Age | Commit message (Collapse) | Author | Files | Lines |
|
This replaces the OpCode enum with a new Op enum which is guaranteed to fit in a
single byte. Instead of carrying enum variants with data, every variant that has
runtime data encodes it into the `Vec<u8>` that a `Chunk` now carries.
This has several advantages:
* Less stack space is required at runtime, and fewer allocations are required
while compiling.
* The OpCode doesn't need to carry "weird" special-cased data variants anymore.
* It is faster (albeit, not by much). On my laptop, results consistently look
approximately like this:
Benchmark 1: ./before -E '(import <nixpkgs> {}).firefox.outPath' --log-level ERROR --no-warnings
Time (mean ± σ): 8.224 s ± 0.272 s [User: 7.149 s, System: 0.688 s]
Range (min … max): 7.759 s … 8.583 s 10 runs
Benchmark 2: ./after -E '(import <nixpkgs> {}).firefox.outPath' --log-level ERROR --no-warnings
Time (mean ± σ): 8.000 s ± 0.198 s [User: 7.036 s, System: 0.633 s]
Range (min … max): 7.718 s … 8.334 s 10 runs
See notes below for why the performance impact might be less than expected.
* It is faster while at the same time dropping some optimisations we previously
performed.
This has several disadvantages:
* The code is closer to how one would write it in C or Go.
* Bit shifting!
* There is (for now) slightly more code than before.
On performance I have the following thoughts at the moment:
In order to prepare for adding GC, there's a couple of places in Tvix where I'd
like to fence off certain kinds of complexity (such as mutating bytecode, which,
for various reaons, also has to be part of data that is subject to GC). With
this change, we can drop optimisations like retroactively modifying existing
bytecode and *still* achieve better performance than before.
I believe that this is currently worth it to pave the way for changes that are
more significant for performance.
In general this also opens other avenues of optimisation: For example, we can
profile which argument sizes actually exist and remove the copy overhead of
varint decoding (which does show up in profiles) by using more adequately sized
types for, e.g., constant indices.
Known regressions:
* Op::Constant is no longer printing its values in disassembly (this can be
fixed, I just didn't get around to it, will do separately).
Change-Id: Id9b3a4254623a45de03069dbdb70b8349e976743
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12191
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
Per https://nnethercote.github.io/perf-book/hashing.html, we have
basically no reason to use the default hasher over a faster,
non-DoS-resistant hasher. This gives a nice perf boost basically for
free:
hello outpath time: [704.76 ms 714.91 ms 725.63 ms]
change: [-7.2391% -6.1018% -4.9189%] (p = 0.00 < 0.05)
Performance has improved.
Change-Id: If5587f444ed3af69f8af4eead6af3ea303b4ae68
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12046
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Reviewed-by: Ilan Joselevich <personal@ilanjoselevich.com>
Autosubmit: aspen <root@gws.fyi>
|
|
This was made unnecessary in c92d06271 (feat(tvix/eval): drop
LightSpan::Delayed, 2023-12-08) because it didn't improve benchmarks as
much as expected and has been vestigial since; this continues the
cleanup by just removing it altogether
Change-Id: I21ec7ae9b52a5cccd2092696a5a87f658194d672
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11949
Autosubmit: aspen <root@gws.fyi>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
This commit implements deep_force() nonrecursively, by maintaining
an explicit stack rather than using the call stack for recursion.
As an added bonus, we don't need to pass around the SharedThunkSet
anymore, and can in fact completely eliminate SharedThunkSet.
Change-Id: I7c4f59f37834d451a28bf6be317eb0a90eac4ee6
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10252
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: Adam Joseph <adam@westernsemico.com>
|
|
The comment explaining ThunkSet makes it seem like it does the same
think as ThunkRepr::Blackhole. In fact neither one is a substitute
for the other. Let's explain the difference.
Change-Id: I89ceaaa9d3c499edbc7d48f70ca5d11f97666c43
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10250
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: Adam Joseph <adam@westernsemico.com>
|
|
This commit rewrites Value::nix_eq() into an equivalent. Except for
calls to Thunk::force(), the new form no longer uses generators, and
is async only because of the fact that it calls Thunk::force().
I believed that the nonrecursive form would be faster. It is, in
fact, slightly slower. I believe this is due to the vec![]
allocation; I am investigating.
