Troubleshooting and questions

A kanji has no grade

gradeOf answers null for a kanji outside the jōyō and jinmeiyō lists. Nothing is wrong: readKanji holds only the 2,999 with a grade, so such a kanji has no row either.

A kanji has no level

levelOf answers null for a kanji on none of Waller's lists. 172 jōyō kanji are on none of them; that says nothing about how hard one is.

A kanji has no look-alike

Yencken's two measures cover the 1,945 jōyō kanji of before 2010, and the shape score that reaches the rest keeps only the ten nearest for each kanji: 9 of the 196 kanji added in 2010 have none. See how the pairs are measured.

A common kanji is said to have no meaning

You may have a compatibility character, which looks the same and is a different code point, and which KANJIDIC2 gives no meaning. "海".normalize("NFC") gives the common one.

A word has no meaning

meaningOf answers null and wordsOf an empty list when JMdict has no word for the text, and also when every sense it has is one JMdict marks vulgar, X-rated, sensitive or derogatory or whose gloss names a sex act, a sex organ or the sex trade, because words are in safe mode unless you pass mode: "unfiltered", and safe mode does not fetch what it leaves out. Nothing is wrong. Look the text up unfiltered to tell the two apart.

A conjugated word finds nothing

wordsOf and loadWords look up exactly what the dictionary lists, so 食べなかった finds nothing there. Use findInflected, or load with inflected: true and read with inflectedOf. If that finds nothing too, the form may be one it does not read (the imperative, honorific forms, a noun with だ) or a typo: what is covered lists both.

readText cuts a word wrongly, or does not find a word

It takes the longest word the dictionary has, with no grammar, so a rare long word can win, a name is not found or is cut into small words, and a word with several readings gets the commonest one JMdict ranks (the card's alternatives has the others). In safe mode a word the mode leaves out is not found, so its text is listed in unknown: read it with mode: "unfiltered" to tell the two apart. See read a text.

wordsOf says a text was not loaded

It throws for a text that loadWords was not asked for, instead of answering with no words, which would say the dictionary has none. Load it first, or pass the words you have as previous to loadWords and add to them. The common table, from loadCommonWords, answers for every text.

A lookup rejects with a WordsLoadError

A piece of the dictionary could not be fetched, even on the second try. error.piece says which ("the readings piece 12 of 318"), and error.cause is what the fetch rejected with: the network was down, the host does not serve the file, or a bundler left the piece out of the build. Nothing is remembered of the failure, so asking again tries again. Look words up has the details.

findByMeaning does not find a word

It finds a word by the whole words of its glosses, from an index of twenty words for each English term (up to ten whose gloss is exactly the term, then the commonest that hold it), and a word JMdict does not rank by the terms of its first gloss alone. A word that only holds a very common term in its gloss may be beyond the twenty; try a phrase that follows the gloss (go out), or the term that is rarer. It does not stem (walking does not find "walk") and it is English only.

A server's bundle got very large

Something imported @johnmorrisdotca/jisho/words where a bundler follows it, and the bundler reached every piece of the dictionary, about 50 MB. Look words up says how to keep it out.