[RFC 0062] Content-addressed paths (NixOS#62)

* CAP RFC: First draft

* typo

* Apply @grahamc's suggestions

Co-Authored-By: Graham Christensen <[email protected]>

* nix code -> Nix expression

* Break-up the big introduction paragraph

As suggested in https://github.com/NixOS/rfcs/pull/62/files#r356694585

* Rename to match the PR number

* Rename the drv attribute to __contentAddressed

Makes it more in line with other "magic" attributes like
`__structuredAttributes`

Also fix the orthograph

* Mention the GC issue

* Remove the ambiguity on what an `output` is

* Replace aliases paths by a pathOf mapping

* Move the example after the design description

* Rephrase the design

In particular, replace `static` and `dynamic` by `symbolic` and
`resolved`

* Add shepherd team

* Rewrite the RFC to account for the RFC meeting comments

- Add the notion of `drvOutputId`
- Replace the alias paths by a `PathOf(drvOutputId)` function
- Mention the notion of "truster" in the `PathOf` function

* Add a section about leaking output paths

* Refine the design summary

* Rename dependency-addressed into input-addressed

And define the term at the begining of the RFC

* minor fixup after comments

* Apply suggestions from code review

Co-authored-by: asymmetric <[email protected]>
Co-authored-by: Profpatsch <[email protected]>

* Update rfcs/0062-content-addressed-paths.md

* Update the terminology to match the in the implementation

* Reword the detailed design presentation

* Quote some strings in the yaml frontmatter

* Add a design paragraph about the remote caching

And update the “unresolved questions” to take it into account

* Lift the determinism requirement

It's not required anymore by the current remote caching semantics

* Typo

Co-authored-by: davidak <[email protected]>

* Apply suggestions from code review

Co-authored-by: davidak <[email protected]>

* Rewrite the RFC

This is mostly a full-rewrite of the RFC to

1. Make it more “incremental”: A first part just describes the minimal
   model upon which everything is based, and a second one shows
   different extensions of this model to add more features
2. Remove the big ugly examples that don’t add much value because they
   aren’t really readable
3. Add a python pseudo-code pseudo-formalisation of the RFC.
  This is imho both more readable and precise than nested bullet-points
  of handwaved language

* Make the python samples a bit more pythonic

Co-authored-by: zseri <[email protected]>

* Explicit that unresolved dependencies are eval-time

* Prettify

* Make the end-goal an experiment

Simplify the paperwork and just get this to FCP because right now it’s stuck in a hole

Co-authored-by: Théophane Hufschmitt <[email protected]>
Co-authored-by: Graham Christensen <[email protected]>
Co-authored-by: John Ericson <[email protected]>
Co-authored-by: asymmetric <[email protected]>
Co-authored-by: Profpatsch <[email protected]>
Co-authored-by: Jörg Thalheim <[email protected]>
Co-authored-by: Eelco Dolstra <[email protected]>
Co-authored-by: davidak <[email protected]>
Co-authored-by: zseri <[email protected]>