| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
When dealing with a formal argument in a function argument pattern that
has a default expression, there are two different things that can happen
at runtime: Either we select its value from the passed attribute
successfully or we need to use the default expression. Both of these may
be thunks and both of these may need finalisers. However, in the former
case this is taken care of elsewhere, the value will always be finalised
already if necessary. In the latter case we may need to finalise the
thunk resulting from the default expression. However, the thunk
corresponding to the expression may never end up in the local's stack
slot. Since finalisation goes by stack slot (and not constants), we need
to prevent a case where we don't fall back to the default expression,
but finalise anyways.
Previously, we worked around this by making `OpFinalise` ignore
non-thunks. Since finalisation of already evaluated thunks still
crashed, the faulty compilation of function pattern arguments could
still cause a crash.
As a new approach, we reinstate the old behavior of `OpFinalise` to
crash whenever encountering something that is either not a thunk or
doesn't need finalisation. This can also help catching (similar)
miscompilations in the future. To then prevent the crash, we need to
track whether we have fallen back or not at runtime. This is done using
an additional phantom on the stack that holds a new `FinaliseRequest`
value. When it comes to finalisation we check this value and
conditionally execute `OpFinalise` based on its value.
Resolves b/261 and b/265 (partially).
Change-Id: Ic04fb80ec671a2ba11fa645090769c335fb7f58b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8705
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Autosubmit: sterni <sternenseemann@systemli.org>
|
|
C++ Nix resolves home relative paths at [parse] time. This is not an
option for us, since it prevents being able to separate the compilation
and execution phase later (e.g. precompiled nix expressions). However, a
practical consequence of this is that paths expressions are always
literals (strict) and never thunks.
[parse]: https://github.com/NixOS/nix/blob/7066d21a0ddb421967980094222c4bc1f5a0f45a/src/libexpr/parser.y#L518-L527
Change-Id: Ie4b9dc68f62c86d6c7fd5f1c9460c850d97ed1ca
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7041
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
When comparing to C++ Nix, we notice that the thunking of default
expressions in function formals corresponds to their normal thunking,
e.g. literals are not thunked. This means that we can just invoke
compile() without much of a care and trust that it will sort it out
correctly.
If function formals blow up as a result of this, it likely indicates
that the expression is treated incorrectly by compile(), not
compile_param_pattern().
Change-Id: I64acbff2f251423eb72ce43e56a0603379305e1d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8704
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
C++ Nix forces and typechecks the passed argument even if it is not
necessary in order to compute the return value of the function. I
discovered this when I thought our formals miscompilation might be that
we are too strict, but doesn't look like it in this case.
Change-Id: Ifb3c92592293052c489d1e3ae8c7c54e4b6b4dc6
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8701
Tested-by: BuildkiteCI
Autosubmit: sterni <sternenseemann@systemli.org>
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
Change-Id: I27f9105ddb20d84342550b2a73b479a7764ee3fe
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8699
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
|
|
As cl/8658 and b/274 reveal, lambda expressions are also wrapped in
thunks.
Resolves b/274.
Change-Id: I02fe5c8730ac76748d940e4f4427116587875275
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8662
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
HasAttrs was weird because with longer attribute paths it would
sometimes not turn out to be a thunk. If it was a thunk, it'd usually
still do some eval strictly which we'll want to avoid.
Verified against C++ Nix using a new test suite introduced in a later
CL.
Change-Id: I6d047ccc68d046bb268462f170a3c4f3c5ddeffe
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8656
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
Probably no real world code broken by this overzealous evaluation, but
let's be thorough!
Change-Id: Ib405a677182eab7940ace940c68e107573473a54
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8655
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
|
|
Unary operator applications are thunked which can easily be observed by
nix-instantiate --eval -E '[ (!true) (-1) ]'
Unfortunately, there are few simple expressions where this makes a
difference in the end result. Thus it only cropped up when using nixpkgs
for cross compilation: Here we would compile the expression
!(stdenv.cc.isGNU or false)
to assemble python3Minimal's passthru attribute set (at least this seems
to be the most likely explanation from the backtraces I've studied).
