I recently tried to explain when a developer should be cautious versus when they should freely experiment. This caused me to realize I never learned this lesson explicitly. It’s an intuition I built over time and forgot I had to learn. I believe the answer aligns with what makes a good development environment.
Trying to identify times for caution first led me to answers like
- Any time live data is involved (this tends to be the most powerful flag for caution because it’s also usually the hardest to recover from)
- Any time production assets are involved
- Any chance of permanent loss
- Possibility of breaking other people’s work
- other developers
- assets used by other workers
- public API changes that will effect consumers on release
These are all true, but there a bit of a lie. I consider API changes at a design level while coding, but everything else I don’t usually consider until pull requesting or other forms of integration (e.g. deployment). My most basic criteria for caution is a workflow stage: pull requests (or code integration in genreal).
Stated differently, I feel I can be bold and break things as long as I’m still in my development environment. This means that the development environment can’t violate any of the caution indicators listed above. A development environment should
- never use live data
- never use production assets or risk harming them
- never risk permanent loss of meaningful data or assets
- never risk breaking other contributor’s workflows
To meet these assumptions a development environment should
- be disposable. Easy to delete and start from scratch again
- should not share stateful resources with any other contributors
- should not share data stores, limited capacity infrastructure, non-isolated infrastructure, configuration, source code branches, etc
- require a clear and mandatory step to transition work from development to shared
This is somewhat analogous to why immutability is becoming popular in programming languages. It follows the principle of least surprise. It causes our development environments to more often behave how we naturally expect them to.
As a bonus, these same questions also feed into deployment and disaster recovery factors. For example,
- We should ensure permanent data loss is never possible
- We should deploy such that consumer workflows are never broken
- We should deploy such that broken production assets are quickly recovered from
- Modifying production assets should require authorization, intention, and predictable process (controlled access, no accidental modification)
Flags for developer caution and properties of a good development environment are two sides of the same question. A development environment should avoid any need for caution and draw a clear boundary before caution becomes necessary.