Documentation
¶
Overview ¶
Package symtext is the single definition of what an ELPS symbol looks like as TEXT: which bytes make up one, where the one under a cursor starts and ends, and how to reach a line of a document without taking the document apart.
Two servers answer cursor questions about a document -- the language server (package lsp) and the MCP server (package mcpserver) -- and each carried its own copy of this code. The copies drifted, in both of the ways a duplicated helper drifts (elps#654):
- CORRECTNESS. "&" is a symbol character in ELPS: (defun f (&rest xs) ...) names the symbol "&rest", and "&" is legal inside package and symbol names besides. The lsp copy knew that; the MCP copy did not, so a cursor in "&rest" answered for "rest", and a cursor in "p&q:a&b" answered for "q:a".
- COST. The lsp copy walks to the requested line and stops (elps#641, elps#653); the MCP copy split the whole document on "\n" first, so one hover on a 112 KiB file allocated 270 KiB to read one line.
This package is the lsp code MOVED, byte for byte, so the language server's behaviour is unchanged and the MCP server adopts it. It is under internal/ deliberately: an ELPS embedder gets the servers, not their cursor arithmetic, and keeping it closed leaves the definition free to follow the lexer without an exported API to hold still.
COLUMNS ARE BYTES everywhere here, which is what token.Location counts and what both servers work in internally. A column taken off an LSP wire is in the negotiated position encoding and must be converted before it reaches this package -- lsp does that at its inbound boundary (elps#464).
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func IsSymbolChar ¶
IsSymbolChar reports whether c can appear in an ELPS symbol name.
"&" is on the list and belongs there: the argument markers "&rest", "&optional" and "&key" are ordinary symbols whose first byte it is, and it is accepted inside package and symbol names generally.
func LineAt ¶
LineAt returns the text of the 0-based line within content, without its trailing newline, or "" when the line does not exist.
Lines are split on "\n" exactly as both servers do elsewhere, so a CRLF document leaves the "\r" on the end of the line. That is harmless for column arithmetic: "\r" is ASCII, so it changes neither a byte column nor a UTF-16 column.
It walks to the line rather than splitting the document, so the cost of a lookup is the text ahead of the cursor and the allocation is nothing at all.
func WordAt ¶
WordAt extracts the symbol-like word at the given 0-based line and 0-based byte column of content, or "" when there is none.
func WordBoundsInLine ¶
WordBoundsInLine returns the byte bounds of the symbol-like word containing the 0-based byte column col within ln, and whether there is one. The cursor may sit inside the word or immediately after its last byte; both name the whole word.
A missing line and an empty line both contain no word.
Types ¶
This section is empty.
Source Files