Fix SymbolMappingMsg hd.length on cross-version encode

Project.toml

@@ -1,7 +1,7 @@
 name = "DatabentoBinaryEncoding"
 uuid = "90689371-c8cb-40d1-831f-18033db90f74"
 authors = ["Tyler Beason <tbeas12@gmail.com>"]
-version = "0.1.0"
+version = "0.1.1"
 
 [deps]
 CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"

src/encode.jl

@@ -791,11 +791,21 @@ function write_record_complex(encoder::DBNEncoder, record)
 
     elseif isa(record, SymbolMappingMsg)
         # Spec-compliant SymbolMappingMsg layout (depends on DBN version):
-        #   v1:  stype_in_symbol[22] | stype_out_symbol[22] | pad(4) | start_ts(8) | end_ts(8)
-        #        (no explicit stype_in/stype_out bytes in v1)
+        #   v1:  stype_in_symbol[22] | stype_out_symbol[22] | pad(4) | start_ts(8) | end_ts(8)   (body 64, total 80, hd.length 20)
         #   v2+: stype_in(1) | stype_in_symbol[71] | stype_out(1) | stype_out_symbol[71] |
-        #        start_ts(8) | end_ts(8)
-        write_record_header(io, record.hd)
+        #        start_ts(8) | end_ts(8)                                                          (body 160, total 176, hd.length 44)
+        #
+        # The record's hd.length may reflect a different on-wire version than
+        # the file we're writing into — the Databento Live gateway emits v1
+        # layout (hd.length = 20) even when the consumer is writing a v3 file.
+        # Re-derive the length from the layout we are about to write so the
+        # resulting record header matches the bytes that follow it.
+        body_bytes = encoder.metadata.version == 1 ? 64 : 160
+        fixed_length = UInt8((16 + body_bytes) ÷ LENGTH_MULTIPLIER)
+        hd = record.hd
+        out_hd = hd.length == fixed_length ? hd :
+            RecordHeader(fixed_length, hd.rtype, hd.publisher_id, hd.instrument_id, hd.ts_event)
+        write_record_header(io, out_hd)
         if encoder.metadata.version == 1
             _write_fixed_string(io, record.stype_in_symbol,  22)
             _write_fixed_string(io, record.stype_out_symbol, 22)

test/test_phase8.jl

@@ -247,10 +247,56 @@ using Dates
             @test records[1].stype_in_symbol == "AAPL.NASDAQ"
             @test records[1].stype_out == SType.INSTRUMENT_ID
             @test records[1].stype_out_symbol == "12345"
-            
+
             rm(test_file, force=true)
         end
-
+
+        @testset "SymbolMappingMsg cross-version (v1 wire → v3 file)" begin
+            # Live Databento gateway emits SymbolMappingMsg in v1 wire layout
+            # (hd.length=20, 80 bytes total) even when the consumer writes a
+            # v3 file. The encoder must re-derive the header length from the
+            # v2+ layout it is actually about to serialize (hd.length=44,
+            # 176 bytes total) so that the resulting record header matches
+            # the body and the file remains parseable.
+            v1_wire_length = UInt8(20)
+            mapping = SymbolMappingMsg(
+                RecordHeader(v1_wire_length, RType.SYMBOL_MAPPING_MSG,
+                             UInt16(0), UInt32(1191182337), 1779203015494476543),
+                SType.INSTRUMENT_ID,
+                "SPX.OPT",
+                SType.INSTRUMENT_ID,
+                "SPX   271217C02800000",
+                Int64(-1),
+                Int64(-1),
+            )
+
+            # Pair with a follow-up record so we exercise the stream offset:
+            # if hd.length on the SymbolMappingMsg is wrong, the decoder
+            # mis-positions for the next read and the file is unparseable.
+            # TradeMsg on-wire: 16-byte header + 32-byte body = 48 bytes, hd.length = 12
+            trade = TradeMsg(
+                RecordHeader(UInt8(12), RType.MBP_0_MSG, UInt16(0), UInt32(1191182337),
+                             1779203015494476600),
+                Int64(100_000_000_000), UInt32(10),
+                Action.TRADE, Side.BID, 0x00, 0x00,
+                Int64(1779203015494476600), Int32(0), UInt32(1),
+            )
+
+            write_dbn(test_file, metadata, [mapping, trade])
+            records = read_dbn(test_file)
+
+            @test length(records) == 2
+            @test records[1] isa SymbolMappingMsg
+            @test records[1].stype_in_symbol  == "SPX.OPT"
+            @test records[1].stype_out_symbol == "SPX   271217C02800000"
+            @test records[1].hd.instrument_id == UInt32(1191182337)
+            @test records[1].hd.length        == UInt8(44)   # rewritten to v3 layout
+            @test records[2] isa TradeMsg
+            @test records[2].hd.instrument_id == UInt32(1191182337)
+
+            rm(test_file, force=true)
+        end
+
         @testset "SystemMsg Write/Read" begin
             # Create a SystemMsg. hd.length is in 4-byte units.
             msg_text = "Market open notification"