Skip to contents

Changelog

Source: NEWS.md

anystyleR (development version)

Changes compared with anystyleR 0.0.999. There are breaking changes in the arguments and in the columns of the results.

Renaming and infrastructure

  • The package is renamed from Ranystyle to anystyleR. Its repository and website move to Codeberg (https://agoutsmedt.codeberg.page/anystyler/).
  • The package now needs R >= 4.1 and no longer imports magrittr, purrr and tidyr (it imports stringi). It has tests (testthat), run with or without anystyle installed.
  • The vignette is replaced by two articles on the website: a workflow guide, From documents to references, and How good is the parsing?, which measures anystyle and the cleaning rules against Scopus references.

Running anystyle

  • New anystyle_path() and check_anystyle(): the anystyle executable is looked for in the option anystyleR.anystyle, the environment variable ANYSTYLE_PATH, the PATH and the user gem directories. Functions fail early with an informative message when it is missing.
  • anystyle is now run without a shell: paths with spaces or quotes work. Its errors, including those it reports with a success exit status, stop the function with anystyle’s message. Commands are no longer printed.
  • Nothing is written in the working directory any more: temporary files go to a temporary directory and are removed.
  • install_anystyle() installs the gems in the user’s directory by default (user_install = TRUE, no administrator rights needed), reports where the executable is, and checks for pdftotext.
  • Messages, warnings and errors use cli.

find_ref() and parse_ref()

  • Arguments renamed: path becomes output_dir (NULL, the default, returns the output instead of writing files), output_format becomes format (default "json", validated with rlang::arg_match() instead of a warning), no_layout becomes layout (default TRUE).
  • New arguments finder_model and parser_model to use custom anystyle models, and solo in find_ref() to keep references outside reference sections.
  • find_ref() accepts text files as well as PDFs, and the "csl", "txt" and "ttx" formats ("ttx" shows the label given to each line, to see why a bibliography was missed). It shows a progress bar.
  • The output is a named character vector (one element per document) instead of a list or a single element depending on the number of documents.
  • parse_ref() takes input as files only when all elements are existing files (previously, any element without .txt made the whole input references). Line breaks inside a reference are replaced by spaces and empty references dropped, so that the output matches the input.
  • Existing output files are checked before anything runs.

find_ref_to_df() and parse_ref_to_df()

  • All the references of a call are parsed in one run of anystyle, in chunks of 1,000, instead of one run per document: much faster on many documents. A reference that makes anystyle fail no longer stops the run: it is isolated, returned empty and flagged anystyle_failed.
  • full_ref is now aligned with the parsed fields by construction (previously, the two were joined by position and could be shifted).
  • New doc argument in parse_ref_to_df(): the document each reference comes from, to number references within documents.
  • New column ref_order (position of the reference in its document). id_doc is the position of the document in input.
  • Argument clean_ref renamed clean; no_layout replaced by layout; new arguments solo, finder_model and parser_model.
  • With clean = FALSE, the result has a stable structure: one list-column per anystyle field (a tibble of name parts for names), instead of columns unnested or not depending on the data.
  • A ditto mark opening a document no longer takes the authors of the last reference of the previous document.
  • Before parsing, a space is added after a closing quotation mark glued to the next word and after an initial glued to a family name (“H.Genberg”); full_ref keeps the original text.

clean_ref()

clean_ref() is rewritten. It returns flat columns (first_author, n_authors, authors, et_al, year, year_suffix, date, title, container_title, volume, issue, number, pages, first_page, last_page, editors, doi…; field names use _ instead of -), and a flag column listing the rules that changed each reference. One-value fields keep the first value anystyle found; further titles and dates go to other_title and other_date (separated by " | " instead of line breaks). It is vectorised and much faster. New rules, each checked against Scopus fields or central bank bibliographies:

  • Names written “Family I.” without comma, read by anystyle with the initials as family name, are swapped.
  • Given names in capitals split into initials by anystyle (“D.O.N.A.L.D.”) are restored; acronyms alone (“ECB”) become institutions.
  • Institutions read as persons (“Bank of England”) or as titles (“Committee on the Global Financial System (2012)”) become one literal name.
  • Ditto marks (“_.”, “–”) are filled with the authors of the previous reference of the same document; “et al.” becomes et_al.
  • The first editor stands for the author of an edited book without author in first_author.
  • Years are read from the reference when anystyle finds no date (in brackets, or the only year of the reference); the letter of “1985a” is kept in year_suffix; a title made only of a year is dropped.
  • Pages are split into first_page and last_page, abbreviated ranges (“1145-64”) expanded.
  • “Vol.” and “No.” are removed (no longer the “No” of “November”), anystyle’s “ue 2” (from “Issue 2”) is repaired, glued volume and issue (“47 2”) split, roman volumes converted.
  • Working papers and other report series are typed report, with their number in number; speeches, remarks and testimonies are typed speech.
  • The DOI is taken from the doi field, the URL or the reference, without resolver and trailing punctuation.
  • Quotation marks and unbalanced brackets around fields are removed (previously, all leading and trailing punctuation of every column, including closing brackets and question marks).
  • Lines that are probably not references are flagged not_a_reference.

Flattened bibliographies

  • New split_ref() cuts a block of text holding a whole bibliography without line breaks (OCR, databases) into references, with rules on where a reference starts. On Scopus bibliographies joined into one block, 92% of cuts are right and 78% of true boundaries are found.
  • parse_ref_to_df() gains split = TRUE to split blocks before parsing.

Word bibliographies

  • find_ref_to_word_bibliography() and parse_ref_to_word_bibliography() call pandoc directly instead of rendering an R Markdown document, and share their code.
  • Argument csl_path renamed csl; no_layout replaced by layout; new argument overwrite.
  • The default output is in the working directory instead of the home directory, and only .docx is accepted. The path is returned invisibly.

anystyleR 0.0.999