, 0, 0, 'E', 'r', 'r', 'o', 'r', 0, 0, 0,
             3, 0, 0, 0, 0, 0, 0, 0,
@@ -163,4 +173,45 @@ TEST(Sink, Error)
     // clang-format on
 }
 
+TEST(WireFormatGenerator, exampleMessage)
+{
+    auto gen = []() -> WireFormatGenerator {
+        std::set<std::string> foo{"a", "longer string", ""};
+        co_yield 42;
+        co_yield foo;
+        co_yield std::string_view("test");
+        co_yield true;
+    }();
+
+    std::vector<char> full;
+    while (auto s = gen.next()) {
+        full.insert(full.end(), s->begin(), s->end());
+    }
+
+    ASSERT_EQ(
+        full,
+        (std::vector<char>{
+            // clang-format off
+            // 42
+            42, 0, 0, 0, 0, 0, 0, 0,
+            // foo
+            3, 0, 0, 0, 0, 0, 0, 0,
+            /// ""
+            0, 0, 0, 0, 0, 0, 0, 0,
+            /// a
+            1, 0, 0, 0, 0, 0, 0, 0,
+            'a', 0, 0, 0, 0, 0, 0, 0,
+            /// longer string
+            13, 0, 0, 0, 0, 0, 0, 0,
+            'l', 'o', 'n', 'g', 'e', 'r', ' ', 's', 't', 'r', 'i', 'n', 'g', 0, 0, 0,
+            // foo done
+            // test
+            4, 0, 0, 0, 0, 0, 0, 0,
+            't', 'e', 's', 't', 0, 0, 0, 0,
+            // true
+            1, 0, 0, 0, 0, 0, 0, 0,
+            //clang-format on
+            }));
+}
+
 }