feat: add tr_variant::visit() (#7923)

* refactor: make tr_variant work with std::visit()

* refactor: use tr_variant::visit() in tr_variant_serde

* refactor: use tr_variant::visit() in tr_variant::merge()

* refactor: simplify JsonWriter

* fix: clang-tidy misc-use-internal-linkage

* test: simplify VariantTest.visitsNodesDepthFirst()

* test: simplify VariantTest.visitNestedJsonSummarizesStructure()

* feat: add tr_variant::clone()

* test: simplify variant tests

* docs: add code comments for tr_variant::merge() and tr_variant::clone()

* fix: add stack-smashing handling for JSON parsing

* fix: clang-tidy modernize-raw-string-literal

* Use writer.Key() for object key names

* refactor: remove unnecessary FMT_COMPILE() macros

* refactor: fix tr_variant::VisitAdapter to preserve the visitor category

chore: remove unnecessary std::cref() calls when calling tr_variant::visitor()

---------

Co-authored-by: Yat Ho <lagoho7@gmail.com>
This commit is contained in:
Charles Kerr
2025-12-13 13:47:54 -06:00
committed by GitHub
parent 64268a3a7f
commit 0fcda4e977
5 changed files with 559 additions and 729 deletions

View File

@@ -13,6 +13,7 @@
#include <string>
#include <string_view>
#include <utility>
#include <vector>
#include <fmt/compile.h>
#include <fmt/format.h>
@@ -21,7 +22,6 @@
#include "libtransmission/benc.h"
#include "libtransmission/quark.h"
#include "libtransmission/tr-buffer.h"
#include "libtransmission/utils.h"
#include "libtransmission/variant.h"
@@ -275,65 +275,99 @@ namespace
{
namespace to_string_helpers
{
using OutBuf = libtransmission::StackBuffer<1024U * 8U, std::byte>;
using OutBuf = fmt::memory_buffer;
void saveNullFunc(tr_variant const& /*var*/, std::nullptr_t /*val*/, void* vout)
[[nodiscard]] auto sorted_entries(tr_variant::Map const& map)
{
static_cast<OutBuf*>(vout)->add("0:"sv);
auto entries = std::vector<std::pair<std::string_view, tr_variant const*>>{};
entries.reserve(map.size());
for (auto const& [key, child] : map)
{
entries.emplace_back(tr_quark_get_string_view(key), &child);
}
std::sort(std::begin(entries), std::end(entries));
return entries;
}
void saveIntFunc(tr_variant const& /*var*/, int64_t const val, void* vout)
struct BencWriter
{
auto out = static_cast<OutBuf*>(vout);
void operator()(std::monostate /*unused*/) const
{
}
auto const [buf, buflen] = out->reserve_space(64U);
auto* walk = reinterpret_cast<char*>(buf);
auto const* const begin = walk;
walk = fmt::format_to(walk, FMT_COMPILE("i{:d}e"), val);
out->commit_space(walk - begin);
}
void operator()(std::nullptr_t) const
{
write_string(""sv);
}
void saveBoolFunc(tr_variant const& /*var*/, bool const val, void* vout)
{
static_cast<OutBuf*>(vout)->add(val ? "i1e"sv : "i0e"sv);
}
void operator()(bool val) const
{
append_literal(val ? "i1e"sv : "i0e"sv);
}
void saveStringImpl(OutBuf* out, std::string_view sv)
{
// `${sv.size()}:${sv}`
auto const [buf, buflen] = out->reserve_space(std::size(sv) + 32U);
auto* begin = reinterpret_cast<char*>(buf);
auto* const end = fmt::format_to(begin, FMT_COMPILE("{:d}:{:s}"), std::size(sv), sv);
out->commit_space(end - begin);
}
void operator()(int64_t val) const
{
write_int(val);
}
void saveStringFunc(tr_variant const& /*var*/, std::string_view const val, void* vout)
{
saveStringImpl(static_cast<OutBuf*>(vout), val);
}
void operator()(double val) const
{
write_real(val);
}
void saveRealFunc(tr_variant const& /*val*/, double const val, void* vout)
{
// the benc spec doesn't handle floats; save it as a string.
auto buf = std::array<char, 64>{};
auto const* const out = fmt::format_to(std::data(buf), FMT_COMPILE("{:f}"), val);
saveStringImpl(static_cast<OutBuf*>(vout), { std::data(buf), static_cast<size_t>(out - std::data(buf)) });
}
void operator()(std::string_view sv) const
{
write_string(sv);
}
void saveDictBeginFunc(tr_variant const& /*val*/, void* vbuf)
{
static_cast<OutBuf*>(vbuf)->push_back('d');
}
void operator()(tr_variant::Vector const& vec) const
{
out_.push_back('l');
for (auto const& child : vec)
{
child.visit(*this);
}
out_.push_back('e');
}
void saveListBeginFunc(tr_variant const& /*val*/, void* vbuf)
{
static_cast<OutBuf*>(vbuf)->push_back('l');
}
void operator()(tr_variant::Map const& map) const
{
out_.push_back('d');
auto entries = sorted_entries(map);
for (auto const& [key, child] : entries)
{
write_string(key);
child->visit(*this);
}
out_.push_back('e');
}
void saveContainerEndFunc(tr_variant const& /*val*/, void* vbuf)
{
static_cast<OutBuf*>(vbuf)->push_back('e');
}
OutBuf& out_;
private:
void write_string(std::string_view sv) const
{
fmt::format_to(fmt::appender(out_), "{:d}:{:s}", std::size(sv), sv);
}
void write_int(int64_t val) const
{
fmt::format_to(fmt::appender(out_), "i{:d}e", val);
}
void write_real(double val) const
{
auto buf = std::array<char, 64>{};
auto const* const out_ptr = fmt::format_to(std::data(buf), "{:f}", val);
write_string({ std::data(buf), static_cast<size_t>(out_ptr - std::data(buf)) });
}
void append_literal(std::string_view literal) const
{
out_.append(std::data(literal), std::data(literal) + std::size(literal));
}
};
} // namespace to_string_helpers
} // namespace
@@ -342,18 +376,7 @@ std::string tr_variant_serde::to_benc_string(tr_variant const& var)
{
using namespace to_string_helpers;
static auto constexpr Funcs = WalkFuncs{
saveNullFunc, //
saveIntFunc, //
saveBoolFunc, //
saveRealFunc, //
saveStringFunc, //
saveDictBeginFunc, //
saveListBeginFunc, //
saveContainerEndFunc, //
};
auto buf = OutBuf{};
walk(var, Funcs, &buf, true);
return buf.to_string();
var.visit(BencWriter{ buf });
return fmt::to_string(buf);
}

