polars-refkit 0.0.2

Polars expression plugin for refkit citation and BibTeX workflows

polars-refkit

polars-refkit adds Rust-backed Polars expressions for BibTeX and BibLaTeX columns. It imports as polars_refkit.

Install

pip install polars-refkit

polars-refkit requires Polars >=1.29 and supports CPython 3.11 through 3.14. It also publishes a Python 3.14 PyEmscripten wheel for Pyodide.

Render And Inspect Rows

import polars as pl
import polars_refkit as prk

df = pl.DataFrame(
    {
        "bibtex": [
            """
@article{doe2024, title={Fast Citations}, year={2024}}
@book{roe2022, title={Batch References}, year={2022}}
""",
        ],
        "key": ["doe2024"],
        "keys": [["doe2024", "roe2022"]],
    }
)

out = df.select(
    citation=pl.col("bibtex").refkit.cite(pl.col("key")),
    literal_citation=pl.col("bibtex").refkit.cite(pl.lit("doe2024")),
    each_citation=pl.col("bibtex").refkit.cite_each(pl.col("keys")),
    grouped_citation=pl.col("bibtex").refkit.cite_group(pl.col("keys")),
    bibliography=pl.col("bibtex").refkit.full_bibliography_html(),
    count=pl.col("bibtex").refkit.entry_count(),
    keys=pl.col("bibtex").refkit.keys(),
    entries=pl.col("bibtex").refkit.entries(),
)

Each row is one BibTeX or BibLaTeX source. The expressions run inside eager DataFrame.select and lazy LazyFrame.select(...).collect() plans. Row-level parse failures return null for value expressions and a diagnostic list from diagnostics or parse_report.

recovery="error" uses strict parsing. In a Polars expression, a strict row parse failure returns null for value expressions, False from can_parse, and a failed parse_report instead of aborting the query. recovery="report" keeps recoverable entries in that row and preserves parser diagnostics.

String arguments name columns. Use pl.lit(...) for literal BibTeX sources or citation keys:

keys = pl.DataFrame({"key": ["doe2024"]})
out = keys.select(citation=pl.lit(df["bibtex"][0]).refkit.cite(pl.col("key")))

Use cite_each when one row has an ordered list of citation keys and each key should render as a separate citation:

batch = pl.DataFrame({"keys": [["doe2024", "roe2022"]]})
out = batch.select(citations=pl.lit(df["bibtex"][0]).refkit.cite_each(pl.col("keys")))

Use cite_group when one row has an ordered list of citation keys and the list should render as one grouped citation:

batch = pl.DataFrame({"keys": [["doe2024", "roe2022"]]})
out = batch.select(citation=pl.lit(df["bibtex"][0]).refkit.cite_group(pl.col("keys")))

Top-level functions and namespace methods use stable default output names, so multiple expressions over the same bibtex column can be selected without manual aliases. Use alias or named select expressions when a result column needs a different name.

Capabilities

Capability	Polars surface
Read normalized bibliography data	entry_count, can_parse, has_diagnostics, keys, entries, parse_report, diagnostics
Render citations	cite, cite_html, cite_rendered, cite_each, cite_group, and their HTML or struct variants
Render bibliographies	full_bibliography_text, full_bibliography_html, full_bibliography_rendered
Inspect entries	keys, entries, to_hayagriva_json
Process dataframe columns	eager DataFrame.select and lazy LazyFrame.select(...).collect()
Use expression namespace	pl.Expr.refkit methods with the same capability set

