Upgrading

From 2.5.0 to 2.7.0

Nothing you call changes, and everything 2.5.0 gave is kept. New, each in an entry point of its own that nothing fetches until it is called: loadEntry, readEntry and referenceOf (the whole entry of a word) at @johnmorrisdotca/jisho/entries; loadExamples, exampleCredit and furiganaOfExample (the example sentences of a word) at @johnmorrisdotca/jisho/examples; loadStrokes, loadAllStrokes, strokesOf, strokeOrderSvg, pathPoints and findByDrawing (stroke order, and a kanji found by drawing it) at @johnmorrisdotca/jisho/strokes; and suggestWords in words. A Word has id, the number JMdict gives its entry, and each row carries it: the rows are 318 pieces of the same size where they were 294, and the words entry is 1 MB larger gzipped as one bundle. The package is 14.8 MB larger to download (13.8 to 28.6 MB), all of it in the new entry points, whose data (the entries, the sentences and the strokes) has licences of its own: NOTICE.md names them, and a page that shows a sentence or a stroke has to credit it.

From 2.3.0 to 2.4.0

Nothing you call changes. New: readText, studyList, studyListText, furiganaOf and furiganaOfSurface (read a text, with the words, their furigana, meanings and levels, and a study list), and an InflectedWord now has form, the dictionary form as it was looked up. The words entry gains about 17 kB gzipped when a bundler follows every piece (the reader, and the grade and JLPT tables of the kanji, 14 kB, which it fetches only when readText is called).

From 2.2.0 to 2.3.0

Nothing you call changes. New: findInflected, inflectedOf and the inflected option of loadWords (a conjugated text found by its dictionary form, with the path), and deinflect, also at its own entry point, @johnmorrisdotca/jisho/deinflect. A Word's kinds can now hold n, for a な adjective (静か), where it held nothing: code that took an empty kinds to mean "not a verb or い adjective" must also allow n. The package is about 2 kB gzipped larger for the rules, and the data is the same JMdict, cut into pieces at slightly different places. The Hikidashi recipe is no longer needed, and Hikidashi is no longer a development dependency.

From 2.1.0 to 2.2.0

Nothing you call correctly changes. A call that had a mistake in it, which used to be quietly accepted, now throws a TypeError before anything is fetched:

• A mode that is not "safe" or "unfiltered" ("Safe", "SAFE", "", "bogus"). It used to load the unfiltered words. No mode is still safe.
• A block or allow that is not a list of texts (block: "神"), and the same for the lists withoutWords, onlyWords, withoutMarks and hasMarks take.

gradeOf and levelOf answer null for anything but one character, where an empty text matched. The rows a lookup gives are frozen, so code that pushed to word.senses[0].glosses or sorted word.forms in place must copy first ([...word.forms].sort()). findByMeaning and searchMeanings put a word whose first gloss is the English text first (飲む for drink), so the order of a result can differ from 2.1.0.

From 2.0.0 to 2.1.0

Nothing you call changes, and everything 2.0.0 gave is kept. A Word has two more fields, rank and priority, and a text is made plain before it is looked up (ｶﾐ, a stray space or an invisible character find what カミ does), so a lookup that threw or found nothing for such a text now finds the word. A piece that cannot be fetched rejects with a WordsLoadError where it rejected with whatever the bundler or the browser threw. New: findByMeaning, wordsWithKanji, kanjiOfWord, searchMeanings, romajiToKana, kanaFromRomaji, normalizeText, commonestFirst, priorityTags, rankOfTags and the romaji option. The package is 3.7 MB larger to download (10.2 to 13.9 MB) for the new indexes, which only the new lookups fetch.

Before 2.0.0

forChildren became mode: see the changelog.

A data refresh

The EDRDG asks that a package using its files keeps them current, so KANJIDIC2 and JMdict are each fetched every month, and a change is a patch release. A reading, a meaning or a word may change in a patch, and a word JMdict adds or drops is added or dropped; the grades, levels and pair scores come from files that rarely do.