View File

@@ -15,6 +15,7 @@
#include <type_traits>
#include <utility>
#include <variant>
#include <vector>
#include <fmt/format.h>
@@ -95,8 +96,13 @@ struct json_to_variant_handler : public rapidjson::BaseReaderHandler<>
bool StartObject()
{
tr_variantInitDict(push_stack(), prealloc_guess());
return true;
if (auto* node = push_stack())
{
tr_variantInitDict(node, prealloc_guess());
return true;
}
return false;
}
bool Key(Ch const* const str, rapidjson::SizeType const len, bool const copy)
@@ -121,8 +127,13 @@ struct json_to_variant_handler : public rapidjson::BaseReaderHandler<>
bool StartArray()
{
tr_variantInitList(push_stack(), prealloc_guess());
return true;
if (auto* node = push_stack())
{
tr_variantInitList(node, prealloc_guess());
return true;
}
return false;
}
bool EndArray(rapidjson::SizeType const len)
@@ -140,7 +151,7 @@ private:
tr_variant* push_stack() noexcept
{
return stack_.emplace(get_leaf());
return std::size(stack_) < MaxDepth ? stack_.emplace(get_leaf()) : nullptr;
}
void pop_stack(rapidjson::SizeType const len) noexcept
@@ -228,11 +239,14 @@ std::optional<tr_variant> tr_variant_serde::parse_json(std::string_view input)
{
return std::optional<tr_variant>{ std::move(top) };
}
if (auto err_code = reader.GetParseErrorCode(); err_code == rapidjson::kParseErrorDocumentEmpty)
{
error_.set(EINVAL, "No content");
}
else if (err_code == rapidjson::kParseErrorTermination)
{
error_.set(E2BIG, "Max stack depth reached; unable to continue parsing");
}
else
{
error_.set(
@@ -254,112 +268,101 @@ namespace
{
namespace to_string_helpers
{
// implements RapidJSON's Stream concept, so that the library can output
// directly to a std::string, and we can avoid some copying by copy elision
// http://rapidjson.org/md_doc_stream.html
struct string_output_stream
// implements RapidJSON's write-only stream concept using fmt::memory_buffer.
// See <rapidjson/stream.h> for details.
struct FmtOutputStream
{
using Ch = char;
explicit string_output_stream(std::string& str)
: str_ref_{ str }
void Put(Ch const ch)
{
buf_.push_back(ch);
}
void Flush()
{
}
[[nodiscard]] static Ch Peek()
[[nodiscard]] std::string to_string() const
{
TR_ASSERT(false);
return 0;
}
[[nodiscard]] static Ch Take()
{
TR_ASSERT(false);
return 0;
}
static size_t Tell()
{
TR_ASSERT(false);
return 0U;
}
static Ch* PutBegin()
{
TR_ASSERT(false);
return nullptr;
}
void Put(Ch const c)
{
str_ref_ += c;
}
static void Flush()
{
}
static size_t PutEnd(Ch* /*begin*/)
{
TR_ASSERT(false);
return 0U;
return fmt::to_string(buf_);
}
private:
std::string& str_ref_;
fmt::memory_buffer buf_;
};
using writer_var_t = std::variant<rapidjson::Writer<string_output_stream>, rapidjson::PrettyWriter<string_output_stream>>;
void jsonNullFunc(tr_variant const& /*var*/, std::nullptr_t /*val*/, void* vdata)
[[nodiscard]] auto sorted_entries(tr_variant::Map const& map)
{
std::visit([](auto&& writer) { writer.Null(); }, *static_cast<writer_var_t*>(vdata));
}
void jsonIntFunc(tr_variant const& /*var*/, int64_t const val, void* vdata)
{
std::visit([val](auto&& writer) { writer.Int64(val); }, *static_cast<writer_var_t*>(vdata));
}
void jsonBoolFunc(tr_variant const& /*var*/, bool const val, void* vdata)
{
std::visit([val](auto&& writer) { writer.Bool(val); }, *static_cast<writer_var_t*>(vdata));
}
void jsonRealFunc(tr_variant const& /*var*/, double const val, void* vdata)
{
std::visit([val](auto&& writer) { writer.Double(val); }, *static_cast<writer_var_t*>(vdata));
}
void jsonStringFunc(tr_variant const& /*var*/, std::string_view sv, void* vdata)
{
std::visit([sv](auto&& writer) { writer.String(std::data(sv), std::size(sv)); }, *static_cast<writer_var_t*>(vdata));
}
void jsonDictBeginFunc(tr_variant const& /*var*/, void* vdata)
{
std::visit([](auto&& writer) { writer.StartObject(); }, *static_cast<writer_var_t*>(vdata));
}
void jsonListBeginFunc(tr_variant const& /*var*/, void* vdata)
{
std::visit([](auto&& writer) { writer.StartArray(); }, *static_cast<writer_var_t*>(vdata));
}
void jsonContainerEndFunc(tr_variant const& var, void* vdata)
{
auto& writer_var = *static_cast<writer_var_t*>(vdata);
if (var.holds_alternative<tr_variant::Map>())
auto entries = std::vector<std::pair<std::string_view, tr_variant const*>>{};
entries.reserve(map.size());
for (auto const& [key, child] : map)
{
std::visit([](auto&& writer) { writer.EndObject(); }, writer_var);
}
else /* list */
{
std::visit([](auto&& writer) { writer.EndArray(); }, writer_var);
entries.emplace_back(tr_quark_get_string_view(key), &child);
}
std::sort(std::begin(entries), std::end(entries));
return entries;
}
template<typename WriterT>
struct JsonWriter
{
WriterT& writer;
void operator()(std::monostate /*unused*/) const
{
}
void operator()(std::nullptr_t) const
{
writer.Null();
}
void operator()(bool const val) const
{
writer.Bool(val);
}
void operator()(int64_t const val) const
{
writer.Int64(val);
}
void operator()(double const val) const
{
writer.Double(val);
}
void operator()(std::string_view const val) const
{
writer.String(std::data(val), std::size(val));
}
void operator()(tr_variant::Vector const& val) const
{
writer.StartArray();
for (auto const& child : val)
{
child.visit(*this);
}
writer.EndArray();
}
void operator()(tr_variant::Map const& val) const
{
writer.StartObject();
for (auto const& [key, child] : sorted_entries(val))
{
writer.Key(std::data(key), std::size(key));
child->visit(*this);
}
writer.EndObject();
}
};
template<typename WriterT>
JsonWriter(WriterT&) -> JsonWriter<WriterT>;
} // namespace to_string_helpers
} // namespace
@@ -367,30 +370,16 @@ std::string tr_variant_serde::to_json_string(tr_variant const& var) const
{
using namespace to_string_helpers;
static auto constexpr Funcs = WalkFuncs{
jsonNullFunc, //
jsonIntFunc, //
jsonBoolFunc, //
jsonRealFunc, //
jsonStringFunc, //
jsonDictBeginFunc, //
jsonListBeginFunc, //
jsonContainerEndFunc, //
};
auto out = std::string{};
out.reserve(rapidjson::StringBuffer::kDefaultCapacity);
auto stream = string_output_stream{ out };
auto writer = writer_var_t{};
auto buf = FmtOutputStream{};
if (compact_)
{
writer.emplace<0>(stream);
auto writer = rapidjson::Writer{ buf };
var.visit(JsonWriter{ writer });
}
else
{
writer.emplace<1>(stream);
auto writer = rapidjson::PrettyWriter{ buf };
var.visit(JsonWriter{ writer });
}
walk(var, Funcs, &writer, true);
return out;
return buf.to_string();
}