Expressions

Function	Return	Behavior
cite(bibtex_col, key_col, style="apa", locale="en-US", recovery="error")	String	Renders one citation as text. Missing keys and row parse failures return null.
cite_html(bibtex_col, key_col, style="apa", locale="en-US", recovery="error")	String	Renders one citation as escaped HTML.
cite_rendered(bibtex_col, key_col, style="apa", locale="en-US", recovery="error")	Struct[text, html]	Renders one citation with both text and HTML fields.
cite_each(bibtex_col, keys_col, style="apa", locale="en-US", recovery="error")	List[String]	Renders each key in a List[String] column as a separate citation. Missing keys and row parse failures return null for the row.
cite_each_html(bibtex_col, keys_col, style="apa", locale="en-US", recovery="error")	List[String]	Renders each key as separate citation HTML.
cite_each_rendered(bibtex_col, keys_col, style="apa", locale="en-US", recovery="error")	List[Struct[text, html]]	Renders each key as a separate citation struct.
cite_group(bibtex_col, keys_col, style="apa", locale="en-US", recovery="error")	String	Renders one grouped citation from a List[String] key column.
cite_group_html(bibtex_col, keys_col, style="apa", locale="en-US", recovery="error")	String	Renders one grouped citation as HTML.
cite_group_rendered(bibtex_col, keys_col, style="apa", locale="en-US", recovery="error")	Struct[text, html]	Renders one grouped citation with both text and HTML fields.
full_bibliography_html(bibtex_col, style="apa", locale="en-US", recovery="error")	String	Renders all entries in the row as an HTML bibliography. Row parse failures return null.
full_bibliography_text(bibtex_col, style="apa", locale="en-US", recovery="error")	String	Renders all entries in the row as plain text.
full_bibliography_rendered(bibtex_col, style="apa", locale="en-US", recovery="error")	Struct[text, html]	Renders all entries in the row with both bibliography formats.
entry_count(bibtex_col, recovery="error")	UInt32	Counts normalized entries in each BibTeX string.
can_parse(bibtex_col, recovery="error")	Boolean	Returns whether the row can produce a normalized library.
has_diagnostics(bibtex_col, recovery="error")	Boolean	Returns whether parsing produced diagnostics.
keys(bibtex_col, recovery="error")	List[String]	Returns normalized entry keys in source order.
entries(bibtex_col, fields=("key", "title", "doi", "volume"), recovery="error")	List[Struct]	Projects normalized entries into Polars-native rows.
parse_report(bibtex_col, recovery="error")	Struct[ok, entry_count, keys, diagnostics]	Parses each row once and returns a summary struct.
diagnostics(bibtex_col, recovery="error")	List[String]	Returns an empty list for valid rows and parse messages for invalid rows.
to_hayagriva_json(bibtex_col, recovery="error")	String	Returns normalized Hayagriva entry JSON with id and key fields.

Expression Namespace

The same operations are available from pl.Expr.refkit.

out = df.select(
    keys=pl.col("bibtex").refkit.keys(),
    count=pl.col("bibtex").refkit.entry_count(),
    citation=pl.col("bibtex").refkit.cite(pl.col("key")),
    each_citation=pl.col("bibtex").refkit.cite_each(pl.col("keys")),
    grouped_citation=pl.col("bibtex").refkit.cite_group(pl.col("keys")),
    entries=pl.col("bibtex").refkit.entries(),
    hayagriva_json=pl.col("bibtex").refkit.to_hayagriva_json(),
)

Top-level functions and namespace methods expose one name per capability. They return expressions with names that match the method, such as keys, entry_count, cite, and to_hayagriva_json. Name outputs in select, with_columns, or alias when a call site needs a different column name.

Typed code can cast the namespace when the type checker does not know Polars plugin namespaces:

from typing import cast

namespace = cast(prk.RefkitExprNamespace, pl.col("bibtex").refkit)
out = df.select(citation=namespace.cite(pl.col("key")))

entries returns a list of structs. Explode and unnest it to query entries as rows:

entries = (
    df.select(entries=pl.col("bibtex").refkit.entries())
    .explode("entries")
    .unnest("entries")
)

Scope

Use polars-refkit when BibTeX source lives in a dataframe and the result should stay in a Polars query plan. Use refkit for raw BibTeX repair with comments, preambles, string definitions, failed blocks, ordering, and source spans.

Development

uv sync --all-packages --group dev
(cd packages/polars-refkit && uv run maturin develop)
uv run pytest packages/polars-refkit/tests --no-cov

License

polars-refkit is licensed under the Apache License, Version 2.0, available in LICENSE. See NOTICE for upstream citation and bibliography component acknowledgements.