The following changes are not yet released, but are code complete:

Features:

• None

Changes:

• None

Fixes:

• Strengthens error handling during the loading of the cached Hyperscan database. This ensures that an invalid cache triggers a rebuild.

2.6.4 - 2024-06-03

Fixes:

• Bump eyecite to for InvalidError/hyperscan bug

2.6.3 - 2024-04-09

Fixes:

• Addresses compatibility issues identified with the current version of hyperscan(0.7.7).

2.6.2 - 2024-03-19

Fixes:

• Adds missing closing parentheses to the corrected_citation_full method in FullCaseCitation and FullJournalCitation classes.

2.6.1 - 2024-02-29

Fixes:

• Improves reliability of utils.hash_sha256() by providing a default function to handle non-serializable objects.

2.6.0 - 2024-02-07

Features:

• Removes Python 3.8 and 3.9 support
• Adds Python 3.12 support

Changes:

• The hashing and equality behavior of CitationBase objects has changed in subtle ways in order to conform with user intuitions and define previously ill-defined behavior. Most importantly, to compare two eyecite objects going forward, simply use the native Python syntax citation1 == citation2. You can also take the hash of each object and compare that with identical results: hash(citation1) == hash(citation2). This broad change has several more specific implications:

  • Citation objects that are created from text with a normalized/corrected reporter are now treated as equal to objects created from text without a normalized/corrected reporter (e.g., 1 U.S. 1 versus 1 U. S. 1 are now treated as equal)
  • Citation objects that are created from text with a nominative reporter are now treated as equal to objects created from text without a nominative reporter (e.g., 5 U.S. 137 versus 5 U.S. (1 Cranch) 137 are now treated as equal)
  • Any IdCitation and UnknownCitation object will always be treated as unequal to any other object
  • Previously, CitationBase objects had a comparison_hash() method. This method was never intended to be a "public" method and has been removed.
  • Citation hashes are now stable and reproducible across module loadings of eyecite, as we are now using hashlib.sha256 under the hood. However, note that due to hashlib's implementation details, hashes with NOT be consistent across 32 and 64 machines.

• As noted in 2.3.3 (2021-03-23), the old NonopinionCitation class was renamed UnknownCitation to better reflect its purpose. Support for the old class name has now been completely deprecated. This change is purely semantic -- there is no change in how these citations are handled.

Fixes:

• Update dependencies for reporters-db

2.5.5 - 2024-01-10

Yanked.

2.5.4 - 2024-01-10

Yanked.

2.5.3 - 2024-01-10

Yanked.

2.5.2 - 2023-05-23

Fixes:

• Update dependencies and add Python 3.11 support.

2.5.1 - 2023-03-09

Fixes:

• Fix & update dependencies

2.5.0 - 2023-01-20

Features:

• Citations now have a full_span property that returns the start and end indexes for the full citation, including any pre- or post-citation attributes.

2.4.0 - 2022-07-22

Features:

• Unnumbered citations like "22 U.S. ___" are now identified by eyecite. They will have a page attribute set to None, but otherwise work just like everything else.

2.3.3 - 2021-03-23

Features:

• None

Changes:

• The NonopinionCitation class has been renamed UnknownCitation to better reflect its purpose. This change is purely semantic -- there is no change in how these citations are handled.
• Updates to courts-db

Fixes:

• Initial support for finding short cites with non-standard regexes, including fixing short cite extraction for Mich., N.Y.2d and Pa..

2.3.2 - 2021-03-23

Yanked.

2.3.1 - 2021-03-23

Yanked.

2.3.0 - 2021-09-23

Features:

• Greatly improved documentation
• Autogenerated documentation

Changes:

• This version lands one more iteration of the APIs to make them more consistent. Sorry. Hopefully this will be the last of its kind for a while. The need for these changes became obvious when we began generating documentation. The changes are all in name only, not in functionality. So: 1) the annotate function is renamed as annotate_citations; 2) The find_citations module has been renamed find (so, do from eyecite.find import get_citations instead of from eyecite.find_citations import get_citations); 3) The cleaners module is now named clean; and 4) The clean_text function has been moved from utils to clean (so, do from eyecite.clean import clean_text instead of from eyecite.utils import clean_text).

