i18ncheck

Overview ¶

i18ncheck reports where a translated Hugo content tree has drifted from the language it was translated from, in the four ways a translation must not change a page:

• a page missing in a language, or present in one language only;

• a fenced code block that is not byte-identical, in the same order: a transcript is what a program printed, and a translated transcript is one the program never printed;

• a heading whose anchor differs: the translation keeps the source language's anchor as an explicit {#id}, so that every link into the page keeps its fragment in every language;

• a relref target, or a weight, that differs: the same links, the same place in the sidebar.

  i18ncheck [-base en] <content-dir>

<content-dir> holds one directory per language (content/en, content/fr...). It exits 1 when it finds a difference, 2 on a usage error, 0 otherwise, and says how many pages and blocks it compared, because a check that compared nothing passes everything.

It was written after a night of translating go-authn's documentation into three languages with the same checks redone by hand -- an awk loop per run, one helper script per agent -- until three agents' helpers, sharing one name in one scratch directory, overwrote each other. A check that matters gets a home, a name nobody else uses, and tests.

Heading anchors are computed as Hugo's default (autoHeadingIDType "github") computes them: lower case, letters, digits, "-" and "_" kept, a space becoming "-", everything else dropped, and "-1", "-2"... for repeats on one page. Inline code and emphasis markers and link targets are removed first, as rendering would.