View File

@@ -3,13 +3,13 @@
// or any future license endorsed by Mnemosyne LLC.
// License text can be found in the licenses/ folder.
#include <algorithm> // std::sort
#include <cstddef>
#include <cstdint>
#include <iterator>
#include <optional>
#include <string>
#include <string_view>
#include <type_traits>
#include <utility>
#include <vector>
#include <variant>
@@ -20,8 +20,6 @@
#include <fmt/format.h>
#include <small/vector.hpp>
#define LIBTRANSMISSION_VARIANT_MODULE
#include "libtransmission/error.h"
@@ -35,21 +33,6 @@ using namespace std::literals;
namespace
{
[[nodiscard]] constexpr bool variant_is_container(tr_variant const* const var)
{
return var != nullptr && (var->holds_alternative<tr_variant::Vector>() || var->holds_alternative<tr_variant::Map>());
}
[[nodiscard]] constexpr size_t variant_index(tr_variant const* const var)
{
if (var != nullptr)
{
return var->index();
}
return tr_variant::NoneIndex;
}
template<typename T>
[[nodiscard]] bool value_if(tr_variant const* const var, T* const setme)
{
@@ -199,63 +182,55 @@ tr_variant::StringHolder& tr_variant::StringHolder::operator=(StringHolder&& tha
// ---
tr_variant::Merge::Merge(tr_variant& tgt)
: tgt_{ tgt }
tr_variant tr_variant::clone() const
{
auto ret = tr_variant{};
ret.merge(*this);
return ret;
}
void tr_variant::Merge::operator()(std::monostate const& src)
tr_variant& tr_variant::merge(tr_variant const& that)
{
tgt_ = src;
}
void tr_variant::Merge::operator()(std::nullptr_t const& src)
{
tgt_ = src;
}
void tr_variant::Merge::operator()(bool const& src)
{
tgt_ = src;
}
void tr_variant::Merge::operator()(int64_t const& src)
{
tgt_ = src;
}
void tr_variant::Merge::operator()(double const& src)
{
tgt_ = src;
}
void tr_variant::Merge::operator()(tr_variant::StringHolder const& src)
{
tgt_ = src.sv_;
}
void tr_variant::Merge::operator()(tr_variant::Vector const& src)
{
auto const n_items = std::size(src);
auto& tgt = tgt_.val_.emplace<Vector>();
tgt.resize(n_items);
for (size_t i = 0; i < n_items; ++i)
{
std::visit(Merge{ tgt[i] }, src[i].val_);
}
}
void tr_variant::Merge::operator()(tr_variant::Map const& src)
{
// if tgt_ isn't already a map, make it one
if (tgt_.index() != tr_variant::MapIndex)
{
tgt_.val_.emplace<tr_variant::Map>();
}
if (auto* tgt = tgt_.get_if<tr_variant::MapIndex>(); tgt != nullptr)
{
tgt->reserve(std::size(*tgt) + std::size(src));
for (auto const& [key, val] : src)
that.visit(
[this](auto const& value)
{
std::visit(Merge{ (*tgt)[tr_quark_convert(key)] }, val.val_);
}
}
using ValueType = std::decay_t<decltype(value)>;
if constexpr (
std::is_same_v<ValueType, std::monostate> || std::is_same_v<ValueType, std::nullptr_t> ||
std::is_same_v<ValueType, bool> || std::is_same_v<ValueType, int64_t> || std::is_same_v<ValueType, double> ||
std::is_same_v<ValueType, std::string_view>)
{
*this = value;
}
else if constexpr (std::is_same_v<ValueType, Vector>)
{
auto& dest = val_.emplace<Vector>();
dest.resize(std::size(value));
for (size_t i = 0; i < std::size(value); ++i)
{
dest[i].merge(value[i]);
}
}
else if constexpr (std::is_same_v<ValueType, Map>)
{
if (index() != MapIndex)
{
val_.emplace<Map>();
}
if (auto* dest = this->template get_if<MapIndex>(); dest != nullptr)
{
dest->reserve(std::size(*dest) + std::size(value));
for (auto const& [key, child] : value)
{
(*dest)[tr_quark_convert(key)].merge(child);
}
}
}
});
return *this;
}
// ---
@@ -552,267 +527,6 @@ bool tr_variantDictRemove(tr_variant* const var, tr_quark key)
return false;
}
// --- BENC WALKING
class WalkNode
{
public:
WalkNode() = default;
explicit WalkNode(tr_variant const* const var)
: var_{ var }
{
}
std::pair<tr_quark, tr_variant const*> next_child()
{
if (var_ == nullptr)
{
return {};
}
if (auto const* const map = var_->get_if<tr_variant::MapIndex>(); map != nullptr)
{
if (auto idx = next_index(); idx < std::size(*map))
{
auto iter = std::cbegin(*map);
std::advance(iter, idx);
return { iter->first, &iter->second };
}
}
else if (auto const* const vec = var_->get_if<tr_variant::VectorIndex>(); vec != nullptr)
{
if (auto idx = next_index(); idx < std::size(*vec))
{
