PLDI 2027 ARTIFACT - AutoDiDact: Input Reduction over Inferred Structure
Authors/Creators
Description
Replication package for the PLDI 2027 submission "AutoDiDact: Input Reduction over Inferred Structure."
The package contains the implementation, benchmark inputs and oracles for the paper's 125 selected cases across 18 format families, pinned subject build recipes, experiment runners, saved measurements and reduced outputs, and scripts that validate the results and recompute the paper’s numerical summaries. The experiments compare AutoDiDact with byte-level drdd, grammar-based HDD, and C-Reduce.
Two archives are provided:
- a source package (autodidact-replication-1.0.0-source.tar.gz), and
- a prebuilt container image (autodidact-1.0.0-image.tar.gz): the source package and all required subject runtimes, reference lexers, and comparison tools.
Start with README.md in the source package. Validating and summarizing saved results requires only Python 3.11 or newer and its standard library. The README guides readers through a small live experiment and full reruns. Container instructions and smoke checks are in image/README.md. The image targets Linux x86-64 and has been validated with rootless Podman.
Benchmark subjects derive from public research artifacts, datasets, and bug trackers; they are the work of their respective authors. Per-case provenance and licensing notes are included.