Debug vs. Release builds

Protocol Design
1

Editor’s note: this is an older architecture discussion from before we posted them all on the forums right away.

Table of Contents

  1. Debug vs. Release Builds
    1. prod
    2. test/dev
    3. docs
    4. interactive

Debug vs. Release Builds

In the early days, there were small, slow disks, and people saved on
disk space and load time by using strip(1) on binaries to remove all the
symbols. This is, as far as I can tell, the origin of the debug/release
distinction.

Elixir, being by web devs, has web dev concepts; MIXENV supports :dev
and :prod, the latter mostly intended to be used with releases. (There’s
also :test.) Web developers have to deal with such things as “dev” and
“production” databases, API secrets, and so forth, and often use these
environments to do so.

We don’t really have these concerns, and should move away from things
like them in some areas; despite the default names Mix uses for
configurations, what we actually want is a debug/release build
distinction. In particular, despite having a database, we don’t have a
“production database” in the sense a webapp does, and therefore have no
desire to distinguish between a locally spun-up “dev database” vs. a
“production database” with customer data.

prod

The default “prod” is our release build, and it’s to be considered
normal operation. This is the mode used to make releases. Basically, any
difference in program behavior from this mode is wrong.

test/dev

I can think of little we would want to guard under the :test environment
specifically, so I’m lumping it in with dev here; there may be
something.

As the “debug build”, and the one developers end up in most often (mix
tasks and iex will assume :dev), these modes are allowed to do things
that don’t change application behavior in any way. Some things they can
do:

  • use things like our macros to turn all defp into def and put
    docstrings on them, making debugging a bit simpler
  • crank up the logging level, resulting in more things dumped to the
    screen
  • set up iex to be a bit more convenient (i.e., i think this should be
    guarded under “not prod”; i have to look into this a bit more)

Some things we should avoid:

  • things along the lines of “dev databases”; this should be uniform
    among all methods of running
    • there’s a caveat here. the standard data directory will get
      clobbered between dev and prod, and it’s not unreasonable to expect
      some random user, using the released software, to download the
      source and run it in dev mode to try something out. we don’t want to
      clobber their files! while I don’t know precisely how this should be
      handled, it shouldn’t be more distinct than having files tagged
      “dev” for dev mode.
  • hardcoded defaults for things configured or configurable in the
    release mode.
    • if there’s useful debugging/test defaults… make a module and put
      them in it! then load them from the repl! if we’re not capable of
      doing this, change things until we are.
  • relatedly to the above, autostarting in any mode is a bit dubious to
    me.
    • startup tasks should be a task given as argument to the release
    • opening iex shouldn’t start things up
    • saying “iex” isn’t saying “start”
    • at any rate, debug/release builds are not where to have this
      distinction

docs

We may want docs to have their own mode; this section is a placeholder
until we consider it.

interactive

It’s possible we want interactive use to have its own configuration; I
can’t think of when we would.

2

As per this issue on GH: Logging level in application · Issue #1386 · anoma/anoma · GitHub

Currently the log level is set to :error, which might be at least :warning.

I’m in favor of putting :debug in config/config.exs and :warning in config/prod.exs for four reasons:

  • It allows you to leave in debug statements (like “starting indexer with arguments…”) and nudges the author to write useful and proper messages because they’re part of the code base.
  • Logging statements get removed at compile-time depending on the logging level set in the configuration so there is zero overhead at runtime if they’re not shown.
  • You cannot set the log level at runtime in your personal ~/.iex.exs because it’s a compile-time parameter.
  • A more personal reason to raise this issue: I always change the level to :debug in config.exs and I accidentally committed this before, which should be avoided.

I think this is in line with the point raised above “crank up the logging level, resulting in more things dumped to the screen”.