return { {}, &vec->at(idx) };
}
}
return {};
}
[[nodiscard]] constexpr auto is_visited() const noexcept
{
return is_visited_;
}
constexpr void set_visited() noexcept
{
is_visited_ = true;
}
[[nodiscard]] tr_variant const* current() const noexcept
{
return var_;
}
protected:
friend class VariantWalker;
tr_variant const* var_ = nullptr;
bool is_visited_ = false;
void assign(tr_variant const* v_in)
{
var_ = v_in;
is_visited_ = false;
child_index_ = 0;
sorted_.clear();
}
struct ByKey
{
std::string_view key;
size_t idx = {};
};
template<typename Container>
void sort(Container& sortbuf)
{
auto const* const map = var_ != nullptr ? var_->get_if<tr_variant::MapIndex>() : nullptr;
if (map == nullptr)
{
return;
}
auto idx = size_t{};
auto const n = std::size(*map);
sortbuf.resize(n);
for (auto const& [key, val] : *map)
{
sortbuf[idx] = { tr_quark_get_string_view(key), idx };
++idx;
}
std::sort(std::begin(sortbuf), std::end(sortbuf), [](ByKey const& a, ByKey const& b) { return a.key < b.key; });
// keep the sorted indices
sorted_.resize(n);
for (size_t i = 0; i < n; ++i)
{
sorted_[i] = sortbuf[i].idx;
}
}
private:
// When walking `v`'s children, this is the index of the next child
size_t child_index_ = 0;
// When `v` is a dict, this is its children's indices sorted by key.
// Bencoded dicts must be sorted, so this is useful when writing benc.
small::vector<size_t, 128U> sorted_;
[[nodiscard]] size_t next_index()
{
auto idx = child_index_++;
if (idx < std::size(sorted_))
{
idx = sorted_[idx];
}
return idx;
}
};
class VariantWalker
{
public:
void emplace(tr_variant const* v_in, bool sort_dicts)
{
stack_.emplace_back(v_in);
if (sort_dicts)
{
top().sort(sortbuf_);
}
}
void pop()
{
TR_ASSERT(!std::empty(stack_));
if (auto const size = std::size(stack_); size != 0U)
{
stack_.resize(size - 1U);
}
}
[[nodiscard]] bool empty() const noexcept
{
return std::empty(stack_);
}
WalkNode& top()
{
TR_ASSERT(!std::empty(stack_));
return stack_.back();
}
private:
static auto constexpr InitialCapacity = size_t{ 24U };
small::vector<WalkNode, InitialCapacity> stack_;
small::vector<WalkNode::ByKey, InitialCapacity> sortbuf_;
};
/**
* This function's previous recursive implementation was
* easier to read, but was vulnerable to a smash-stacking
* attack via maliciously-crafted data. (#667)
*/
void tr_variant_serde::walk(tr_variant const& top, WalkFuncs const& walk_funcs, void* user_data, bool sort_dicts)
{
auto stack = VariantWalker{};
stack.emplace(&top, sort_dicts);
while (!stack.empty())
{
auto& node = stack.top();
tr_variant const* v = nullptr;
if (!node.is_visited())
{
v = node.current();
node.set_visited();
}
else
{
auto [key, child] = node.next_child();
v = child;
if (v != nullptr)
{
if (node.current()->holds_alternative<tr_variant::Map>())
{
auto const keystr = tr_quark_get_string_view(key);
walk_funcs.string_func(tr_variant::unmanaged_string(keystr), keystr, user_data);
}
}
else // finished with this node
{
if (variant_is_container(node.current()))
{
walk_funcs.container_end_func(*node.current(), user_data);
}
stack.pop();
continue;
}
}
switch (variant_index(v))
{
case tr_variant::NullIndex:
walk_funcs.null_func(*v, *v->get_if<tr_variant::NullIndex>(), user_data);
break;
case tr_variant::BoolIndex:
walk_funcs.bool_func(*v, *v->get_if<tr_variant::BoolIndex>(), user_data);
break;
case tr_variant::IntIndex:
walk_funcs.int_func(*v, *v->get_if<tr_variant::IntIndex>(), user_data);
break;
case tr_variant::DoubleIndex:
walk_funcs.double_func(*v, *v->get_if<tr_variant::DoubleIndex>(), user_data);
break;
case tr_variant::StringIndex:
walk_funcs.string_func(*v, *v->get_if<tr_variant::StringIndex>(), user_data);
break;
case tr_variant::VectorIndex:
if (v == node.current())
{
walk_funcs.list_begin_func(*v, user_data);
}
else
{
stack.emplace(v, sort_dicts);
}
break;
case tr_variant::MapIndex:
if (v == node.current())
{
walk_funcs.dict_begin_func(*v, user_data);
}
else
{
stack.emplace(v, sort_dicts);
}
break;
default: // NoneIndex:
break;
}
}
}
// ---
bool tr_variantDictChild(tr_variant* const var, size_t pos, tr_quark* key, tr_variant** setme_value)

