Named, not written inline
Regexes.uuid instead of a 36-character literal pasted into the third file that needs one.
Patterns named for what they match — UUIDs, IP addresses, ISO 8601 times, IBANs — each a frozen Regexp that chains anchors, repetition and captures, tested against Ruby's own parsers and runs the same in JavaScript.
The patterns worth naming are the ones that are easy to get almost right.
The IPv4 regex everyone writes finds addresses that aren't there.
log = 'Retry 10.0.0.1, not 1.2.3.456'
log.scan(/\d{1,3}(?:\.\d{1,3}){3}/) # => ["10.0.0.1", "1.2.3.456"]
log.scan(Regexes.ipv4_address) # => ["10.0.0.1"]IPv6 has nine compressed forms and an embedded IPv4 one. Nobody writes it from memory.
Regexes.ipv6_address.whole.match?('::ffff:192.0.2.1') # => true
Regexes.ipv6_address.whole.match?('1::2::3') # => false^ and $ are line anchors in Ruby, so a validation written with them lets a second line through.
"ok\nevil".match?(/^[a-z]+$/) # => true
Regexes.lowercase_letters.one_or_more.whole.match?("ok\nevil") # => falseA date's fields have ranges, and \d{2} doesn't know them.
Regexes.iso8601_date_time.whole.match?('2026-11-21T14:30:05+05:30') # => true
Regexes.iso8601_date_time.whole.match?('2026-13-21T25:61') # => falseParts come back by name, in the standard's own terms.
version = Regexes.semantic_version.match('2.1.3-rc.1+sha.5114f85')
version[:prerelease] # => "rc.1"
version[:build] # => "sha.5114f85"
Regexes.http_url.match('http://[::1]:8080/a?b=c#d')[:host] # => "[::1]"The same pattern runs in the browser and in a JSON Schema, translated rather than copied by hand.
Regexes.e164_phone_number.whole.ecma_source
# => "^(?<![0-9+])\\+[1-9][0-9]{1,14}(?![0-9])$"