How to percent-encode only some characters - sparql

Question

The SPARQL function ENCODE_FOR_URI escapes all except unreserved URI characters in the input. How do I change it to ignore certain (non-ASCII characters for use in IRI for example) characters?

Answer 1

This is a non-standard solution, as it requires additional regex support (lookahead) beyond what the SPARQL specification mandates, but it works for some triple stores/SPARQL engines (e.g. Wikidata). Here's the full solution: it also requires to pick a character that should not (and cannot) be replaced (_ in this case) and a character not present in the input (\u0000 cannot be stored in RDF so this is a good pick)
BIND("0/1&2]3%4#5_" AS ?text)
BIND(REPLACE(?text, "[^\u0001-\u005E\u0060-\u007F]+", "") AS ?filtered) # the characters to keep
BIND(REPLACE(?filtered, "(.)(?=.*\\1)", "", "s") AS ?shortened) # leaves only one of each character
BIND(REPLACE(?shortened, "(.)", "_$1", "s") AS ?separated) # separates the characters via _
BIND(CONCAT(?separated, ENCODE_FOR_URI(?separated)) AS ?encoded) # appends the encoded variant after it
BIND(CONCAT("_([^_]*)(?=(?:_[^_]*){", STR(STRLEN(?shortened) - 1), "}_([^_]*))?") AS ?regex)
BIND(REPLACE(?encoded, ?regex, "$1$2\u0000", "s") AS ?replaced) # groups the character and replacement together, separated by \u0000
BIND(REPLACE(?shortened, "([-\\]\\[])", "\\\\$1") AS ?class) # converts the remaining characters to a valid regex class
BIND(CONCAT(?text, "\u0000", ?replaced) AS ?prepared) # appends the replacement groups after the original text
BIND(CONCAT("([", ?class, "])(?=.*?\u0000\\1([^\u0000]*))|\u0000.*") AS ?regex2)
BIND(REPLACE(?prepared, ?regex2, "$2", "s") AS ?result) # replaces each occurrence of the character by its replacement in the group at the end
If you know the precise replacements beforehand, only the last 3 lines are necessary, to form the string.