Formatters

We rely on formatters — like normalize_email() or PhoneNumberFormatter.format() — to make data look the same everywhere. In a microservices world, the same record can be saved dozens of times in one workflow, so an inconsistent formatter can quietly corrupt data and break audit trails.

How many times have we all seen code like this in the model layer of our frameworks? I’m showing both Ruby and Java because this pattern pops up across languages:

before_validation :normalize_email

def normalize_email
  self.email = format_email(email)
end

or

@PrePersist
public void normalizePhone() {
    this.phone = PhoneFormatter.format(phone);
}

Looks familiar, doesn’t it?

The Formatter Under Test

import com.google.i18n.phonenumbers.NumberParseException;
import com.google.i18n.phonenumbers.PhoneNumberUtil;
import com.google.i18n.phonenumbers.Phonenumber;

class PhoneNumberFormatter {
  private static final PhoneNumberUtil phoneUtil = PhoneNumberUtil.getInstance();

  public static String format(String input) throws NumberParseException {
    Phonenumber.PhoneNumber number = phoneUtil.parse(input, "US");
    return phoneUtil.format(number, PhoneNumberUtil.PhoneNumberFormat.NATIONAL);
  }
}

Simple, right? Parse a string, format it as a U.S. number, move on (see the libphonenumber docs for details).

Putting PBT to Work

I used JQwik—a JUnit 5‑style PBT library—to throw 1,000,000 random digit strings at the formatter. The rule was simple:

Formatting twice should give the same result

The test:

import net.jqwik.api.*;

@Property(tries = 1_000_000) // 1,000,000 trials usually finish in milliseconds
void formattingIsConsistent(@ForAll("phoneNumbers") String input) throws NumberParseException {
  String once;
  try {
    once = PhoneNumberFormatter.format(input);
  } catch (NumberParseException ignored) {
    return; // skip invalid inputs
  }

  String twice = PhoneNumberFormatter.format(once);
  
  assertThat(twice)
      .withFailMessage("Failed for: '%s' → '%s' → '%s'", input, once, twice)
      .isEqualTo(once);
}

@Provide
Arbitrary<String> phoneNumbers() {
  return Arbitraries.strings().withChars("1234567890").ofMaxLength(14);
}

The Surprise

Consistency failed for input: ‘0118251144’ → ‘051-144’ → ‘051144’

Hypothesis: the formatter treats national dial codes differently on re‑parse.

On the first pass Google’s library inserts a hyphen; on the second it drops it. That one‑character drift can ripple through logs, duplicate‑detection jobs, or billing systems—and it would be hard to spot with hand‑picked test cases.

Takeaway

Ten minutes of PBT saved what could have been days of production firefighting. If a single assertion can surface issues like this, imagine what a small suite could do for your codebase!

Have you tried PBT? Share your story