This means that an unthunked
<stdenv.cc.isGNU or false>
OpForce
OpInvert
would be performed in order to assemble this attribute set, causing
stdenv.cc to be evaluated too early, causing an infinite recursion.
Resolves b/273.
It seems that having a test suite that doesn't use --strict and relies
on thunks rendered as <CODE> would be beneficial for catching such
issues. I've not been able to find a test case with --strict that
demonstrates the problem fixed in this CL.
Change-Id: I640a5827b963f5b9d0f86fa2142e75e3a6bbee78
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8654
Tested-by: BuildkiteCI
Autosubmit: sterni <sternenseemann@systemli.org>
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
Change-Id: Ie66cb1b163a544d45d113fd0f866286f230b0188
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7960
Tested-by: BuildkiteCI
Reviewed-by: grfn <grfn@gws.fyi>
|
|
This feature allows the compiler to detect situations where the
created thunk is useless and can be merged into the parent chunk
instead.
The only case where the compiler does this initially is when
statically optimising a select expression.
For example, previously the expression `builtins.length` compiled into
two thunks:
1. An "inner" thunk which contained an `OpConstant` that had the
optimised `length` builtin in it.
2. An "outer" thunk which contained an `OpConstant` to access the
inner thunk, and the trailing OpForce of the top-level program.
With this change, the inner thunk is skipped completely and the outer
chunk directly contains the `length` builtin access.
This can be applied in several situations, some easier than others,
and we will add them in as we go along.
Change-Id: Ie44521445fce1199f99b5b17712833faea9bc357
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7959
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
Emits the span of the `set` that is being accessed in the `force`
operation of an attribute access.
Looking at traces, it's a lot more useful to get information about
*what* is being forced, as in cases like `foo.bar` it can be
misleading to have an error highlight `bar`, when the error occured
while forcing `foo` to be able to access `bar` in the first place.
Change-Id: Id46ff28f20c67cb4971727ac52cc4811795cea2d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8272
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
This produces traces in which we can see what kind of native code was
run. Note that these "names" are named after the generator message, so
these aren't *really* intended for end-user consumption, but we can
give them saner names later.
Example:
https://gist.github.com/tazjin/82b24e92ace8e821008954867ee05057
This already makes the traces a little easier to parse.
Change-Id: Idcd601baf84f492211b732ea0f04b377112e10d0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8268
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Autosubmit: tazjin <tazjin@tvl.su>
|
|
When emitting an error at runtime, the VM will now use the new
`NativeError` and `BytecodeError` error kinds (which just wrap inner
errors) to create a set of diagnostics to emit.
The primary diagnostic is emitted last, with `error` type (so it will
be coloured red in terminals), the other ones will be emitted with
`note` type, highlighting the causal chain.
Example:
https://gist.github.com/tazjin/25feba7d211702453c9ebd5f8fd378e4
This is currently quite verbose, and we can cut down on this further,
but the purpose of this commit is to surface more information first of
all before worrying about the exact display.
Change-Id: I058104a178c37031c0db6b4b3e4f4170cf76087d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8266
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
|
|
Formals can be initialised with deferred default values (see the test
cases), in which case they need an extra thunk to have something that
can be finalised appropriately when the setup is done.
Fixes: b/255
Change-Id: I380e3770be68eaa83ace96d450c7cead32dacc9f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8196
Tested-by: BuildkiteCI
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
|
|
This shaves another 8 bytes off Value. How did that type get so big?!
Change-Id: I65e9b59a1636bd57e3cc4aec5fea16887070b832
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8153
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Tested-by: BuildkiteCI
|
|
Warning: This is probably the biggest refactor in tvix-eval history,
so far.
This replaces all instances of trampolines and recursion during
evaluation of the VM loop with generators. A generator is an
asynchronous function that can be suspended to yield a message (in our
case, vm::generators::GeneratorRequest) and receive a
response (vm::generators::GeneratorResponsee).
