Update hashes only on successful deployments

[haroldadmin]

Problem

functions-differ writes new hashes to the .differspec.json file even if the detected changes were not deployed to Firebase.
This means that if you run functions-differ again after a failed deployment, you get an empty list of changed functions.

Proposed Solutions

The goal is to make functions-differ command invocation idempotent, meaning that multiple invocations of functions-differ should output the same list of changed functions until a successful deployment.

There's no easy solution to this because we can not detect if the deployment was successful. Some possible solutions are:

• 💡 Use the source code deployed on GCF to calculate old hashes of the functions, instead of relying on hashes stored in .differspec.json.

  • ✔️ Pros:
    • Ensures that old hashes are always calculated from functions that were deployed successfully
  • 🛑 Cons:
    • Increases time of invocation significantly, as we would need to download source code of deployed functions.
    • Increases complexity, as we would need to authenticate with GCP to download sources.
    • Calculating hashes from downloaded source code is complicated, as the source code of the function includes all other files in the project as well.

• 💡 Execute firebase deploy command as a sub process from functions-differ and only update hashes when the sub process exits successfully

  • ✔️ Pros:
    • Ensures that old hashes are always calculated from functions that were deployed successfully
    • Makes the tool more fully-featured, as it provides an out of the box solution for deploying functions.
  • 🛑 Cons:
    • Imposes a dependency on the Firebase CLI, and would require changes whenever the CLI interface changes.
    • In case of a partially successful deployment, it is difficult to detect which functions were deployed sucessfully.