#include <array>
#include <cstdint>
#include <expected>
#include <functional>
#include <ostream>
#include <string>
#include <vector>
#include "node/process_data_ring.h"

Include dependency graph for process_data_dump.h:

This graph shows which files directly or indirectly include this file:

Classes
struct	mm::node::DumpPdoEntry
	One PDO-mapped object as recorded in the dump header. More...

struct	mm::node::DumpDevice
	One device's identity and the PDO objects it contributes to the image. More...

struct	mm::node::DumpHeader
	The embedded process-image header: everything needed to decode the rows offline. More...

Namespaces
namespace	mm

namespace	mm::node

Typedefs
using	mm::node::DumpRecordReader = std::function< bool(uint64_t seq, ProcessDataRing::Record &out)>
	Reads the record for `seq` into `out`; returns `false` if it is no longer available (lapped or a torn read). Wraps `ProcessDataRing::readRecord`.

Functions
std::expected< uint64_t, std::string >	mm::node::writeProcessDataDump (std::ostream &out, const DumpHeader &header, uint64_t startSeq, uint64_t endSeq, const DumpRecordReader &read)
	Serialises the frozen span [startSeq, endSeq) to `out` as a `.mmpd` dump.

Variables
constexpr std::array< char, 4 >	mm::node::kDumpMagic = {'M', 'M', 'P', 'D'}
	Magic at the very start of every dump file.

constexpr uint16_t	mm::node::kDumpFormatVersion = 1
	Current dump format version. A reader rejects an unknown major shape; bump on layout changes so old files stay identifiable.

constexpr std::streamoff	mm::node::kDumpRowCountOffset = 16
	Byte offset of the `rowCount` field within the fixed prefix. The writer streams rows first, then seeks here to patch the final count — so the stream must be seekable.

Detailed Description

The .mmpd binary dump of the process-data recorder: exactly what the recorder ring holds (full raw inputs + outputs for every cycle in a frozen span) plus the process image embedded as a header, so a run decodes fully offline — no running Motion Master and no live bus.

Layout (all multi-byte integers little-endian):

File prefix (fixed 36 bytes — rowCount sits at a fixed offset so it can be patched after the rows are streamed): 0 char magic[4] = 'M','M','P','D' 4 u16 formatVersion = kDumpFormatVersion 6 u16 flags = 0 (reserved) 8 u64 startSequence (sequence of the first row; oldest cycle in the span) 16 u64 rowCount (number of rows actually written) <- patched at the end 24 u32 inputBytes (per-row input region size) 28 u32 outputBytes (per-row output region size) 32 u32 deviceCount Device table (deviceCount entries): u16 slavePosition u32 vendorId · u32 productCode · u32 revisionNumber · u32 serialNumber str name (u16 length + UTF-8 bytes) u32 entryCount per PDO entry: u16 index · u8 subindex · u8 direction (1 = output/RxPDO, 0 = input/TxPDO) u16 dataType (0 = unknown) · u16 bitLength · u32 bitOffset str name (empty when the object dictionary is not enumerated) Rows (rowCount entries, fixed stride 16 + inputBytes + outputBytes): u64 sequence · u64 timestampNs · u8 inputs[inputBytes] · u8 outputs[outputBytes]

A row's input/output regions are padded with zeros (or truncated) to the header sizes, so the stride is fixed and a reader can index any cycle directly. The timestamp keeps full epoch-ns precision (unlike the live monitoring row, which reduces it to microseconds for JavaScript).

Classes

Namespaces

Typedefs

Functions

Variables

Detailed Description