The `genawaiter` crate provides an interpreter for generators that can
drive their execution and lets us move control flow between the VM and
suspended generators.
To do this, massive changes have occured basically everywhere in the
code. On a high-level:
1. The VM is now organised around a frame stack. A frame is either a
call frame (execution of Tvix bytecode) or a generator frame (a
running or suspended generator).
The VM has an outer loop that pops a frame off the frame stack, and
then enters an inner loop either driving the execution of the
bytecode or the execution of a generator.
Both types of frames have several branches that can result in the
frame re-enqueuing itself, and enqueuing some other work (in the
form of a different frame) on top of itself. The VM will eventually
resume the frame when everything "above" it has been suspended.
In this way, the VM's new frame stack takes over much of the work
that was previously achieved by recursion.
2. All methods previously taking a VM have been refactored into async
functions that instead emit/receive generator messages for
communication with the VM.
Notably, this includes *all* builtins.
This has had some other effects:
- Some test have been removed or commented out, either because they
tested code that was mostly already dead (nix_eq) or because they
now require generator scaffolding which we do not have in place for
tests (yet).
- Because generator functions are technically async (though no async
IO is involved), we lose the ability to use much of the Rust
standard library e.g. in builtins. This has led to many algorithms
being unrolled into iterative versions instead of iterator
combinations, and things like sorting had to be implemented from scratch.
- Many call sites that previously saw a `Result<..., ErrorKind>`
bubble up now only see the result value, as the error handling is
encapsulated within the generator loop.
This reduces number of places inside of builtin implementations
where error context can be attached to calls that can fail.
Currently what we gain in this tradeoff is significantly more
detailed span information (which we still need to bubble up, this
commit does not change the error display).
We'll need to do some analysis later of how useful the errors turn
out to be and potentially introduce some methods for attaching
context to a generator frame again.
This change is very difficult to do in stages, as it is very much an
"all or nothing" change that affects huge parts of the codebase. I've
tried to isolate changes that can be isolated into the parent CLs of
this one, but this change is still quite difficult to wrap one's mind
and I'm available to discuss it and explain things to any reviewer.
Fixes: b/238, b/237, b/251 and potentially others.
Change-Id: I39244163ff5bbecd169fe7b274df19262b515699
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8104
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Reviewed-by: Adam Joseph <adam@westernsemico.com>
Tested-by: BuildkiteCI
|
|
We currently send two warnings in case of detecting dead code - W008
inside compile_dead_code, and a more detailed warning in all places that
invoke compile_dead_code:
```
warning[W007]: useless operation on boolean: this expression is always false
--> /nix/store/qz3gjn95gazab4fkb7s8lm6hz17rdzzy-414z9nnj1wy66ymq6vgb693x9xjz6hf2-nixpkgs-src/pkgs/top-level/perl-packages.nix:12079:15
|
12079 | doCheck = false && !stdenv.isDarwin;
| ^^^^^^^^^^^^^^^^^^^^^^^^^
warning[W008]: this code will never be executed
--> /nix/store/qz3gjn95gazab4fkb7s8lm6hz17rdzzy-414z9nnj1wy66ymq6vgb693x9xjz6hf2-nixpkgs-src/pkgs/top-level/perl-packages.nix:12079:24
|
12079 | doCheck = false && !stdenv.isDarwin;
| ^^^^^^^^^^^^^^^^
```
The place invoking `compile_dead_code` has more context to why the code
is unused, so it's error message is much more useful.
Stop emitting the less informative warning inside compile_dead_code
(W008), and update the comment that we expect the caller to emit a
warning.
I kept W008 itself still around, in case we end up having places this
will get used again.
Change-Id: I2c5d84fc0cb4035872cd4b71cc3e9e34e120eb37
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8024
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
|
|
As applies are thunked, there was no situation where OpCall could be
emitted. In practice, all calls were already tail calls.
