kausalflow/lightkone / components/bibtex-import/versions/1/symbolon.toml · raw

[contract]
id = "lightkone/bibtex-import"
version = 1
title = "Read BibTeX into lightkone paper drafts"
description = """
Reads BibTeX text, as exported by Paperpile, Zotero, JabRef or Google
Scholar, into paper drafts the way lightkone's importer does. It never fails:
malformed parts are skipped. Whitespace below means JavaScript's \\s: tab,
LF, VT, FF, CR, space, U+00A0, U+1680, U+2000-U+200A, U+2028, U+2029,
U+202F, U+205F, U+3000 and U+FEFF (and nothing else: not U+0085 or
U+001C-U+001F). Trimming, collapsing and splitting all use this set.

READING ENTRIES. Scan for "@". Read the entry type (characters up to
whitespace or one of , = { } ( ) " #), lower-cased, then skip whitespace. If
the next character is not "{" or "(", skip that "@". The entry closes with
"}" or ")" to match.
  @comment: skip only the "@"; scanning continues inside the comment.
  @preamble: read one value and ignore it.
  @string{name = value}: define a macro (name lower-cased).
  Any other type: the key is everything up to the first "," or the closing
  character outside braces, trimmed ("Erd\\H{o}s1959" is one key). Then fields "name = value" separated by commas. A
  field name is read like a type and lower-cased; a field without a name or
  "=" is skipped up to the next "," or the entry's closing character, whichever
comes first; braced and quoted values in it are skipped whole, read as VALUES
reads them. Reaching the closing character this way ends the entry. A later field
  with the same name replaces the earlier one. An entry with no key and no
  fields is dropped, and so is any entry whose title is missing or empty.
VALUES are parts joined by "#": {braced} text (nested braces, a backslash
escapes the next character, braces are kept inside), "quoted" text (a quote
inside braces does not end it), digits, or a bare word, which is replaced by
a macro of that name (case-insensitive) or kept as written. Month macros
jan..dec mean January..December. Unterminated values run to the end of the
text. Whitespace runs in a value collapse to one space, and the value is
trimmed.

PLAIN TEXT. plain(v) applies these steps in order, each replacing every
non-overlapping match from left to right:
  1. optional "{", then backslash and one of " ' ` ^ ~ = .  (or one of
     c v u H when followed by whitespace or "{"), optional whitespace,
     optional "{", one ASCII letter or \\i or \\j (the dotless letters, read as
     i and j, when not followed by a letter, digit or "_"), optional "}",
     optional "}"  ->  the
     letter plus the combining mark (U+0308, U+0301, U+0300, U+0302, U+0303,
     U+0304, U+0307, U+0327, U+030C, U+0306, U+030B in that order), NFC.
  2. {\\ss} {\\o} {\\O} {\\ae} {\\AE} {\\l} {\\L} {\\i}, or the same without braces
     when the next character is not an ASCII letter, digit or "_" (one
     following space is consumed)  ->  ß ø Ø æ Æ ł Ł ı.
  3. \\emph{x} \\textit{x} \\textbf{x} \\mathrm{x} \\textrm{x}, x without braces  ->  x.
  4. backslash before one of & % $ # _ { }  ->  that character.
  5. remove every { and } not directly after a backslash.
  6. ~ -> space; collapse whitespace runs to one space; trim.

DRAFT FIELDS from an entry's fields (each one used is removed from
"fields" unless stated):
  key, type   from the entry.
  title       plain(title).
  authors     author, or editor if there is no author field. Split on
              whitespace-"and"-whitespace (any case). For each part: plain
              it and drop it if empty; keep it as is if it contains a comma,
              is "others" (any case) or starts with "{" and ends with "}";
              otherwise move the last space-separated word to the front:
              "Ada King Lovelace" -> "Lovelace, Ada King".
  year        the first run of 4 ASCII digits in year, as a number.
  venue       the first of plain(journal), plain(booktitle), plain(publisher)
              that is not empty, so journal = {} does not hide a booktitle;
              absent when all are empty or missing. journal always leaves
              the fields; booktitle leaves them only when journal was empty;
              publisher never does.
  abstract    plain(abstract); empty means absent.
  doi         with a leading http(s)://doi.org/ or dx.doi.org/ removed
              (case-insensitive); empty means absent.
  url, file   as written; an empty value stays "" rather than absent.
  notes       annote, else note, as written.
  arxiv       arxivid; else eprint when archiveprefix or eprinttype is
              "arxiv" (any case); else the id in an arxiv.org/abs|pdf|html/
              url (matched ignoring case, so a V3 version suffix and .PDF are removed
              too); else eprint when it
              starts with NNNN.NNNN. A leading "arXiv:" and a trailing vN are
              removed. When an arxiv id is found, archiveprefix, eprinttype
              and primaryclass are removed; otherwise eprint stays in fields.
  tags        keywords split on ";" or ",", each plained; empty ones and
              ones starting with doi: url: arxiv: isbn: issn: pmid: pmc:
              (any case) or containing http:// or https:// are dropped.
  fields      every remaining field, raw (collapsed) values.
"""
regenerability = "disposable"
license = "Proprietary (kausalflow)"
authors = ["kausalflow/lightkone"]

[entrypoint]
function = "run"

[requires]
network = []
credentials = []
state = []
permissions = []

[spec]
seed = 0
max_examples = 150
timeout_seconds = 300