View File

@@ -8,6 +8,7 @@
#include <algorithm> // std::move()
#include <cstddef> // size_t
#include <cstdint> // int64_t
#include <functional> // std::invoke
#include <initializer_list>
#include <optional>
#include <string>
@@ -250,11 +251,15 @@ public:
constexpr tr_variant() noexcept = default;
~tr_variant() = default;
tr_variant(tr_variant const&) = delete;
tr_variant(tr_variant&& that) noexcept = default;
tr_variant& operator=(tr_variant const&) = delete;
tr_variant& operator=(tr_variant&& that) noexcept = default;
// Copying a variant is potentially expensive, so copy assignment
// and copy construct are deleted here to prevent accidental copies.
// Use clone() instead.
tr_variant(tr_variant const&) = delete;
tr_variant& operator=(tr_variant const&) = delete;
template<typename Val>
tr_variant(Val&& value) // NOLINT(bugprone-forwarding-reference-overload, google-explicit-constructor)
{
@@ -430,12 +435,28 @@ public:
val_.emplace<std::monostate>();
}
tr_variant& merge(tr_variant const& that)
template<typename Visitor>
[[nodiscard]] constexpr decltype(auto) visit(Visitor&& visitor)
{
std::visit(Merge{ *this }, that.val_);
return *this;
return std::visit(make_visit_adapter(std::forward<Visitor>(visitor)), val_);
}
template<typename Visitor>
[[nodiscard]] constexpr decltype(auto) visit(Visitor&& visitor) const
{
return std::visit(make_visit_adapter(std::forward<Visitor>(visitor)), val_);
}
// Usually updates `this` to hold a clone of `that`, with two exceptions:
// 1. If both sides hold maps, recursively merge each entry and overwrite
// duplicate keys from `this`.
// 2. Any unmanaged string taken from `that` is copied so `this` owns its copy.
tr_variant& merge(tr_variant const& that);
// Returns a new copy of `this`.
// Any unmanaged strings in `this` are copied so the new variant owns its copy.
[[nodiscard]] tr_variant clone() const;
private:
// Holds a string_view to either an unmanaged/external string or to
// one owned by the class. If the string is unmanaged, only sv_ is used.
@@ -458,23 +479,59 @@ private:
std::string str_;
};
class Merge
template<typename Visitor>
class VisitAdapter
{
public:
explicit Merge(tr_variant& tgt);
void operator()(std::monostate const& src);
void operator()(std::nullptr_t const& src);
void operator()(bool const& src);
void operator()(int64_t const& src);
void operator()(double const& src);
void operator()(tr_variant::StringHolder const& src);
void operator()(tr_variant::Vector const& src);
void operator()(tr_variant::Map const& src);
explicit constexpr VisitAdapter(Visitor visitor)
: visitor_{ std::move(visitor) }
{
}
// These ref-qualified overloads preserve the visitor's cv/ref category
// (lvalue/rvalue, const/non-const) to match std::visit() semantics.
template<typename T>
[[nodiscard]] constexpr decltype(auto) operator()(T&& value) &
{
return call(*this, std::forward<T>(value));
}
template<typename T>
[[nodiscard]] constexpr decltype(auto) operator()(T&& value) const&
{
return call(*this, std::forward<T>(value));
}
template<typename T>
[[nodiscard]] constexpr decltype(auto) operator()(T&& value) &&
{
return call(std::move(*this), std::forward<T>(value));
}
private:
tr_variant& tgt_;
template<typename Self, typename T>
[[nodiscard]] static constexpr decltype(auto) call(Self&& self, T&& value)
{
auto&& visitor = std::forward<Self>(self).visitor_;
if constexpr (std::is_same_v<std::decay_t<T>, StringHolder>)
{
return std::invoke(std::forward<decltype(visitor)>(visitor), value.sv_);
}
else
{
return std::invoke(std::forward<decltype(visitor)>(visitor), std::forward<T>(value));
}
}
Visitor visitor_;
};
template<typename Visitor>
[[nodiscard]] static constexpr auto make_visit_adapter(Visitor&& visitor)
{
using AdaptedVisitor = VisitAdapter<Visitor>;
return AdaptedVisitor{ std::forward<Visitor>(visitor) };
}
std::variant<std::monostate, std::nullptr_t, bool, int64_t, double, StringHolder, Vector, Map> val_;
};
@@ -637,18 +694,6 @@ private:
Json
};
struct WalkFuncs
{
void (*null_func)(tr_variant const& var, std::nullptr_t val, void* user_data);
void (*int_func)(tr_variant const& var, int64_t val, void* user_data);
void (*bool_func)(tr_variant const& var, bool val, void* user_data);
void (*double_func)(tr_variant const& var, double val, void* user_data);
void (*string_func)(tr_variant const& var, std::string_view val, void* user_data);
void (*dict_begin_func)(tr_variant const& var, void* user_data);
void (*list_begin_func)(tr_variant const& var, void* user_data);
void (*container_end_func)(tr_variant const& var, void* user_data);
};
explicit tr_variant_serde(Type type)
: type_{ type }
{
@@ -660,8 +705,6 @@ private:
[[nodiscard]] std::string to_json_string(tr_variant const& var) const;
[[nodiscard]] static std::string to_benc_string(tr_variant const& var);
static void walk(tr_variant const& top, WalkFuncs const& walk_funcs, void* user_data, bool sort_dicts);
Type type_;
bool compact_ = false;

