Provide separate control on error propagation for encoding and decoding.
Default is throw during encode but bool for encode to reduce sensitivity to DoS attacks (see benchmarks).

Add an operator() in the class dumper (printer.hpp) for IE_TAG that is used in print_all()

template <class IE>
constexpr void operator() (IE const& ie, IE_TAG) const noexcept {}

Example:

sink s;
diameter::slh::proto proto;
med2::print_all(s, proto); // error
..

//print all (named and not) IEs in full depth
template <class SINK, class IE, std::size_t MAX_LINE = 128>
void print_all(SINK&& sink, IE const& ie)

Compile error:

cannot convert 'med2::IE_TAG{}' (type 'med2::IE_TAG') to type 'med2::PRIMITIVE'
encoder(ie, IE_TAG{});

In file included from ../med/med.hpp:9,
from ../ut/ut.hpp:14,
from ../ut/med.cpp:1:
../med/choice.hpp:13:10: fatal error: brigand.hpp: No such file or directory
13 | #include <brigand.hpp>

public:
	template <class PAD_TRAITS, class FUNC>
	struct padder_type
	{
		constexpr explicit padder_type(FUNC& func) noexcept {}
		static constexpr void enable_padding(bool) noexcept { }
		static constexpr void add_padding() noexcept {}
	};

Example:

gtpu::proto proto;
char const* err = gtpu::decode(data, size, proto);
sink s;
med2::print(s, proto, max_depth); // error

Compile error:

../microcore/gate/rfms/src/msg_dump.cpp:217:13: required from here
../microcore/common/codec/med2/encode.hpp:274:47: error: no class template named 'padder_type' in 'class med2::printer<sink_t<100000>&, 128>'
274 | using pad_t = typename ENCODER::template padder_type<pad_traits, ENCODER>;

choice.hpp

when exceptions are used for error reporting and malformed IE appears inside length constrained sequence the exception will be caught and treated as end of sequence.
need to introduce typed exceptions to resolve this but 1st include UT to detect such case.

Денис, привет.

В продолжение нашего разговора, оставляю обновление для ascii_string. Предлагаю добавить этот метод в выше упомянутый класс.

template <std::size_t N>
bool set(char const(&arr)[N]) { return this->set_encoded(std::strlen(arr), arr); }

Тест для этого случая

TEST(common, ascii_string_set)
{
	char arr[8] = {0};
	strcpy(arr, "str");
	med2::ascii_string s;
	s.set(arr);
	EXPECT_EQ(s.get().size(), strlen(arr));
}

extend mandatory (optional?) with condition and default value for example to describe GTP-C message priority.

Case:

struct MySeq : med::sequence<
    M<Flag>,
    O<Case1, Flag::is_case1>, // is_case1 returns true if Flag.get() == 1
    O<Case2, Flag::is_case2, med::inf>// is_case1 returns true if Flag.get() == 2
>
struct Msg : med::sequence<
    M<L, MySeq>
>
{
};

Decoder will try to decode Case2 until the end of Msg, but not end of MySeq with specified length. Probably, it happens because the loop

now padding is only supported in container but should be available for single IE with length (e.g. like in RTCP User-Id).

padding also s.b. defined via meta-info

to avoid complicated constructions in the code like:

auto& teps = m_data.ref<udsp::smf::qos_info>().ref<udsp::smf::traffic_endpoints>();
    for (auto& tep : teps.get())
    {
        if (!tep.get<udsp::smf::valid>().get())
        {
            udsp::smf::traffic_endpoint& nc_tep = const_cast<udsp::smf::traffic_endpoint&>(tep);
            nc_tep.ref<udsp::smf::local_f_teid>().clear();
            nc_tep.ref<udsp::smf::network_instance>().clear();
            return &nc_tep;
        }
    }

should be like std::array w/o ability to erase, etc.

could be useful in some cases:

some ie1, ie2;
...
if (ie1 == ie2)

med::placeholder::_length lacks of the clear() method.

possible to use type-erasure at some common functions to reduce code bloat

Test:

#include "ut.hpp"
#include "ut_proto.hpp"

struct o_params : med::sequence<
        O<C<0x01>, FLD_UC>,
        O<C<0x02>, FLD_U8>,
        O<C<0x03>, FLD_U16>>
{
};

struct m_params : med::sequence<
        M<FLD_UC>,
        M<FLD_U8>,
        M<FLD_U16>>
{
};

struct params : med::sequence<
        M<FLD_U8>,
        O<C<0x04>, o_params>,
        O<C<0x05>, m_params>>
{

};


TEST(sequence, optional)
{
    params p1;

    p1.ref<FLD_U8>().set(0);

    p1.ref<o_params>().ref<FLD_U8>().set(1);

    p1.ref<m_params>().ref<FLD_UC>().set(2);
    p1.ref<m_params>().ref<FLD_U8>().set(3);
    p1.ref<m_params>().ref<FLD_U16>().set(4);

    uint8_t e_buffer[64];
    med::encoder_context e_ctx{e_buffer};
    encode(med::octet_encoder{e_ctx}, p1);

    uint8_t d_buffer[64];
    med::allocator d_alloc{d_buffer};
    med::decoder_context<med::allocator> d_ctx{e_ctx.buffer().get_start(), e_ctx.buffer().get_offset(), &d_alloc};
    params p2;
    med::decode(med::octet_decoder{d_ctx}, p2);

    EXPECT_NE(p2.get<m_params>(), nullptr);
}

Adding length to o_params fixes the problem (O<C<0x04>, L, o_params>) .

clear in choice only resets the header w/o clearing currently selected value

seems to be possible to factor out loops in one template, put ies in typelist and iterate with functor instead of multiple crafted loops.

perhaps it's worth to have the same interface as std::pmr::memory_resource to allow use of any external allocators

may need to operate on items in multi_field.
at least erase is needed so far. anything more?