Artificial Intelligence

Accelerating code migrations with AI

February 22, 2025
roosho

As Google’s codebase and its merchandise evolve, assumptions made previously (generally over a decade in the past) now not maintain. For instance, Google Adverts has dozens of numerical distinctive “ID” varieties used as handles — for customers, retailers, campaigns, and many others. — and these IDs have been initially outlined as 32-bit integers. However with the present development within the variety of IDs, we count on them to overflow the 32-bit capability a lot prior to anticipated.

This realization led to a big effort to port these IDs to 64-bit integers. The undertaking is troublesome for a number of causes:

  • There are tens of 1000’s of places throughout 1000’s of information the place these IDs are used.
  • Monitoring the adjustments throughout all of the concerned groups can be very troublesome if every staff have been to deal with the migration of their information themselves.
  • The IDs are sometimes outlined as generic numbers (int32_t in C++ or Integer in Java) and are usually not of a novel, simply searchable sort, which makes the method of discovering them via static tooling non-trivial.
  • Modifications within the class interfaces have to be taken into consideration throughout a number of information.
  • Exams have to be up to date to confirm that the 64-bit IDs are dealt with appropriately.

The total effort, if carried out manually was anticipated to require many, many software program engineering years.

To speed up the work, we employed our AI migration tooling and devised the next workflow:

  1. An professional engineer identifies the ID they need to migrate and, utilizing a mixture of Code Search, Kythe, and customized scripts, identifies a (comparatively tight) superset of information and places emigrate.
  2. The migration toolkit runs autonomously and produces verified adjustments that solely comprise code that passes unit assessments. Some assessments are themselves up to date to mirror the brand new actuality.
  3. The engineer shortly checks the change and probably updates information the place the mannequin failed or made a mistake. The adjustments are then sharded and despatched to a number of reviewers who personal the a part of the codebase affected by the change.

Be aware that the IDs used within the inside code base have applicable privateness protections already utilized. Whereas the mannequin migrates them to a brand new sort, it doesn’t alter or floor them, so all privateness protections will stay intact.

For this workstream we discovered that 80% of the code modifications within the landed CLs have been AI-authored, the remainder have been human-authored. The whole time spent on the migration was diminished by an estimated 50% as reported by the engineers doing the migration. There was important discount in communication overhead as a single engineer might generate all mandatory adjustments. Engineers nonetheless wanted to spend time on the evaluation of the information that wanted adjustments and on their overview. We discovered that in Java information our mannequin predicted the necessity to edit a file with 91% accuracy.

The toolkit has already been used to create tons of of change lists on this and different migrations. On common we obtain >75% of the AI-generated character adjustments efficiently touchdown within the monorepo.