Collect the package dependencies a set of Python files declare
Source:R/reproducibility_check_python.R
repro_dependencies_py.RdThe Python sibling of repro_dependencies(): walks each file's import/
from ... import statements via code_library_names() (lang = "Python"),
and additionally reads any requirements.txt/pyproject.toml the
repository supplies for version pins — static analysis of the code alone
cannot see those (a script that does import numpy says nothing about
which version), so when a manifest is present its pin is attached to the
matching imported package instead of being reported separately. A top-
level package name is tagged base (the Python standard library; a run
never needs to pip install these) or pypi — unlike R's CRAN/
Bioconductor/GitHub/URL split, this does not attempt to tell a PyPI
install apart from a pip install git+.../local-path install named in a
requirements.txt line: both are reported source = "pypi", with ref
carrying the raw requirement line verbatim when it is not a bare
package==version pin, so the install step can still attempt it.
Arguments
- code_text
the code text for a single file (character vector), OR a list of such vectors (one per file) to pool across files
- manifest_text
optional character vector: the text of a
requirements.txt(one requirement per line) or the[project]/[tool.poetry]dependency lines of apyproject.toml, if the repository has one.NULL(default) when none was found — dependencies are then name-only, exactly like a code-only R scan.
Value
a data frame with columns package, source (pypi or base),
ref (the manifest's own pin/requirement text for that package, or NA
when the package was seen only via import), and base (logical, TRUE
for a standard-library module). One row per distinct package. Empty
frame (same columns) when none are found.
Examples
code_text <- c("import numpy as np", "from sklearn.linear_model import LinearRegression")
repro_dependencies_py(code_text)
#> package source ref base
#> 1 numpy pypi <NA> FALSE
#> 2 scikit-learn pypi <NA> FALSE