Prev-Nixpkgs-Benchmark: {"attrpath":"pkgsCross.aarch64-multiplatform.hello.outPath","peak-kbytes":"459068","system-seconds":"0.71","user-seconds":"5.39"}
This-Nixpkgs-Benchmark: {"attrpath":"pkgsCross.aarch64-multiplatform.hello.outPath","peak-kbytes":"460048","system-seconds":"0.68","user-seconds":"5.73"}
Change-Id: I10f4868891e4b7475df13f0cbc41ec78dd985dd8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10118
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Autosubmit: Adam Joseph <adam@westernsemico.com>
|
|
This commit rewrites Thunk::force() so that it is not (directly)
self-recursive. It maintains a Vec of all the
previously-encountered thunks which point to the one it is currently
forcing, rather than recursively calling itself.
Benefits:
- Short term:
This commit saves the cost of a round-trip through the generator
machinery for the generators::request_force() which is removed by
this commit.
- Medium term:
Once a similar transformation has been applied to nix_cmp(),
nix_add(), nix_eq(), and coerce_to_string(), those four functions,
along with Thunk::force(), will make non-tail calls only to each
other. They can then be merged into a single tail-recursive
function which does not use the generator machinery at all:
enum Task { Cmp, Add, Eq, CoerceToString, Force};
fn Value::walk(task:Task, v1:Value, v2:Value) {
// ...
- Long term:
The long-term goal here is to use generators **only for builtins**
and [Marionette]-style remote control of the VM. In other words:
use `async` for things that actually involve concurrency. Calls
from the VM to builtins can then be blocking calls, because even
cppnix will overflow the stack if you make a MAX_STACK_DEPTH-deep
recursive call which passes through a builtin at every stack frame
(e.g. `{ func = builtins.sort (a: b: ... func ...) ...}`).
This way the inner "tight loop" of the interpreter doesn't pay the
costs of `async` and generators. These costs manifest in terms
of: performance, complex nonlocal control flow, and language
impediments (async Rust is a restricted subset of real Rust, and
is missing things like traits).
[Marionette]: https://firefox-source-docs.mozilla.org/testing/marionette/Intro.html
Change-Id: I6179b8abb2ea0492180fcb347f37595a14665777
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10039
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
Change-Id: I4eab5c81fb82337da06327248845cd2f3a4490d3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10038
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Autosubmit: Adam Joseph <adam@westernsemico.com>
|
|
This commit adds Thunk::unwrap_or_clone(), which uses
Rc::try_unwrap() to avoid cloning the Value out of a an Rc which has
only one strong reference.
Change-Id: Icacefe0c823dcddf046d90c0c5cd5ed59fe976d4
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10037
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: Adam Joseph <adam@westernsemico.com>
Tested-by: BuildkiteCI
|
|
mapAttrs, map and genList call Nix functions provided by the caller and
store the result of applying them in a Nix data structure that does not
force all of its contents when forced itself. This means that when such
a builtin application is forced, the Nix function calls performed by the
builtin should not be forced: They may be forced later, but it is also
possible that they will never be forced, e.g. in
builtins.length (builtins.map (builtins.add 2) [ 1 2 3 ])
it is not necessary to compute a single application of builtins.add.
Since request_call_with immediately performs the function call
requested, Tvix would compute function applications unnecessarily before
this change. Because this was not followed by a request_force, the
impact of this was relatively low in Nix code (most functions return a
new thunk after being applied), but it was enough to cause a lot of
bogus builtins.trace applications when evaluating anything from
`lib.modules`. The newly added test includes many cases where Tvix
previously incorrectly applied a builtin, breaking a working expression.
To fix this we add a new helper to construct a Thunk performing a
function application at runtime from a function and argument given as
`Value`s. This mimics the compiler's compile_apply(), but does itself
not require a compiler, since the necessary Lambda can be constructed
independently.
I also looked into other builtins that call a Nix function to verify
that they don't exhibit such a problem:
- Many builtins immediately use the resulting value in a way that makes
it necessary to compute all the function calls they do as soon as
the outer builtin application is forced:
* all
* any
* filter
* groupBy
* partition
- concatMap needs to (shallowly) force the returned list for
concatenation.
- foldl' is strict in the application of `op` (I added a comment that
makes this explicit).
- genericClosure needs to (shallowly) force the resulting list and some
keys of the attribute sets inside.
Resolves b/272.
Change-Id: I1fa53f744bcffc035da84c1f97ed25d146830446
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8651
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
Change-Id: Ie4c563e933f571f45cb4f4efe650d1b65f119e8d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8324
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
Autosubmit: tazjin <tazjin@tvl.su>
|
|
This reports the span
1. of the code within a thunk,
2. of the place where the thunk was instantiated,
3. of the place where the thunk was first forced,
4. of the place where the thunk was forced again,
when yielding an infinite recursion error, which hopefully makes it
easier to debug them.
The spans are tracked in the ThunkRepr::Blackhole variant when putting
a thunk under evaluation.
