chapter 06 · draft · edit the source

Two worlds

One idea: one burial, as many worlds as you like.

Everything so far has been one program and one answer. Now the same trace against two different worlds. Write post.nc:

// post.nc — one burial, and as many worlds as you care to try.
//
// `env` is stratum 2 and `--declare` is how a rite answers it, so the world
// this program runs against fits on the command line. Two exhumings of one
// trace reach two sealed traces, and both keep their names.

Str@2 who = must(descend env { env("WHO") });

U0 greet()
{
  concat("Hello, ", who);
}

demand greet();

env is stratum 2 — shallower than the disk, because reading a frozen environment costs nothing and changes nothing. Bury it once:

$ nether bury post.nc
buried   post.nc → e7e0b6de   depth 2   holes 1   4 nodes
  hole ①  env("WHO")                     stratum 2  env

Now try to answer it, and get told off:

$ nether exhume e7e0b6de --grant env
nether: granting `env` means pinning it: `--clock` and `--target`.

  §8.3.1: there is no default, because a default would be this
  machine's, and a rite whose answer depends on which machine ran
  it is what §6.7 exists to prevent.

That message is worth stopping on. Granting the environment means the rite has to say what the environment is — including the time and the target triple — because a default would quietly be this machine's, and then the trace would depend on which machine happened to run it. There is no convenience here on purpose. Somewhere there is a reproducible build that is only reproducible on the laptop it was invented on, and this is that bug, refused at the door.

So say what the world is:

$ nether exhume e7e0b6de --grant env --declare WHO=world --clock 0 --target x86_64-unknown-linux-gnu
  ①  env("WHO")  →  world  10960a1b
sealed   e7e0b6de + 10960a1b → 732a2f23   depth 2   holes 0

And then say it differently:

$ nether exhume e7e0b6de --grant env --declare WHO=moon --clock 0 --target x86_64-unknown-linux-gnu
  ①  env("WHO")  →  moon  c26253b8
sealed   e7e0b6de + c26253b8 → c9bcc6a6   depth 2   holes 0

Two rites, one burial, two worlds. Both of them are still there:

$ nether lamp 732a2f23
Hello, world
$ nether lamp c9bcc6a6
Hello, moon

What did not happen

post.nc was parsed, checked and evaluated once. The second exhuming did not re-read the file, did not re-typecheck it, and did not redo the part of the work that does not depend on WHO. It answered a hole in a trace it already had.

That is the ordinary shape of work here. Bury the expensive, deterministic part once and keep its name. Answer the part that depends on the world as many times as there are worlds you care about. Nothing is invalidated, because nothing was ever overwritten — e7e0b6de still has its hole, and will have it forever.

The names are the record

732a2f23 and c9bcc6a6 are different names because they are different things, and they are different things because they were told different things. The environment that produced each is not a note somebody kept; it is reachable from the trace, through the witness, by anyone holding the name.

If you built something for two customers and one of them says the wrong thing shipped, the question "which world did this come out of" has an answer, and the answer is not "let me check the CI logs".

Next: asking a trace whether the world has moved.