Repository navigation
MATERIAL (30 records): per-map block layout needs a corpus file with a non-default material #49
Description
#47 withdrew MATERIAL's wrong field list (it decoded ambient_color_method = 542113793 on the very first field) and kept only proven content: names/description from the string stream plus the measured data budget (1264 bits R2004, 1284 R2010/R2013, 516/516/1028 R2018). Twelve BD slots decode to exactly 1/48 at data bits 46, 120, 250, 324, 510, 584, 706, 780, 900, 974, 1096, 1170 - six pairs 74 bits apart, one per texture map - but inter-pair gaps (204/260/196/194/196) pin no repeating block. Every file has only ByLayer/ByBlock/Global, bit-identical apart from the name, so the joint-boundary search has nothing to bite on. Needs a redistributable DWG with a user-defined material (a synthetic one from examples/build_fixtures.rs once the writer can emit MATERIAL, or a permissively-licensed sample). Tracked as Unhandled until then.
Activity
DrunkOnJava commentedon Oct 21, 2026
A MATERIAL joint-boundary probe,
examples/probe_material.rs, was written in a local worktree (feat/xrecord-material) but never committed. It is preserved atrefs/archive/2026-10-21/xrecord-material-probes(commit d6ac794, based on 44cc130; fetch it withgit fetch origin refs/archive/2026-10-21/xrecord-material-probes). It has not been rebuilt against currentmain. It runs a candidate field spec against every MATERIAL record and reports each record's delta to its boundary. A spec only counts when every record of every file lands at delta 0.DrunkOnJava commentedon Oct 26, 2026
MATERIAL survey, 2026-10-26,
examples/probe_material.rs(#222) on the 19-file corpusThe corpus holds 65 MATERIAL records, not 30, and eight of them are not defaults. All in
sample_AC1032.dwg:0x96/0x97/0x980xDAA0xDB60xDC00xDCA0xDD40xDDE0xDE80xDF2The other 18 files carry only the three defaults, with the budgets the module documents, now measured on every band: R14, R2000 and R2004 all 1340 / 1340 / 1332 with the two
TVs inline (1264 afterTV,TVon all three), R2007, R2010 and R2013 all 1284.So the joint-boundary search does have something to bite on, on R2018 only: eight records whose budgets differ by map count (the ones with texture paths are the long ones), which should separate a per-map block from the fixed fields.
sample_AC1032.dwghas no known source (#96), so it serves as measurement evidence, not as a redistributable fixture; the request for a redistributable file with a user material still stands for CI.Two corrections for whoever picks this up: the module table in
src/objects/acad_material.rslists 3 records forsample_AC1032.dwg(there are 11), and this issue's title count (30) is the R2004, R2010 and R2013 files plus the three R2018 defaults: the R14, R2000 and R2007 walks (#72) add 27, and the eight library materials the other 8.Reproduce:
cargo run --release --example probe_material -- ../../samples/*.dwg [spec];MAT_BD=1reproduces the twelve1/48slots at the offsets above onarc_2010.dwg.feat(objects): probe XRECORD on every release band and MATERIAL's dat…
feat(objects): decode MATERIAL from AutoCAD's own DWG and DXF output (#…
feat(objects): decode five small classes from AutoCAD's own DWG and D…
feat(objects): decode the dynamic-block evaluation objects from AutoC…
feat(objects): decode DIMASSOC from AutoCAD's own DWG and DXF output (#…
feat(objects): decode ACSH_HISTORY_CLASS and ACSH_BOX_CLASS from Auto…
fix(entities): decode PDFUNDERLAY on every release and check it again…
feat(entities): decode SUN from AutoCAD's own DWG and DXF output (#152)…
feat(objects): decode DATATABLE from AutoCAD's own DWG and DXF output (…
feat(objects): decode ACDB_TEXTOBJECTCONTEXTDATA_CLASS from AutoCAD's…
feat(objects): decode ACAD_EVALUATION_GRAPH from AutoCAD's own DWG an…
feat(objects): decode ACSH_EXTRUSION_CLASS from AutoCAD's own DWG and…