Change-Id: Id0d441dcdd86f804d7cddd0cc14f589bbfc75e5b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8147
Tested-by: BuildkiteCI
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
|
|
Because they do not use it, and it can not be passed with the coming
generator refactoring.
Change-Id: I0d96f2357a7ee79cd8a0f401583d4286230d4a6b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8146
Tested-by: BuildkiteCI
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
|
|
Change-Id: I4c02f0104c455ac00a3f299c1fbf75cbb08e8972
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8142
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Autosubmit: tazjin <tazjin@tvl.su>
|
|
Plain paths like `foo/bar.nix` are also allowed, so we can not
determine this based on the prefix.
The upstream PR that is referenced in a comment here has a
significantly different interface than we expected, so I'm not
touching that comment yet in this CL before I've had more time to
digest it.
Change-Id: Iea33bbb35de9c00a7d7fedf64d02253c75c1cc9e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8032
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: Alyssa Ross <hi@alyssa.is>
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
|
|
When resolving a select expression (`attrs.name` or `attrs.name or
default`), if the set compiles to a constant attribute set (as is most
notably the case with `builtins`) we can backtrack and replace that
attribute set directly with the compiled value.
For something like `builtins.length`, this will directly emit an
`OpConstant` that leaves the `length` builtin on the stack.
Change-Id: I639654e065a06e8cfcbcacb528c6da7ec9e513ee
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7957
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
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>
|
|
This makes it possible to inject builtins into the builtin set that
are written in Nix code, and which at runtime are represented by a
thunk that will compile them the first time they are used.
Change-Id: Ia632367328f66fb2f26cb64ae464f8f3dc9c6d30
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7891
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
These don't apply anymore since the "antidote-CL".
Change-Id: I40ee73ef43d44bbfc650a8fe6c2b33263dd06959
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7890
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
The codebase contains a lot of complexity and odd roundabout
handling for shadowing globals. I'm pretty sure none of this is
necessary, and all of it disappears if you simply make the globals
part of the ordinary identifier resolution chain, with their own
scope up above the root scope. Then the ordinary shadowing routines
do the right thing, and no special cases or new terminology are
required.
This commit does that.
Note by tazjin: This commit was originally abandoned when Adam decided
not to take away reviewer bandwidth for this at the time (eval was
still in a much earlier stage). As we've recently done some
significant refactoring of globals initialisation this came up again,
and it seems we can easily cover the use-cases of the poison tracking
in other ways now, so I've rebased, updated and resurrected the CL.
Co-Authored-By: Vincent Ambo <tazjin@tvl.su>
Signed-off-by: Adam Joseph <adam@westernsemico.com>
Change-Id: Ib3309a47a7b31fa5bf10466bade0d876b76ae462
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7089
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
This makes it possible for users to add additional context to an
error, which will then be rendered as an additional secondary span in
the formatted error output.
We should strive to do this basically anywhere errors are raised that
can occur multiple times, *especially* during type casts. This was
triggered by me debugging a type cast error attached to a fairly
large-ish span (a builtin invocation).
Change-Id: I51be41fabee00cf04de973935daf34fe6424e76f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7849
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
This is unnecessary, Rc already provides all the boxing we need.
Change-Id: I08cf0939c48da43f04c847526c7e5dae5336d528
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7749
Tested-by: BuildkiteCI
Reviewed-by: grfn <grfn@gws.fyi>
Reviewed-by: sterni <sternenseemann@systemli.org>
|
|
This is a somewhat terrifying hack that enables us to support
`builtins.builtins`, by running a "fake compilation" inside of a
suspended native thunk that can resolve the weak pointer to the
globals.
With this implementation, the thunk at `builtins.builtins` actually
resolves to the "real" `builtins` (verified with a new test).
