Skip to content

Latest commit

 

History

History

formats

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.md

Format library

Ready-made Iris 0.1 reader specs for six benchmark formats from Gecode's examples. This is the format library in PLAN.md §8.9, part of milestone M2; its layout follows §5.

Format Samples
DIMACS CNF Three variables; an out-of-range literal.
TSPLIB Five-city EXPLICIT and EUC_2D files; missing weights.
TSPTW Three nodes; a reversed time window.
Job shop Three jobs and machines in OR-Library and JSPLIB layouts; a repeated machine.
Sudoku A 4×4 grid with 2×2 boxes; incompatible box dimensions.
Nonogram A 5×5 X; a clue wider than the board.

Each <format>/<format>.iris contains instance_spec from the corresponding Gecode example at commit 599f977b7855dba2a1a5f67db1892722372067d3, formatted with iris fmt according to the formatting guide. Formatting changes only whitespace and line breaks. All six were extracted from C++ raw string literals without escape decoding; extraction was checked against bytes emitted by compiling the original literal. The Gecode notices accompany these specs.

Samples and cases

Samples live directly in cases/<name>/input.txt, alongside their expected.json or error.json and optional params.json. There is no second sample or expectation directory. Each case follows the conformance contract. Its spec.iris is a relative file symlink to ../../<format>.iris. This keeps a single reader spec per format, so case copies cannot drift. The runner supports file symlinks and discovers these cases from formats/.

All samples were written by hand for this repository and are freely redistributable under its MIT license. No benchmark instances were copied. Each format README cites its format documentation and describes the subset read by the pinned spec.

Expected results come from running the pinned gecode-instance tool on every sample. Successful dzn output was compared byte for byte with the adapter's dzn output before capturing canonical JSON through the adapter. Invalid samples' reference diagnostics were checked against the adapter's canonical errors, including locations and record paths. These cases need no known-failure annotations.

Run the cases

From the repository root:

eval "$(scripts/gecode-reference.sh --print-env)"
adapter=$(conformance/adapters/gecode/build.sh)
(cd rust && cargo run --locked -p gecode-iris-conformance -- run --driver "\"$adapter\"" \
  --driver-id gecode-6.5.0-599f977b7855 ../formats)

For a direct read with the reference tool:

"$GECODE_REFERENCE_INSTANCE_TOOL" -spec formats/tsplib/tsplib.iris -dzn \
  formats/tsplib/cases/euc-2d-five/input.txt

Changes to the meaning of these specs belong in a later language version. Format limitations and awkward constructions feed into Iris 1.0 ideas.