Distribution · text comparison

Doublephone

Works

Double Metaphone over a bundled C implementation — one call turns a word into two four-letter phonetic keys.

Version: 0.1.3 zef:jonathanstowe
Depends: none at runtime
License: Artistic-2.0
Its own test suite: 3 files, green
Checked: 2026-09-15 against Raku++ 3.28.0 and Rakudo 2026.08
Where it lives: raku.land · source

Install it #

$ rakupp install Doublephone

zef install Doublephone writes the same store; either installer leaves the module usable by both engines.

What it is for #

Soundex was designed for English surnames in 1918 and shows it. Double Metaphone is the modern replacement: it handles Germanic, Slavic, Romance and Chinese name patterns, and it returns two keys per word — a primary and an alternative — because a name like Schmidt has more than one plausible pronunciation depending on where the bearer's family came from.

Two names match if any of their keys match.

Coding a name #

File

use Doublephone;

for <Smith Smyth Schmidt Wright Rait Knight Nite
     Jackson Jaxon Gonzalez Gonsalez Pfeiffer Xavier Czerny> -> $w {
    my ($p, $s) = double-metaphone($w);
    say sprintf('%-10s primary=%-6s secondary=%-6s same=%s', $w, $p, $s, $p eq $s);
}

Output

Smith      primary=SM0    secondary=XMT    same=False
Smyth      primary=SM0    secondary=XMT    same=False
Schmidt    primary=XMT    secondary=SMT    same=False
Wright     primary=RT     secondary=RT     same=True
Rait       primary=RT     secondary=RT     same=True
Knight     primary=NT     secondary=NT     same=True
Nite       primary=NT     secondary=NT     same=True
Jackson    primary=JKSN   secondary=AKSN   same=False
Jaxon      primary=JKSN   secondary=AKSN   same=False
Gonzalez   primary=KNSL   secondary=KNSL   same=True
Gonsalez   primary=KNSL   secondary=KNSL   same=True
Pfeiffer   primary=PFFR   secondary=PFFR   same=True
Xavier     primary=SF     secondary=SFR    same=False
Czerny     primary=SRN    secondary=XRN    same=False

Smith and Smyth share both keys. Jackson and Jaxon share both. Gonzalez and Gonsalez agree. And Schmidt gets XMT primary with SMT alternative, which is exactly the two-pronunciation case the algorithm exists for — Smith's alternative is XMT too, so the two names match on the crossed pair.

Matching #

File

use Doublephone;

sub sounds-alike($a, $b) {
    my @a = double-metaphone($a);
    my @b = double-metaphone($b);
    so @a.grep(*.chars) (&) @b.grep(*.chars);
}

for <Smith Schmidt>, <Wright Rait>, <Knight Nite>,
    <Smith Jones>, <Thompson Tomson> -> ($a, $b) {
    say sprintf('%-10s %-10s -> %s', $a, $b, sounds-alike($a, $b) ?? 'match' !! 'no');
}

Output

Smith      Schmidt    -> match
Wright     Rait       -> match
Knight     Nite       -> match
Smith      Jones      -> no
Thompson   Tomson     -> no

Thompson and Tomson do not match, which is the algorithm working correctly: the p is audible.

What the input can be #

File

use Doublephone;

sub show($label, $in) {
    my ($p, $s) = double-metaphone($in);
    say sprintf('%-22s -> (%s, %s)', $label, $p.raku, $s.raku);
}

show 'empty string',    '';
show 'digits only',     '12345';
show 'lowercase',       'smith';
show 'mixed case',      'SmItH';
show 'hyphenated',      'Smith-Jones';
show 'leading spaces',  '  Smith';
show 'accented',        "Ren\c[LATIN SMALL LETTER E WITH ACUTE]e";
show 'Cyrillic',        "\c[CYRILLIC CAPITAL LETTER I]\c[CYRILLIC SMALL LETTER VE]";

Output

empty string           -> ("", "")
digits only            -> ("", "")
lowercase              -> ("SM0", "XMT")
mixed case             -> ("SM0", "XMT")
hyphenated             -> ("SM0J", "XMTJ")
leading spaces         -> ("SM0", "SMT")
accented               -> ("RN", "RN")
Cyrillic               -> ("", "")

Case is folded. Non-Latin input yields two empty strings rather than an error, which is worth guarding: an empty key matches every other empty key, so a corpus with mixed scripts will bucket all of its non-Latin names together.

Note the leading-whitespace row: 'Smith' and ' Smith' give different secondary keys. Trim before calling.

The one thing to know #

An undefined Str segfaults the process, and try cannot save you.

The signature is Str $str, which accepts a type object; the C function then dereferences a null pointer. Exit status 139 is SIGSEGV — not an exception, not a backtrace, just a program that is gone.

The realistic path to it is ordinary: a database column that is NULL, a regex capture that did not match, a hash key that was absent. Guard with // '' or declare Str:D at your own call site.

Where the two engines differ #

Nowhere, including the segfault — Raku++ and Rakudo both die the same way on an undefined argument, and every key in this page is byte-identical on the two.

One difference in noise rather than behaviour: on a cold precompilation cache Rakudo prints a deprecation report at exit about Distribution::Resource.Str, and it does not reappear once the unit is precompiled. Raku++ never prints it.

The thing to size before you rely on the keys: every code is capped at four characters, so Constantinople, Constantine, Constant and Konstanz all collide on KNST. That is fine for a phonetic bucket and wrong if you treat the key as an identifier.