2.2.0 - 2021-06-04

Features:

• Adds support for parsing statutes and journals and includes new json files with associated regular expressions and data. This introduces FullLawCitation and FullJournalCitation.

• Id and Supra citations now have a metadata.parenethical attribute, to mirror FullCaseCitation objects and make them more useful. PR #71

• A new tool, dump_citations() is added to inspect extracted citations.

• The readme is updated with a new tutorial.

• We now use page-based heuristics while looking up the citation that a pin cite refers to. For example, if an opinion says:

  1 U.S. 200. blah blah. 2 We Missed This 20. blah blah. Id. at 22.

  We might miss the second citation for whatever reason. The pin cite refers to the second citation, not the first, and you can be sure of that because the first citation begins on page 200 and the pin cite references page 22. When resolving the pin cite, we will no longer link it up to the first citation.

  Similarly, an analysis of the Caselaw Access Project's dataset indicates that all but the longest ~300 cases are shorter than 150 pages, so we also now ignore pin cites that don't make sense according to that heuristic. For example, this (made up) pin cite is also likely wrong because it's overwhelmingly unlikely that 1 U.S. 200 is 632 pages long:

  1 U.S. 200 blah blah 1 U.S. 832

  The longest case in the Caselaw Access Project collection is United States v. Philip Morris USA, Inc, at 986 pages, in case you were wondering. Figures.

  Issue #74, PR #79.

Changes:

• To harmonize the API while adding laws and journals, a large API reorganization was completed. See PR 64 for discussion. Here are the details:
  • All of the metdata that we capture before and after each citation is now organized into a metadata object. Thus, if they make sense for the citation type, all of the following attributes are now in some_citation.metadata: publisher, day, month, antecedent_guess, court, extra, defendant, plaintiff, parenthetical, pin_cite.
  • The canonical_reporter attribute is removed from citation objects. It wasn't used much and was duplicated in edition_guess.reporter.short_name. Where applicable, use that instead going forward.
  • The reporter_found attribute is removed from citation objects in favor of groups['reporter'].
  • Similarly, the volume and page attributes are removed from citation objects in favor of groups["volume"]and groups["page"].
  • The reporter attribute has been removed from citations and replaced with the corrected_reporter() method.
  • Similarly, the base_citation() method has been renamed as corrected_citation(), and the formatted() method has been renamed as corrected_citation_full().
  • The do_defendant and do_post_citation arguments to get_citations have been removed. They're fast enough to just always do. No need to think about these further.\
  • The resolve_fullcase_citation parameter in the resolve_citations function has been renamed to resolve_full_citation.

Fixes:

• Support for reporter citations with volume=None is added. Some reporters don't use volumes, for example, "Bankr. L. Rep. (CCH) P12,345".
• Upgrades courts-db subdependency to latest that provides lazy-loading. This should speed up imports of eyecite.

2.1.0 - 2021-05-13

Features:

• Adds support for resolving id, supra, and short form citations into their targets. See readme for details on "Resolving Citations."
• Pin cites are now matched across more citation types.
• Summarizing parentheticals are now included in the match.

Changes:

• The shape of various citation objects has changed to better handle pages and pin citations. See #61 for details.

Fixes:

• Fixes crashing errors on some partial supra, id, and short form citations.
• Fixes unbalanced tags created by annotation.
• Fixes year parsing to move away from isdigit, which can capture unicode superscript numbers like "123 U.S. 456 (196⁴)"
• Allow years all the way back to 1600 instead of 1754. Anybody got a citation from before then?
• Page number matching is tightened to be much more strict about how it matches Roman numerals. This change will prevent some citations from being matched if they have extremely common Roman numerals. See #56 for a full discussion.

2.0.2 - Adds missing dependency to toml file, nukes setup.py and requirements.txt. We're now fully in the poetry world.

2.0.1 - Major rewrite to efficiently build and use hundreds of regular expressions to parse the text, and to use merging algorithms to annotate it. These changes bring better speed, accuracy, and flexibility to the library.

2.0.0 - Broken, bad release process.

1.1.0 - Standardize the __eq__() and __hash__() methods and remove the unused fuzzy_hash() method.

0.0.1 - Initial release with CL-compatible API.

0.0.1 to 0.0.5 - Continuous deployment debugging