Moldova (MD) postcode pattern rejects valid codes and accepts invalid ones

[vadviktor]

Hey @dgilperez , first of all, thank you for this gem. We use this in our app and a client has found an issue that we've patched internally, though I wish to contribute the fix upstream for the benefit of others.

---

The MD entry in ZIPCODES_REGEX is both too strict and too loose:

MD: /\A(([A-Z]){2})(|\s)\d{4}\z/i,

It requires any two letters followed by four digits (optionally separated by a space). That has two problems:

Input	Real status	Gem result	Correct?
MD-2001	valid — canonical international form	rejected (no hyphen allowed)	❌
2001	valid — bare domestic form	rejected (prefix required)	❌
XY2001	not a Moldovan postcode	accepted (any 2 letters pass)	❌
MD2001	valid	accepted	✅

So a genuine address written MD-2001 fails validation, while a nonsense code like XY2001 passes.

Root cause

The gem cites CLDR's postalCodeData.xml as its source. CLDR's MD value was simply:

\d{4}

— four digits, no letter prefix at all. The current gem regex doesn't match its own cited source: a mandatory two-letter prefix was added that CLDR never specified, which is what lets any XY/ZZ prefix through and what blocks the bare 4-digit form.

CLDR's \d{4} alone is also incomplete today: it rejects the canonical international MD-#### form (e.g. MD-2001) used by the Universal Postal Union and Google's address metadata.

Proposed fix

MD: /\A(MD[- ]?)?\d{4}\z/i,

This is CLDR's \d{4}, extended to also accept the documented MD-#### / MD #### / MD#### prefix. It accepts every real Moldovan code and rejects non-MD prefixes:

• valid: 2001, MD2001, MD-2001, MD 2001 (case-insensitive)
• invalid: XY2001, MD-202 (wrong digit count)

I verified it against the full UPU / GeoNames Moldovan postcode set (1,215 distinct codes, including the Transnistria region — Tiraspol MD-3300, Bender MD-3200); all pass, none falsely rejected.

Note on the existing test

spec/spec_helper.rb currently asserts MD-2100 is invalid, which mirrors the broken regex rather than reality. A fix needs to flip that to valid (and add 2001 / a non-MD prefix as the new regression cases).

PR

I have a small PR ready (regex + updated MD fixture + CHANGELOG) — happy to open it if this looks right to you.