This is kind of ugly, and it's something users shouldn't use, but
bubbling a warning out of this is difficult at the moment due to a
little bit of trickery with how the spans in suspended native thunks
work (they don't) (see b/237, b/238)
Change-Id: I67d0e93246dd5b279c960aeda00402031aa12af3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7748
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
This adds a mechanism to the compiler to compile an expression without
emitting any code. This allows for detected dead code to still be
compiled to detect errors & warnings inside of it.
Change-Id: Ie78479173570e9c819d8f32ae683ce34234a4c5d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7767
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
|
|
This optimiser can rewrite some expressions into more efficient forms,
and warn users about those cases.
As a proof-of-concept, only some simple boolean comparisons are
supported for now.
Change-Id: I7df561118cfbad281fc99523e859bc66e7a1adcb
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7766
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
This adds a very minimal amount of additional Rc-increments (~1 per
compilation), but makes it a lot easier to add an AST-optimising
compiler pass without incurring a lot of extra cost.
Change-Id: I57208bdfc8882e3ae21c5850e14aa380d3ccea36
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7765
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
|
|
This would make it possible to implement something like a linter based
on the tvix-eval compiler warnings.
Change-Id: I1feb4e7c4a44be7d1204b0a962ab522fd32b93c6
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7763
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
Previously the construction of globals (a compiler-only concept) and
builtins (a (now) user-facing API) was intermingled between multiple
different modules, and kind of difficult to understand.
The complexity of this had grown in large part due to the
implementation of `builtins.import`, which required the notorious
"knot-tying" trick using Rc::new_cyclic (see cl/7097) for constructing
the set of globals.
As part of the new `Evaluation` API users should have the ability to
bring their own builtins, and control explicitly whether or not impure
builtins are available (regardless of whether they're compiled in or
not).
To streamline the construction and allow the new API features to work,
this commit restructures things by making these changes:
1. The `tvix_eval::builtins` module is now only responsible for
exporting sets of builtins. It no longer has any knowledge of
whether or not certain sets (e.g. only pure, or pure+impure) are
enabled, and it has no control over which builtins are globally
available (this is now handled in the compiler).
2. The compiler module is now responsible for both constructing the
final attribute set of builtins from the set of builtins supplied
by a user, as well as for populating its globals (that is
identifiers which are available at the top-level scope).
3. The `Evaluation` API now carries a `builtins` field which is
populated with the pure builtins by default, and can be extended by
users.
4. The `import` feature has been moved into the compiler, as a
special case. In general, builtins no longer have the ability to
reference the "fix point" of the globals set.
This should not change any functionality, and in fact preserves minor
differences between Tvix/Nix that we already had (such as
`builtins.builtins` not existing).
Change-Id: Icdf5dd50eb81eb9260d89269d6e08b1e67811a2c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7738
Reviewed-by: sterni <sternenseemann@systemli.org>
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
It's been a while since the last time, so quite a lot of stuff has
accumulated here.
Change-Id: I0762827c197b30a917ff470fd8ae8f220f6ba247
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7597
Reviewed-by: grfn <grfn@gws.fyi>
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
This type carries the information required for calculating a
span (i.e. the chunk and offset), instead of the span itself. The span
is then only calculated in cases where it is required (when throwing
errors).
This reduces the eval time for
`builtins.length (builtins.attrNames (import <nixpkgs> {}))` by *one
third*!
The data structure in chunks that carries span information reduces
in-memory size by trading off the speed of retrieving span
information. This is because the span information is only actually
required when throwing errors (or emitting warnings).
However, somewhere along the way we grew a dependency on carrying span
information in thunks (for correctly reporting error chains). Hitting
the code paths for span retrieval was expensive, and carrying the
spans in a different way would still be less cache-efficient. This
change is the best tradeoff I could come up with.
Refs: b/229.
Change-Id: I27d4c4b5c5f9be90ac47f2db61941e123a78a77b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7558
Reviewed-by: grfn <grfn@gws.fyi>
Tested-by: BuildkiteCI
|
|
Instead of finding locals by doing 2x O(n) walks over the compiler's
locals list, use a secondary name-based index for resolving locals by
name.
