tvl-depot/tvix
Vincent Ambo 3419f63575 fix(tvix/eval): ensure all evaluated thunks are correctly memoized
This fixes a very complicated bug (b/246). Evaluation
progresses *much* further after this, leading to several less
complicated bugs likely being uncovered by this

What was the problem?
=====================

Previously, when evaluating a thunk, we had a code path that looked
like this:

    match *thunk {
        ThunkRepr::Evaluated(Value::Thunk(ref inner_thunk)) => {
            let inner_repr = inner_thunk.0.borrow().clone();
            drop(thunk);
            self.0.replace(inner_repr);
        }
        /* ... */
    }

This code path created a copy of the inner `ThunkRepr` of a nested
thunk, and moved that copy into the `ThunkRepr` of the parent.

The effect of this was that the original `ThunkRepr` (unforced!) lived
on in the original thunk, without the memoization of the subsequent
forcing applying to it.

This had the result that Tvix would repeatedly evaluate these thunks
without ever memoizing them, if they occured repeatedly as shared
inner thunks. Most notably, this would *always* occur when
builtins.import was used.

What's the solution?
====================

I have completely rewritten `Thunk::force_trampoline_self` to make all
flows that can occur in it explicit. I have also removed the outer
loop inside of that function, and resorted to more use of trampolining
instead.

The function is now well-commented and it should be possible to read
it from top-to-bottom and get a general sense of what is going on,
though the trampolining itself (which is implemented in the VM) needs
to be at least partially understood for this.

What's the new problem(s)?
==========================

One new (known) problem is that we have to construct `Error` instances
in all error types here, but we do not have spans available in some
thunk-related situations. Due to b/238 we cannot ask the VM for an
arbitrary span from the callsite leading to the force. This means that
there are now code paths where, under certain conditions, causing an
evaluation error during thunk forcing will panic.

To fix this we will need to investigate and fix b/238, and/or add a
span tracking mechanism to thunks themselves.

What other impacts does this have?
==================================

With this commit, eval of nixpkgs mostly succeeds (things like stdenv
evaluate to the same hashes for us and C++ Nix, meaning we now
construct identical derivations without eval breaking).

Due to this we progress much further into nixpkgs, which lets us
uncover more additional bugs. For example, after this commit we can
quickly see that cl/7949 introduces some kind of behavioural issue and
should not be merged as-is (this was not apparent before).

Additionally, tvix-eval is now seemingly very fast. When doing
performance analysis of a nixpkgs eval, we now mostly see the code
path for shelling out to C++ Nix to add things to the store in there.
We still need those code paths, so we can not (yet) do a performance
analysis beyond that.

Change-Id: I738525bad8bc5ede5d8c737f023b14b8f4160612
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8012
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
2023-02-03 10:47:18 +00:00
..
.vscode chore(tvix): fix vscode rust-analyzer recommendation 2022-10-15 16:54:28 +00:00
cli fix(tvix/cli): keep tracking full paths in known_paths 2023-02-02 23:37:34 +00:00
docs docs(tvix): fix minor spelling problems in pointer equality document 2023-01-25 14:30:50 +00:00
eval fix(tvix/eval): ensure all evaluated thunks are correctly memoized 2023-02-03 10:47:18 +00:00
nix-compat feat(tvix/nix-compat/derivation): Display -> to_aterm_string() 2023-02-01 16:31:56 +00:00
nix_cli chore(tvix): upgrade to clap 4.0 2022-12-21 13:23:38 +00:00
proto chore(tvix/store): move castore.proto 2022-12-04 10:41:39 +00:00
serde refactor(tvix/serde): allow dead_code in struct 2023-01-31 15:35:46 +00:00
store feat(tvix/store): add write_nar function 2023-01-31 15:28:22 +00:00
verify-lang-tests test(tvix/eval): add test for builtins parity 2023-01-06 12:00:38 +00:00
.gitignore feat(tvix/): .gitignore target folders 2022-11-11 19:55:12 +00:00
Cargo.lock refactor(tvix/cli): use Wu-Manber string scanning for drv references 2023-02-02 17:50:44 +00:00
Cargo.nix refactor(tvix/cli): use Wu-Manber string scanning for drv references 2023-02-02 17:50:44 +00:00
Cargo.toml refactor(tvix/nix-compat): absorb nar writer 2023-01-31 15:18:39 +00:00
crate-hashes.json refactor(tvix/cli): use Wu-Manber string scanning for drv references 2023-02-02 17:50:44 +00:00
default.nix fix(tvix): add dummy target to attach extra-step to 2023-02-01 17:34:08 +00:00
LICENSE chore(tvix): Bootstrap Tvix folder 2021-03-27 00:09:49 +00:00
OWNERS chore(gerrit): migrate OWNERS files to code-owners style 2022-09-19 11:13:28 +00:00
README.md docs(tvix): add more information to README 2023-02-02 16:25:52 +00:00

