- Generated by
1.9.8
|
Email Generator 2.1.3
Email addresses in each country's style for C++23
|
The data in resources/ and how email-generator uses it (rule 6 of the dasmig standard).
| Tier | Countries | What it holds |
|---|---|---|
lite | 195 | The sovereign states of country-generator's lite tier |
full | 250 | The territories of its full tier |
Both tiers hold the same style, particle, fold and word tables; profiles.tsv and providers.tsv list the tier's countries.
countries.tsv has region and population, copied from country-generator (the population weights the country draw).
Built by scripts/prepare_email.py. All are curated for this library (rule 6) except where noted; none comes from a survey of real addresses, so the shares are plausible, not measured.
| File | Columns | Source and reliability |
|---|---|---|
fold.tsv | from, to | Unicode decomposition of Latin-1, Latin Extended-A/B and Latin Extended Additional, plus letters that do not decompose (ß ø æ œ đ ð þ ł). Letters of other scripts are not folded |
particles.tsv | particle | Curated list of family-name particles (de, da, van, von, bin, ...) that email names skip |
profiles.tsv | region, profile | Curated: which style profile a country uses (anglo, germanic, romance, slavic, eastasian, japan, arabic, brazil, default) |
email_styles.tsv | profile, pattern, weight | Curated. The shapes were modelled on Faker's username formats and on the name transforms of nickname-generator; the weights are a judgement. Separator variants (. _ - none) are written out as rows, weighted per profile |
providers.tsv | domain, country, kind, weight | Seeded from Faker's locale free_email_domains (MIT), curated: global webmail under *, local domains per country, extra domains for countries Faker lacks. A row for a country replaces the * row of the same domain (weight 0 switches it off there). kind is free (webmail) or isp Eleven countries (DE, GB, RU, JP, KR, FR, IT, ES, BR, US, IN) have exact shares in percent (SHARES in the script), rounded from general knowledge of public market-share reports and not a survey: those countries get exactly the listed domains and every global domain they do not list has weight 0. A weight of 0 switches a domain off, in weighted and uniform draws alike |
words.tsv | list, word | Copied from nickname-generator: adj (adjectives), animal, jp (romaji), lowercased to [a-z0-9] |
Pattern language (email_styles.tsv): {f} given name, {l} family name, {fi} {li} their initials, {n} nickname, {y2} {y4} birth year, {r} one to three random digits, {r2} two, {adj} {animal} {jp} a handle word, anything else a literal from [a-z0-9._-]. A style whose inputs are missing (no family name, no birth year, ...) is not eligible, so a call with no names draws among the handle styles.
Names are folded to lowercase ASCII; the given part is the first word, the family part the first word that is not a particle (so de la Cruz is cruz, Müller-Schmidt is muller). A name in a script with no ASCII form leaves no part, and a handle style is used.
A country (by population, stream "country"), a style of the country's profile among those the request can fill ("style", by weight), handle words ("local"), digits ("number") and a domain ("domain", by weight; the reserved example.* domains with safe). Digits in a style are the birth year ({y2}, {y4}) or random ({r}, {r2}).
python scripts/prepare_email.py writes every table to resources/lite and resources/full. It reads Faker's per-locale free_email_domains (downloaded into build/faker/), the word lists from ../nickname-generator/resources/lite and the countries from ../country-generator/resources. All weights and the curated tables live in the script.
SHARES follow rounded market-share figures from memory; the others split Faker's lists evenly among their local domains. Treat the mix of styles and providers as realistic-looking, not statistically exact. 1.9.8