Previously, almost 60% (!!) of eval time on some expressions over
nixpkgs was spent in `Local::has_name`. This function doesn't even
exist anymore now, and eval speed about doubles as a result.
Note that this doesn't exactly make the locals code easier to read,
but I'm also not sure what we can simplify in there in general.
This fixes b/227.
Change-Id: I29ce5eb9452b02d3b358c673e1f5cf8082e2fef9
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7560
Reviewed-by: grfn <grfn@gws.fyi>
Tested-by: BuildkiteCI
|
|
Change-Id: I595087eff943d38a9fc78a83d37e207bb2ab79bc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7443
Reviewed-by: grfn <grfn@gws.fyi>
Tested-by: BuildkiteCI
|
|
It is very confusing that this opcode is called DataLocalIdx, but it
carries a StackIdx rather than a LocalIdx. It seems like this
really ought to be called DataStackIdx, but maybe I've
misunderstood; if so please explain it to me.
Change-Id: I91f6ffa759412beef0b91d3c19ec0d873fe51b99
Signed-off-by: Adam Joseph <adam@westernsemico.com>
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7088
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
This commit contains two search-and-replace renames which are broken
out from I04131501029772f30e28da8281d864427685097f in order to
reduce the noise in that CL:
- `is_thunk -> is_suspended_thunk`, since there are now
OpThunkClosure and OpThunkSuspended
- `compile_lambda_or_thunk` -> `compile_lambda_or_suspension`
Change-Id: I7cc5bbb75ef6605e3428c7be27e812f41a10c127
Signed-off-by: Adam Joseph <adam@westernsemico.com>
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7037
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
CL/6867 added support for builtins.import, which required a cyclic
reference import->globals->builtins->import. This was implemented
using a RefCell, which makes it possible to mutate the builtins
during evaluation. The commit message for CL/6867 expressed a
desire to eliminate this possibility:
This opens up a potentially dangerous footgun in which we could
mutate the builtins at runtime leading to different compiler
invocations seeing different builtins, so it'd be nice to have
some kind of "finalised" status for them or some such, but I'm not
sure how to represent that atm.
This CL replaces the RefCell with Rc::new_cyclic(), making the
globals/builtins immutable once again. At VM runtime (once opcodes
start executing) everything is the same as before this CL, except
that the Rc<RefCell<>> introduced by CL/6867 is turned into an
rc::Weak<>.
The function passed to Rc::new_cyclic works very similarly to
overlays in nixpkgs: a function takes its own result as an argument.
However instead of laziness "breaking the cycle", Rust's
Rc::new_cyclic() instead uses an rc::Weak. This is done to prevent
memory leaks rather than divergence.
This CL also resolves the following TODO from CL/6867:
// TODO: encapsulate this import weirdness in builtins
The main disadvantage of this CL is the fact that the VM now must
ensure that it holds a strong reference to the globals while a
program is executing; failure to do so will cause a panic when the
weak reference in the builtins is upgrade()d.
In theory it should be possible to create strong reference cycles
the same way Rc::new_cyclic() creates weak cycles, but these cycles
would cause a permanent memory leak -- without either an rc::Weak or
RefCell there is no way to break the cycle. At some point we will
have to implement some form of cycle collection; whatever library we
choose for that purpose is likely to provide an "immutable strong
reference cycle" primitive similar to Rc::new_cyclic(), and we
should be able to simply drop it in.
Signed-off-by: Adam Joseph <adam@westernsemico.com>
Change-Id: I34bb5821628eb97e426bdb880b02e2097402adb7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7097
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
When compiling a lambda, take the name of the outer slot (if
available) and store it as the name on the lambda.
These names are then shown in the observer, and nowhere else (so far).
It is of course common for these things to thread through many
different context levels (e.g. `f = a: b: c: ...`), in this setup only
the outermost closure or thunk gains the name, but it's better than
nothing.
