[wpietri]

This line really struck me: “They will have to change at some point, and you don’t want to recompile and redeploy your application just to change the VAT tax rate.”

Continuous deployment has changed this for me. I'm still not going to leave tunable constants in the middle of random if statements. But nowadays instead of trying to avoid recompiling and redeploying, I'd rather invest that time into making that very easy.

[icambron]

Yeah, CD is a big game changer here. Managing config separately from code is a big pain (as that piece points out in colorful detail), and as the countervailing painfulness of a deploy drops toward zero, it becomes better and better to just hardcode some consts. "Business rules" just become "code you write". Easier to change, test, and reason about. At my last job we moved to CD and it was fun to see everyone (haltingly, sometimes unwillingly) change their approach to this.

There's still a place for configs and even DSLs (multiple deployments, on-prem software, biz logic written by non-programmers, etc) but CD eliminates the huge class of "this is too expensive to deploy changes to" cases.

[mst]

Pulling things out into a configuration file (or environment variable) is valuable when either (a) the value needs to be set per-environment and you have a variable number of environments or (b) the value needs to be changeable by somebody other than the dev team.

Also it's worth considering for constants whether what you actually want is a 'resource file' (which could easily just be a file containing all your constants written in your programming language, no need to use a different syntax if that doesn't gain you anything) - learning and internalising (and also learning how to explain to other developers) that a resource file is -not- the same as a configuration file has been a significant quality of life improvement for me.

[mcguire]

Did you happen to see the recent discussion of the maintainer-hoarks-up-a-dependency problem? How does your CD system handle testing?