Skip to content
internal/prose
Updated
On this page

The shared unit-pickers behind every summary selfdoc emits: a whole first sentence or paragraph, never truncated mid-word and never given punctuation.

#internal/prose

#internal/prose

Package prose holds the shared unit-pickers that extract complete linguistic units from text.

Every summary selfdoc emits (bullet text, frontmatter "description", llms.txt and Atom-feed entries, auto-extracted meta descriptions) is a complete linguistic unit: a whole first sentence or a whole first paragraph. There are no character caps, no ellipses and no synthesized punctuation -- the text is never truncated mid-word and never gains a period it did not already have.

Two granularities: [FirstSentence] returns the first sentence of the first paragraph, and [FirstParagraph] returns the whole first paragraph with its soft-wrapped lines joined into one line.

[JoinWrappedLines] normalizes source-wrapped prose (Go, JSDoc and KDoc doc comments wrap at about 75 columns) by joining soft-wrapped lines within a paragraph while leaving code blocks, list items and doctest lines verbatim.

#FirstParagraph

Go go
func FirstParagraph(text string) string

FirstParagraph returns the first paragraph of text as a single joined line.

Leading blank lines are skipped and the paragraph ends at the next blank line. Soft-wrapped physical lines within the paragraph are joined with a single space, so the result is one complete linguistic unit. No caps, no ellipsis, no synthesized punctuation.

#FirstSentence

Go go
func FirstSentence(text string) string

FirstSentence returns the first sentence of the first paragraph of text.

The sentence includes its terminating ".", "!" or "?". When the paragraph contains no sentence terminator the whole paragraph is returned unchanged -- a complete unit is always emitted and punctuation is never synthesized.

#JoinWrappedLines

Go go
func JoinWrappedLines(text string) string

JoinWrappedLines joins soft-wrapped physical lines within the paragraphs of text.

Doc comments (Go, JSDoc, KDoc) wrap at about 75 columns, so a single sentence spans several physical lines. This joins those lines within a paragraph so downstream unit-pickers see whole sentences. Preserved verbatim:

- blank-line paragraph breaks - fenced code blocks (``` and ~~~) - indented preformatted blocks (the Go doc convention) - list items ("-", "*", "+", "1.") - doctest lines (">>>" and "...") - ATX headings, which end the paragraph above and start nothing: a heading joined to the sentence under it becomes part of the heading

Idempotent: joining already-joined prose is a no-op.

Search