Change-Id: I681ba74e624f2b9e7a147144a27acf364fe6ccc7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7065
Reviewed-by: grfn <grfn@gws.fyi>
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
There are some rare scope cases with deferred access where this
doesn't behave correctly otherwise.
Change-Id: I6c774f5e62c1cb50b598026c54727017a52cd22d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7064
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
|
|
This check is now actually simply equivalent to checking whether the
target has been initialised or not.
Change-Id: I30660d11073ba313358f3a64234a90ed81abf74c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7062
Tested-by: BuildkiteCI
Reviewed-by: grfn <grfn@gws.fyi>
|
|
Checking the computed depth and stack slot against the computed depth
and stack slot is equivalent to just checking the indices into the
locals vector against each other (i.e. "is the slot we're compiling
into the slot we're accessing?")
Change-Id: Ie85a68df073e3b2e3d9aba7fe8634c48eada81fc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7059
Tested-by: BuildkiteCI
Reviewed-by: grfn <grfn@gws.fyi>
|
|
Failures to resolve a nix search path lookup in angle brackets can be
caught using tryEval (if it reaches the runtime). Resolving relative
paths (either to the current directory or the current user's home) can
never be caught, even if they happen inside a thunk at runtime (which is
currently the case for home-relative paths).
Change-Id: I7f73221df66d82a381dd4063358906257826995a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7025
Autosubmit: sterni <sternenseemann@systemli.org>
Reviewed-by: Adam Joseph <adam@westernsemico.com>
Tested-by: BuildkiteCI
Reviewed-by: grfn <grfn@gws.fyi>
|
|
This commit deduplicates the Thunk-like functionality from Closure
and unifies it with Thunk.
Specifically, we now have one and only one way of breaking reference
cycles in the Value-graph: Thunk. No other variant contains a
RefCell. This should make it easier to reason about the behavior of
the VM. InnerClosure and UpvaluesCarrier are no longer necessary.
This refactoring allowed an improvement in code generation:
`Rc<RefCell<>>`s are now created only for closures which do not have
self-references or deferred upvalues, instead of for all closures.
OpClosure has been split into two separate opcodes:
- OpClosure creates non-recursive closures with no deferred
upvalues. The VM will not create an `Rc<RefCell<>>` when executing
this instruction.
- OpThunkClosure is used for closures with self-references or
deferred upvalues. The VM will create a Thunk when executing this
opcode, but the Thunk will start out already in the
`ThunkRepr::Evaluated` state, rather than in the
`ThunkRepr::Suspeneded` state.
To avoid confusion, OpThunk has been renamed OpThunkSuspended.
Thanks to @sterni for suggesting that all this could be done without
adding an additional variant to ThunkRepr. This does however mean
that there will be mutating accesses to `ThunkRepr::Evaluated`,
which was not previously the case. The field `is_finalised:bool`
has been added to `Closure` to ensure that these mutating accesses
are performed only on finalised Closures. Both the check and the
field are present only if `#[cfg(debug_assertions)]`.
Change-Id: I04131501029772f30e28da8281d864427685097f
Signed-off-by: Adam Joseph <adam@westernsemico.com>
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7019
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
Validate "closed formals" (formal parameters without an ellipsis) via a
new ValidateClosedFormals op, which checks the arguments (in an attr set
at the top of the stack) against the formal parameters on the Lambda in
the current frame, and returns a new UnexpectedArgument error (including
the span of the formals themselves!!) if any arguments aren't allowed
Change-Id: Idcc47a59167a83be1832a6229f137d84e426c56c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7002
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
In preparation for both implementing the `functionArgs` builtin and
adding support for validating closed formals, record information about
the formal arguments to a function *on the Lambda itself*. This may seem
a little odd for the purposes of just closed formal checking, but is
something we have to have anyway for builtins.functionArgs so I figured
I'd do it this way to kill both birds with one stone.
Change-Id: Ie3770a607bf352a1eb395c79ca29bb25d5978cd8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7001
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
