Datasets:
File size: 4,985 Bytes
c4574c1 9d81d70 c4574c1 9d81d70 c4574c1 9d81d70 c4574c1 9d81d70 c4574c1 9d81d70 c4574c1 9d81d70 c4574c1 9d81d70 c4574c1 9d81d70 c4574c1 9d81d70 c4574c1 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 | ---
license: cc-by-4.0
pretty_name: py-code.org PyPI code usage
language:
- en
tags:
- python
- pypi
- code
- dependencies
size_categories:
- 10M<n<100M
configs:
- config_name: refs
default: true
data_files: refs.parquet
- config_name: projects
data_files: projects.parquet
- config_name: own_modules
data_files: own_modules.parquet
- config_name: pth_files
data_files: pth_files.parquet
---
# py-code.org: which PyPI packages use which names
For the latest release of 354,763 PyPI projects (every project released between 13 Nov 2025 and
13 Nov 2026, plus the most downloaded projects even when their latest release is older), these
tables list every module, function and class the code imports or reaches by attribute,
such as `requests.Session` or `numpy.linalg.norm`. This is the data behind the search on
[py-code.org](https://py-code.org). The scanner that built it is
[pypi-data/py-code](https://github.com/pypi-data/py-code) (MIT).
The tables hold names, counts and file paths. They hold no source code.
## Tables
### `refs` (39.9 million rows, 792 MB)
One row per project and name. Sorted by `name`, then `project`.
| Column | Type | Meaning |
|---|---|---|
| `project` | string | PyPI project name |
| `name` | string | Full dotted name, e.g. `requests.Session` |
| `module` | string | Top-level module of `name`, e.g. `requests` |
| `files` | int | Number of `.py` files in the project that use `name` |
| `locations` | list of string | Up to 3 places where it is used, as `path:line` |
A name is recorded with every prefix: `os.path.join` also adds `os` and `os.path`. Attribute
chains are followed up to 4 levels after the imported name. Names from the project's own
top-level modules are left out, so a package does not count as a user of itself.
### `projects` (354,763 rows, 22 MB)
| Column | Type | Meaning |
|---|---|---|
| `project` | string | PyPI project name |
| `version` | string | Latest release at scan time |
| `upload_time` | timestamp | Upload time of the scanned file (UTC) |
| `requires_python` | string | `Requires-Python` of the release, if set |
| `summary` | string | PyPI summary, cut at 300 characters |
| `file` | string | File that was scanned |
| `file_size` | int | Size of that file in bytes |
| `bytes_fetched` | int | Bytes the scanner downloaded to read the `.py` files |
| `py_files` | int | `.py` files in the file |
| `py_failed` | int | `.py` files that do not parse as current Python (mostly Python 2) |
| `downloads_30d` | int | Downloads in the 30 days before the scan, for the top 60,000 projects; 0 for the rest |
### `own_modules` (402,861 rows)
Top-level modules each project provides, from its file layout. Use it to go from an import
name to a project: `PIL` to `pillow`, `yaml` to `PyYAML`.
### `pth_files` (748 rows)
`.pth` files that contain an `import` line. Python runs these at every start-up.
## How it was built
1. List every project whose latest event is newer than the cut-off date (PyPI Simple API and
XML-RPC changelog), plus the 60,000 most downloaded projects.
2. For each one, take the latest release and pick one file: a pure-Python wheel, then a
platform wheel, then the source distribution. Wheels are read with HTTP range requests,
so only the `.py` and `.pth` files are downloaded.
3. Parse each `.py` file with Python's `ast` module and record imports and attribute access
on imported names.
4. Add 30-day download counts from ClickHouse's public PyPI dataset.
12.9 million `.py` files were parsed. The index was last rebuilt on 13 Nov 2026. The scanner's README
has the details and the measured run times.
## Known limits
- Only the latest release of each project. Projects with no release in the window are not
included unless they are among the 60,000 most downloaded.
- Dynamic imports (`importlib.import_module`, `__import__`, `getattr`) are missed.
- Names reached through variables (`x = np; x.foo`) are missed.
- Files that do not parse as current Python are counted in `py_failed` but not indexed.
- Projects whose authors asked to be left out are not included.
## Example queries
Which packages use `requests.Session`, by downloads:
```sql
SELECT r.project, p.downloads_30d
FROM refs r JOIN projects p USING (project)
WHERE r.name = 'requests.Session'
ORDER BY p.downloads_30d DESC
LIMIT 20;
```
Which packages still import `distutils`:
```sql
SELECT count(DISTINCT project) FROM refs WHERE module = 'distutils';
```
With DuckDB, without downloading the files first:
```sql
SELECT count(DISTINCT project)
FROM 'hf://datasets/pypi-data/py-code/refs.parquet'
WHERE name = 'httpx.AsyncClient';
```
## Removal
If you maintain a package and want it left out, email dev@py-code.org from an address listed
on its PyPI page. It is removed at the next update.
## Credit
From 2023 to 2026 py-code.org hosted [Tom Forbes](https://tomforb.es)' PyPI Data project. This
dataset is a new, independent project on the same domain and does not reuse its data.
|