File size: 4,525 Bytes
f389146
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
---
license: cc-by-4.0
pretty_name: PolyCI
tags:
  - continuous-integration
  - software-engineering
  - mining-software-repositories
  - devops
  - ci-services
size_categories:
  - 100M<n<1B
configs:
  - config_name: CI_SERVICES
    data_files: data/CI_SERVICES.parquet
  - config_name: EXISTENCE_CHECK
    data_files: data/EXISTENCE_CHECK.parquet
  - config_name: YML_STATS
    data_files: data/YML_STATS.parquet
  - config_name: REPO_DETAILS
    data_files: data/REPO_DETAILS.parquet
  - config_name: REPO_METRICS
    data_files: data/REPO_METRICS.parquet
  - config_name: CI_REPO_CONTRIBUTORS
    data_files: data/CI_REPO_CONTRIBUTORS.parquet
  - config_name: COMMITS_PER_CI_FILES
    data_files: data/COMMITS_PER_CI_FILES.parquet
  - config_name: REPO_COMMIT_DETAILS
    data_files: data/REPO_COMMIT_DETAILS/*.parquet
  - config_name: EXPLANATORY_CI_COMMITS
    data_files: data/EXPLANATORY_CI_COMMITS.parquet
---

# PolyCI

PolyCI is a cross-service, cross-language, longitudinal dataset of continuous
integration (CI) use in open-source software. It contains **135,189 repositories**
adopting one or more of **eight CI services** (GitHub Actions, Travis CI, CircleCI,
GitLab CI, AppVeyor, Azure Pipelines, Bitbucket Pipelines, Cirrus CI),  across 
**seven languages** (C, C++, Go, Java, Python, Ruby, Rust), from **2008 to 2025**. 
Every table is keyed by the GitHub `repo_name` (`owner/repo`), so project, service, 
and commit levels join directly.

Each data type below merges its per-language source files into a single parquet
carrying a lowercase `language` column that identifies the source language. The
one exception is `REPO_COMMIT_DETAILS` (167 M rows): it is split into one
parquet per language to keep every file relatively small, and each file keeps 
the same `language` column.

## Tables

| Config (`config_name`) | Grain | What it holds |
|---|---|---|
| `CI_SERVICES` | one row per CI-using repo | which CI services a repo declares, a per-service presence flag, and the service count |
| `EXISTENCE_CHECK` | repo x service | whether each declared CI configuration `still_exists` (basis for abandonment) |
| `YML_STATS` | repo x service | number of workflow/config YAML files and total YAML lines |
| `REPO_DETAILS` | one row per repo | GitHub metadata: stars, forks, size, timestamps, license, topics, description |
| `REPO_METRICS` | one row per repo | commit, pull-request, issue, and contributor counts |
| `CI_REPO_CONTRIBUTORS` | repo x contributor | contributor login, account `type`, and contribution count |
| `COMMITS_PER_CI_FILES` | commit that edits a CI file | commit and parent SHA, message, author/committer, dates for every commit touching a CI configuration file |
| `REPO_COMMIT_DETAILS` | commit | the full commit history of each repo (SHA, parent SHA, message, author/committer, dates) |
| `EXPLANATORY_CI_COMMITS` | commit | CI commit messages annotated with `multi_ci` and `abandoned` flags |

## Loading

```python
from datasets import load_dataset

ci = load_dataset("taher-ghaleb/PolyCI", "CI_SERVICES", split="train")
```

Or read a single table directly with pandas / DuckDB:

```python
import duckdb
duckdb.sql("SELECT language, count(*) FROM 'data/CI_SERVICES.parquet' GROUP BY 1")
```

## Scope and known boundaries

- Detection is path-based: a table records a *declared* CI configuration, not
  executed builds. A configuration still present is not proof it still runs, so
  the abandonment signal is a lower bound.
- `COMMITS_PER_CI_FILES` and `REPO_COMMIT_DETAILS` give commit and parent SHAs
  and messages, but **not** the list of files changed per commit, and there is
  **no pull-request identifier**. Studies that need same-commit file co-change or
  PR-level grouping resolve the provided SHAs to file lists (GitHub API or a local
  clone) and map commits to PRs themselves.
- The dataset excludes private repositories, forks, and projects below five
  stars. It generalizes to visible, active open-source development, not to GitHub
  as a whole (see Kalliamvakou et al., 2016, on the perils of mining GitHub).
- Commit records include author and committer names and emails as they appear in
  public git history. Do not use these fields to identify or contact individuals.


## Citation

```bibtex
@misc{polyci2027ghaleb,
  title  = {PolyCI: A Large-Scale Longitudinal Multi-Service Dataset of Continuous Integration Adoption},
  author = {Ghaleb, Taher A.},
  year   = {2026},
  howpublished = {\url{https://huggingface.co/datasets/taher-ghaleb/PolyCI}}
}
```