DIRE: Decompiled Identifier Naming Data from 3.2 Million Functions

A 2019 dataset released with the ASE conference paper 'DIRE: A Neural Approach to Decompiled Identifier Naming'. It contains information from 3,195,962 functions decompiled from 164,632 unique binaries generated from C code scraped from GitHub. The dataset was created by Jeremy Lacomis of Carnegie Mellon University and is partitioned into 16 archives by binary hash.

Use Cases

• Training neural models for identifier name prediction based on decompiled function data.
• Benchmarking decompilation and code understanding tools using the large corpus of functions.
• Studying the relationship between source code patterns and decompiled output mentioned in the description.

Strengths

• Large scale with 3,195,962 functions from 164,632 binaries.
• Clear provenance linked to a peer-reviewed ASE 2019 conference paper.
• Data is partitioned and provided in multiple compressed formats (GZIP and BZIP2) for practicality.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Data may reflect source bias inherent to being scraped from GitHub.

Provenance

Source

Carnegie Mellon University

Collection Method

Generated by decompiling binaries from C code scraped from GitHub.