View File

@@ -7,8 +7,11 @@
#include <cerrno>
#include <cstddef> // size_t
#include <cstdint> // int64_t
#include <map>
#include <string>
#include <string_view>
#include <type_traits>
#include <utility>
#include <vector>
#define LIBTRANSMISSION_VARIANT_MODULE
@@ -24,12 +27,22 @@
using namespace std::literals;
class VariantTest : public ::testing::Test
using VariantTest = ::testing::Test;
namespace
{
protected:
static void expectVariantMatchesQuark(tr_quark key);
template<class... Ts>
struct Overloaded : Ts...
{
using Ts::operator()...;
};
template<class... Ts>
Overloaded(Ts...) -> Overloaded<Ts...>;
} // namespace
#ifndef _WIN32
#define STACK_SMASH_DEPTH (1 * 1000 * 1000)
#else
@@ -86,33 +99,41 @@ TEST_F(VariantTest, getType)
EXPECT_EQ(strkey, *sv);
}
// static
void VariantTest::expectVariantMatchesQuark(tr_quark const key)
TEST_F(VariantTest, mergeStringsTakesOwnership)
{
auto const key_sv = tr_quark_get_string_view(key);
auto const is_equal_string = [](std::string_view const a, std::string_view const b)
{
return a == b;
};
auto const var = tr_variant::unmanaged_string(key);
auto const var_sv = var.value_if<std::string_view>();
ASSERT_TRUE(var_sv);
auto const is_same_address = [](std::string_view const a, std::string_view const b)
{
return std::data(a) == std::data(b);
};
// The strings should not just be equal,
// but should point to literally the same memory
EXPECT_EQ(key_sv, *var_sv);
EXPECT_EQ(std::data(key_sv), std::data(*var_sv));
}
// set up `src` to hold an unmanaged string
auto constexpr Original = "this is the string"sv;
auto const src = tr_variant::unmanaged_string(Original);
auto src_sv = src.value_if<std::string_view>().value_or(""sv);
TEST_F(VariantTest, unmanagedStringFromPredefinedQuark)
{
expectVariantMatchesQuark(TR_KEY_name);
}
// set up `tgt` to hold another unmanaged string
auto constexpr WillBeReplaced = "some other string"sv;
static_assert(Original != WillBeReplaced);
auto tgt = tr_variant::unmanaged_string(WillBeReplaced);
auto tgt_sv = tgt.value_if<std::string_view>().value_or(""sv);
TEST_F(VariantTest, unmanagedStringFromNewQuark)
{
static auto constexpr NewString = std::string_view{ "this-string-is-not-already-interned" };
ASSERT_FALSE(tr_quark_lookup(NewString));
// test that `src` and `tgt` hold unmanaged strings
EXPECT_TRUE(is_equal_string(Original, src_sv));
EXPECT_TRUE(is_equal_string(WillBeReplaced, tgt_sv));
EXPECT_TRUE(is_same_address(Original, src_sv));
EXPECT_TRUE(is_same_address(WillBeReplaced, tgt_sv));
auto const key = tr_quark_new(NewString);
expectVariantMatchesQuark(key);
tgt.merge(src);
// test that `tgt` now holds its own copy of `Original`.
auto const actual = tgt.value_if<std::string_view>().value_or(""sv);
EXPECT_TRUE(is_equal_string(Original, actual));
EXPECT_FALSE(is_same_address(Original, actual));
}
TEST_F(VariantTest, parseInt)
@@ -379,73 +400,72 @@ TEST_F(VariantTest, bencToJson)
}
}
TEST_F(VariantTest, merge)
TEST_F(VariantTest, mergeMapsCreatesCombinedMap)
{
auto const i1 = tr_quark_new("i1"sv);
auto const i2 = tr_quark_new("i2"sv);
auto const i3 = tr_quark_new("i3"sv);
auto const i4 = tr_quark_new("i4"sv);
auto const s5 = tr_quark_new("s5"sv);
auto const s6 = tr_quark_new("s6"sv);
auto const s7 = tr_quark_new("s7"sv);
auto const s8 = tr_quark_new("s8"sv);
auto serde = tr_variant_serde::json();
serde.compact();
serde.inplace();
/* initial dictionary (default values) */
auto dest = tr_variant::make_map(6U);
auto* map = dest.get_if<tr_variant::Map>();
map->try_emplace(i1, 1);
map->try_emplace(i2, 2);
map->try_emplace(i4, -35); /* remains untouched */
map->try_emplace(s5, "abc");
map->try_emplace(s6, "def");
map->try_emplace(s7, "127.0.0.1"); /* remains untouched */
/* new dictionary, will overwrite items in dest */
auto src = tr_variant::make_map(6U);
map = src.get_if<tr_variant::Map>();
map->try_emplace(i1, 1); /* same value */
map->try_emplace(i2, 4); /* new value */
map->try_emplace(i3, 3); /* new key:value */
map->try_emplace(s5, "abc"); /* same value */
map->try_emplace(s6, "xyz"); /* new value */
map->try_emplace(s8, "ghi"); /* new key:value */
dest.merge(src);
map = dest.get_if<tr_variant::Map>();
auto i = map->value_if<int64_t>(i1);
ASSERT_TRUE(i);
EXPECT_EQ(1, *i);
i = map->value_if<int64_t>(i2);
ASSERT_TRUE(i);
EXPECT_EQ(4, *i);
i = map->value_if<int64_t>(i3);
ASSERT_TRUE(i);
EXPECT_EQ(3, *i);
i = map->value_if<int64_t>(i4);
ASSERT_TRUE(i);
EXPECT_EQ(-35, *i);
auto sv = map->value_if<std::string_view>(s5);
ASSERT_TRUE(sv);
EXPECT_EQ("abc"sv, *sv);
sv = map->value_if<std::string_view>(s6);
ASSERT_TRUE(sv);
EXPECT_EQ("xyz"sv, *sv);
sv = map->value_if<std::string_view>(s7);
ASSERT_TRUE(sv);
EXPECT_EQ("127.0.0.1"sv, *sv);
sv = map->value_if<std::string_view>(s8);
ASSERT_TRUE(sv);
EXPECT_EQ("ghi"sv, *sv);
auto src = serde.parse(R"({"src_key":123})"sv).value_or(tr_variant{});
auto tgt = serde.parse(R"({"tgt_key":456})"sv).value_or(tr_variant{});
tgt.merge(src);
EXPECT_EQ(R"({"src_key":123,"tgt_key":456})"sv, serde.to_string(tgt));
}
TEST_F(VariantTest, stackSmash)
TEST_F(VariantTest, mergeMapsOverwritesSrcMapEntries)
{
// make a nested list of list of lists.
auto serde = tr_variant_serde::json();
serde.compact();
serde.inplace();
auto src = serde.parse(R"({"src_key": 123, "dup_key":789})"sv).value_or(tr_variant{});
auto tgt = serde.parse(R"({"tgt_key": 456, "dup_key":456})"sv).value_or(tr_variant{});
