Documentation
¶
Overview ¶
Package codehl classifies the spans of a fenced code block for syntax highlighting in the assistant's answers: SQL, Go, Python, JavaScript, JSON and shell.
WHY A HAND-WRITTEN LEXER and not chroma or a grammar: the answers are short, redrawn every frame while they stream, and only a handful of languages matter in a SQL client's answers. A table-driven scanner — comments, strings, numbers, a keyword list — is a few hundred lines with no dependency, and it degrades gracefully: a construct it does not know (a JS regex literal, a Go struct tag) is merely uncolored or colored as its nearest lexical cousin, never an error.
SQL is not lexed here: it delegates to sqlsplit.Lex, the scanner the editor colors with, so a statement looks the same in an answer as it will once inserted into the editor. Shell has a scanner of its own too (lexShell): its quoting, $variables, word-bound comments and heredocs fit none of the C-family switches.
The web pane draws answers in the browser, so web/static/js/hl.js mirrors this file rule for rule. Its keyword lists are checked against Keywords by web's tests; the scanning rules are kept in step by hand.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func Keywords ¶
Keywords returns the sorted keyword list of a language (by fence tag), so the browser's copy in hl.js can be tested against the one source of truth.
func Lang ¶
Lang maps a fence's language tag to the lexer that draws it: "sql", "go", "python", "js", "json", "shell", or "" for a language not highlighted.
An untagged fence is SQL, the same call tui/chat.go isSQLLang makes for ⤓ insert: the preamble asks for ```sql and models often drop the tag, so in a SQL assistant an untagged block is overwhelmingly SQL. Coloring a plain-text block that way costs little — only SQL keywords light up. The one exception, decided by Lex from the content, is untagged JSON.
Types ¶
type Kind ¶
type Kind int
Kind classifies a span. Text is everything no span covers.
const ( Text Kind = iota Keyword // a reserved word (and the literal constants: nil, None, true…) String // any string, rune or template literal Number // a numeric literal Comment // line or block comment Ident // a quoted SQL identifier ("…" or `…`) Param // a SQL bind parameter ($1, ?, :name) // Var is a shell variable ($HOME, ${X:-1}, $?). It was once drawn as // Param, which both front ends color as an error (bind parameters are // rare enough in SQL to deserve the alarm); in a shell block $vars are // on most lines, and a page of error red read as a page of mistakes. Var )
Source Files