Tvix

Tvix is a new implementation of the Nix language and package manager. See the announcement post for information about the background of this project.

Tvix is developed by TVL in our monorepo, the depot, at //tvix. Code reviews take place on Gerrit, bugs are filed in our issue tracker.

For more information about Tvix, feel free to reach out. We are interested in people who would like to help us review designs, brainstorm and describe requirements that we may not yet have considered.

Most of the discussion around development happens on our IRC channel, which you can join in several ways documented on tvl.fyi, or on our mailing list.

Contributions to Tvix follow the TVL review flow and contribution guidelines.

WARNING: Tvix is not ready for use in production. None of our current APIs should be considered stable in any way.

WARNING: Any other instances of this project or repository are josh-mirrors. We do not accept code contributions or issues outside of the tooling and communication methods outlined above.

Components

This folder contains the following components:

  • //tvix/eval - an implementation of the Nix programming language
  • //tvix/nix-compat - library functions for compatibility with C++ Nix
  • //tvix/cli - preliminary REPL & CLI implementation for Tvix
  • //tvix/serde - Rust library for using the Nix language for app configuration
  • //tvix/store - implementation of a file store for Tvix

Some additional folders with auxiliary things exist and can be explored at your leisure.

Building the CLI

The CLI can also be built with standard Rust tooling (i.e. cargo build), as long as you are in a shell with the right dependencies.

  • If you cloned the full monorepo, it can be provided by mg shell // tvix:shell.
  • If you cloned the tvix workspace only (git clone https://code.tvl.fyi/depot.git:workspace=views/tvix.git), nix-shell provides it.

If you're in the TVL monorepo, you can also run mg build //tvix/cli (or mg build from inside that folder) for a more incremental build.

Please follow the depot-wide instructions on how to get mg and use the depot tooling.

Compatibility

Important note: We only use and test Nix builds of our software against Nix 2.3. There are a variety of bugs and subtle problems in newer Nix versions which we do not have the bandwidth to address, builds in newer Nix versions may or may not work.

Rust projects, crate2nix

Some parts of Tvix are written in Rust. To simplify the dependency management on the Nix side of these builds, we use crate2nix in a single Rust workspace in //tvix to maintain the Nix build configuration.

When making changes to Cargo dependency configuration in any of the Rust projects under //tvix, be sure to run mg run //tvix:crate2nixGenerate -- in //tvix itself and commit the changes to the generated Cargo.nix file. This only applies to the full TVL checkout.

License structure

All code implemented for Tvix is licensed under the GPL-3.0, with the exception of the protocol buffer definitions used for communication between services which are available under a more permissive license (MIT).

The idea behind this structure is that any direct usage of our code (e.g. linking to it, embedding the evaluator, etc.) will fall under the terms of the GPL3, but users are free to implement their own components speaking these protocols under the terms of the MIT license.