tgt.merge(src);
EXPECT_EQ(R"({"dup_key":789,"src_key":123,"tgt_key":456})"sv, serde.to_string(tgt));
}
TEST_F(VariantTest, mergeOverwritesDifferingTypes)
{
auto const variants = std::array<std::pair<tr_variant, std::string_view>, 7U>{ {
{ tr_variant{ true }, "true" },
{ tr_variant{ int64_t{ 123 } }, "123" },
{ tr_variant{ 4.5 }, "4.5" },
{ tr_variant{ "foo"sv }, R"("foo")"sv },
{ tr_variant{ nullptr }, "null"sv },
{ tr_variant::make_map(0U), "{}"sv },
{ tr_variant::make_vector(), "[]"sv },
} };
auto serde = tr_variant_serde::json();
serde.compact();
serde.inplace();
for (auto const& [src, src_expected] : variants)
{
for (auto const& [tgt, tgt_expected] : variants)
{
if (&src != &tgt)
{
// set up `var` to be a copy of `src`
auto var = src.clone();
EXPECT_EQ(src_expected, serde.to_string(var));
var.merge(tgt);
// test that `var` is now a copy of `tgt`
EXPECT_EQ(tgt_expected, serde.to_string(var));
}
}
}
}
TEST_F(VariantTest, stackSmashBenc)
{
// set up a nested list of list of lists.
static int constexpr Depth = STACK_SMASH_DEPTH;
std::string const in = std::string(Depth, 'l') + std::string(Depth, 'e');
// confirm that it fails instead of crashing
// test that parsing fails without crashing
auto serde = tr_variant_serde::benc();
auto var = serde.inplace().parse(in);
EXPECT_FALSE(var.has_value());
@@ -453,7 +473,23 @@ TEST_F(VariantTest, stackSmash)
EXPECT_EQ(E2BIG, serde.error_.code());
}
TEST_F(VariantTest, boolAndIntRecast)
TEST_F(VariantTest, stackSmashJson)
{
auto serde = tr_variant_serde::json();
serde.inplace();
// set up a nested array of arrays of arrays.
static auto constexpr Depth = STACK_SMASH_DEPTH;
auto const in = std::string(Depth, '[') + std::string(Depth, ']');
// test that parsing fails without crashing
auto var = serde.inplace().parse(in);
EXPECT_FALSE(var.has_value());
EXPECT_TRUE(serde.error_);
EXPECT_EQ(E2BIG, serde.error_.code());
}
TEST_F(VariantTest, valueIfCanReadBoolsAndIntsInterchangeably)
{
auto const key1 = tr_quark_new("key1"sv);
auto const key2 = tr_quark_new("key2"sv);
@@ -558,134 +594,159 @@ TEST_F(VariantTest, dictFindType)
TEST_F(VariantTest, mapContains)
{
auto const key_bool = tr_quark_new("contains-bool"sv);
auto const key_int = tr_quark_new("contains-int"sv);
auto const key_double = tr_quark_new("contains-double"sv);
auto const key_string = tr_quark_new("contains-string"sv);
auto const key_vector = tr_quark_new("contains-vector"sv);
auto const key_map = tr_quark_new("contains-map"sv);
auto const key_missing = tr_quark_new("contains-missing"sv);
auto const nested_key = tr_quark_new("contains-nested"sv);
auto serde = tr_variant_serde::json();
serde.inplace();
serde.compact();
// populate a test map
auto top = tr_variant::make_map(6U);
// set up a map with some sample entries
static auto constexpr Input = R"({
"id": 42,
"is_finished": true,
"labels": ["a", "b"],
"units": { "speed_units": ["KB/s", "MB/s", "GB/s", "TB/s"] },
"upload_ratio": 4.2,
"version": "5.0"
})"sv;
auto top = serde.parse(Input).value_or(tr_variant{});
auto* const map = top.get_if<tr_variant::Map>();
ASSERT_NE(map, nullptr);
ASSERT_NE(nullptr, map);
map->try_emplace(key_bool, true);
map->try_emplace(key_int, int64_t{ 42 });
map->try_emplace(key_double, 4.2);
map->try_emplace(key_string, "needle"sv);
// test that contains() returns true for entries that exist
EXPECT_TRUE(map->contains(TR_KEY_id));
EXPECT_TRUE(map->contains(TR_KEY_is_finished));
EXPECT_TRUE(map->contains(TR_KEY_labels));
EXPECT_TRUE(map->contains(TR_KEY_units));
EXPECT_TRUE(map->contains(TR_KEY_upload_ratio));
EXPECT_TRUE(map->contains(TR_KEY_version));
auto vec = tr_variant::Vector{};
vec.emplace_back(true);
vec.emplace_back(int64_t{ 7 });
map->try_emplace(key_vector, std::move(vec));
// test that contains() returns false for entries that never existed
EXPECT_FALSE(map->contains(TR_KEY_umask));
auto nested = tr_variant::make_map(1U);
auto* nested_map = nested.get_if<tr_variant::Map>();
ASSERT_NE(nested_map, nullptr);
nested_map->try_emplace(nested_key, "nested"sv);
map->try_emplace(key_map, std::move(nested));
// ---
// test: returns true for entries that exist
EXPECT_TRUE(map->contains(key_bool));
EXPECT_TRUE(map->contains(key_double));
EXPECT_TRUE(map->contains(key_int));
EXPECT_TRUE(map->contains(key_map));
EXPECT_TRUE(map->contains(key_string));
EXPECT_TRUE(map->contains(key_vector));
// test: returns false for entries that never existed
EXPECT_FALSE(map->contains(key_missing));
// test: returns false for entries that were removed
EXPECT_EQ(1U, map->erase(key_vector));
EXPECT_FALSE(map->contains(key_vector));
// test that contains() returns false for entries that were removed
auto const key = TR_KEY_labels;
EXPECT_TRUE(map->contains(key));
EXPECT_EQ(1U, map->erase(key));
EXPECT_FALSE(map->contains(key));
}
TEST_F(VariantTest, mapReplaceKey)
TEST_F(VariantTest, visitStringExposesStringView)
{
auto constexpr IntVal = int64_t{ 73 };
auto const key_bool = tr_quark_new("replace-bool"sv);
auto const key_int = tr_quark_new("replace-int"sv);
auto const key_double = tr_quark_new("replace-double"sv);
auto const key_string = tr_quark_new("replace-string"sv);
auto const key_vector = tr_quark_new("replace-vector"sv);
auto const key_map = tr_quark_new("replace-map"sv);
auto const key_duplicate = tr_quark_new("replace-duplicate"sv);
auto const key_missing_src = tr_quark_new("replace-missing-src"sv);
auto const key_missing_tgt = tr_quark_new("replace-missing-tgt"sv);
auto const key_replacement = tr_quark_new("replace-string-new"sv);
auto const key_nested = tr_quark_new("replace-nested"sv);
static auto const Text = "visit-string"sv;
auto var = tr_variant{ std::string{ Text } };
auto called = false;
// populate a sample map
var.visit(
Overloaded{ [&](std::string_view sv)
{
called = true;
EXPECT_EQ(Text, sv);
},
[](auto&&)
{
FAIL();
} });
auto top = tr_variant::make_map(7U);
auto* const map = top.get_if<tr_variant::Map>();
ASSERT_NE(map, nullptr);
EXPECT_TRUE(called);
}
map->try_emplace(key_bool, true);
map->try_emplace(key_int, IntVal);
map->try_emplace(key_double, 7.3);
map->try_emplace(key_string, "string"sv);
TEST_F(VariantTest, visitConstVariant)
{
auto var = tr_variant::make_vector(1U);
auto* vec = var.get_if<tr_variant::Vector>();
ASSERT_NE(vec, nullptr);
vec->emplace_back(int64_t{ 99 });
auto vec = tr_variant::Vector{};
vec.emplace_back(false);
vec.emplace_back(int64_t{ 99 });
map->try_emplace(key_vector, std::move(vec));
auto const result = std::as_const(var).visit(
Overloaded{ [](tr_variant::Vector const& values) -> int64_t
{
EXPECT_EQ(1U, std::size(values));
return values[0].value_if<int64_t>().value_or(-1);
},
[](auto&&) -> int64_t
{
ADD_FAILURE() << "unexpected alternative";
return -1;
} });
auto nested = tr_variant::make_map(1U);
auto* nested_map = nested.get_if<tr_variant::Map>();
ASSERT_NE(nested_map, nullptr);
nested_map->try_emplace(key_nested, "nested"sv);
map->try_emplace(key_map, std::move(nested));
EXPECT_EQ(99, result);
}
map->try_emplace(key_duplicate, "occupied"sv);
TEST_F(VariantTest, visitsNodesDepthFirst)
{
auto serde = tr_variant_serde::json();
serde.compact();
serde.inplace();
// ---
// set up a test variant to be visited
static auto constexpr Input = R"({
"files": [
{ "name": "file1", "size": 5, "pieces": [1, 2] },
{ "name": "file2", "size": 7, "pieces": [] }
],
"meta": { "active": true }
})"sv;
auto const var = serde.parse(Input).value_or(tr_variant{});
// test: neither src nor tgt exist
auto const serde = tr_variant_serde::json();
auto expected = serde.to_string(top);
EXPECT_FALSE(map->contains(key_missing_src));
EXPECT_FALSE(map->contains(key_missing_tgt));
EXPECT_FALSE(map->replace_key(key_missing_src, key_missing_tgt));
EXPECT_FALSE(map->contains(key_missing_src));
EXPECT_FALSE(map->contains(key_missing_tgt));
auto actual = serde.to_string(top);
EXPECT_EQ(expected, actual); // confirm variant is unchanged
// set up some containers that we'll populate during `var.visit()`
auto visited_counts = std::map<size_t, size_t>{};
auto flattened = tr_variant::Vector{};
flattened.reserve(64U);
// test: src doesn't exist
expected = serde.to_string(top);
EXPECT_FALSE(map->contains(key_missing_src));
EXPECT_EQ(IntVal, map->value_if<int64_t>(key_int).value_or(!IntVal));
EXPECT_FALSE(map->replace_key(key_missing_src, key_int));
EXPECT_FALSE(map->contains(key_missing_src));
EXPECT_EQ(IntVal, map->value_if<int64_t>(key_int).value_or(!IntVal));
actual = serde.to_string(top);
EXPECT_EQ(expected, actual); // confirm variant is unchanged
// set up the visitor
auto flatten = [&](tr_variant const& node, auto const& self) -> void
{
++visited_counts[node.index()];
// test: tgt already exists
expected = serde.to_string(top);
EXPECT_TRUE(map->contains(key_int));
EXPECT_TRUE(map->contains(key_string));
EXPECT_FALSE(map->replace_key(key_int, key_string));
EXPECT_TRUE(map->contains(key_int));
EXPECT_TRUE(map->contains(key_string));
actual = serde.to_string(top);
EXPECT_EQ(expected, actual); // confirm variant is unchanged
node.visit(
[&](auto const& val)
{
using ValueType = std::decay_t<decltype(val)>;
// test: successful replacement
EXPECT_TRUE(map->contains(key_int));
EXPECT_FALSE(map->contains(key_replacement));
EXPECT_TRUE(map->replace_key(key_int, key_replacement));
EXPECT_FALSE(map->contains(key_int));
EXPECT_TRUE(map->contains(key_replacement));
EXPECT_EQ(IntVal, map->value_if<int64_t>(key_replacement).value_or(!IntVal));
if constexpr (
std::is_same_v<ValueType, bool> || //
std::is_same_v<ValueType, double> || //
std::is_same_v<ValueType, int64_t> || //
std::is_same_v<ValueType, std::monostate> || //
std::is_same_v<ValueType, std::nullptr_t> || //
std::is_same_v<ValueType, std::string_view>)
{
flattened.emplace_back(val);
}
else if constexpr (std::is_same_v<ValueType, tr_variant::Vector>)
{
for (auto const& child : val)
{
self(child, self);
}
}
else if constexpr (std::is_same_v<ValueType, tr_variant::Map>)
{
for (auto const& [key, child] : val)
{
flattened.emplace_back(tr_variant::unmanaged_string(key));
self(child, self);
}
}
});
};
flatten(var, flatten);
// test that the nodes were visited depth-first
auto const actual = serde.to_string({ std::move(flattened) });
auto constexpr Expected =
R"(["files","name","file1","size",5,"pieces",1,2,"name","file2","size",7,"pieces","meta","active",true])"sv;
EXPECT_EQ(Expected, actual);
// test that we visited the expected number of nodes
auto const expected_visited_count = std::map<size_t, size_t>{
{ tr_variant::BoolIndex, 1U }, //
{ tr_variant::IntIndex, 4U }, //
{ tr_variant::MapIndex, 4U }, //
{ tr_variant::StringIndex, 2U }, //
{ tr_variant::VectorIndex, 3U }, //
};
EXPECT_EQ(expected_visited_count, visited_counts);
}
TEST_F(VariantTest, variantFromBufFuzz)