Note that we currently have some loss of span precision in the VM loop
when switching between frame types, so spans 3/4 are currently a bit
wonky. Working on it.
Change-Id: Icbd2a9df903d00e8c2545b3fc46dcd2a9e3e3e55
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8270
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Autosubmit: tazjin <tazjin@tvl.su>
|
|
This is step 1 towards being able to use all 4 spans that we know when
dealing with infinite recursion. It tracks the span at which the
force of a thunk was first requested when constructing a blackhole, so
that we can highlight the spans of the first and second forces.
These are actually the least relevant spans, but the easiest to put in
place, more coming soon.
Change-Id: I4c7e82f6211b98756439d4148a4191457cc46807
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8269
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
|
|
Given Rust's current lack of support for tail calls, we cannot avoid
using `async` for builtins. This is the only way to avoid
overflowing the cpu stack when we have arbitrarily deep
builtin/interpreted/builtin/interpreted/... "sandwiches"
There are only five `async fn` functions which are not builtins
(some come in multiple "flavors"):
- add_values
- resolve_with
- force, final_deep_force
- nix_eq, nix_cmp_eq
- coerce_to_string
These can be written iteratively rather than recursively (and in
fact nix_eq used to be written that way!). I volunteer to rewrite
them. If written iteratively they would no longer need to be
`async`.
There are two motivations for limiting our reliance on `async` to
only the situation (builtins) where we have no other choice:
1. Performance.
We don't really have any good measurement of the performance hit
that the Box<dyn Future>s impose on us. Right now all of our
large (nixpkgs-eval) tests are swamped by the cost of other
things (e.g. fork()ing `nix-store`) so we can't really measure
it. Builtins tend to be expensive operations anyways
(regexp-matching, sorting, etc) that are likely to already cost
more than the `async` overhead.
2. Preserving the ability to switch to `musttail` calls.
Clang/LLVM recently got `musttail` (mandatory-elimination tail
calls). Rust has refused to add this mainly because WASM doesn't
support, but WASM `tail_call` has been implemented and was
recently moved to phase 4 (standardization). It is very likely
that Rust will get tail calls sometime in the next year; if it
does, we won't need async anymore. In the meantime, I'd like to
avoid adding any further reliance on `async` in places where it
wouldn't be straightforward to replace it with a tail call.
https://reviews.llvm.org/D99517
https://github.com/WebAssembly/proposals/pull/157
https: //github.com/rust-lang/rfcs/issues/2691#issuecomment-1462152908
Change-Id: Id15945d5a92bf52c16d93456e3437f91d93bdc57
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8290
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Autosubmit: Adam Joseph <adam@westernsemico.com>
|
|
This drops the usage of serde::Serialize, as the trait can not be used
to implement the correct semantics (function colouring!).
Instead, a manual JSON serialisation function is written which
correctly handles toString, outPath and other similar weirdnesses.
Unexpectedly, the eval-okay-tojson test from the C++ Nix test suite
now passes, too.
This fixes an issue where serialising data structures containing
derivations to JSON would fail.
Change-Id: I5c39e3d8356ee93a07eda481410f88610f6dd9f8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8209
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
|
|
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>
|
|
This is a ThunkSet wrapped to be shareable, which will be required
once ThunkSets are embedded in futures.
Change-Id: I5a067b7972ac86e4d354c75ef05c86b2284c1137
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8144
Tested-by: BuildkiteCI
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
|
|
Change-Id: Iea248870a0ea5d38cb02ff059c968fbd563570b6
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8143
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Tested-by: BuildkiteCI
|
|
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>
|
|
This adds addresses of thunk and closure chunks to the debug output
displayed when dumping bytecode.
This makes it possible to see in the dump which thunks are referenced
by constants in other thunks.
Change-Id: I2c98de5227e7cb415666cd3134c947a56979dc80
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8137
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
|
|
This CL removes redundant clone from value which is
going to be dropped without further use.
Change-Id: Ibd2a724853c5cfbf8ca40bf0b3adf0fab89b9be5
Signed-off-by: Aaqa Ishtyaq <aaqaishtyaq@gmail.com>
Reviewed-on: https://cl.tvl.fyi/c/depot/+/8125
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
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 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>
|
|
Instead of having a representation of suspended native thunks that
involves constructing a fake code chunk, make these thunks a
first-class part of the internal thunk representation.
The previous code was not that simple to understand, and actually
contained a critical bug which could lead to Tvix crashes. This
version fixes the particular instance of that bug, but instead
uncovers another (b/238) which can still lead to Tvix crashes.
Fixes: b/237.
Change-Id: I771d03864084d63953bdbb518fec94487481f839
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7750
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
|
|
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>
|
|
Implements `Serialize` for `tvix_eval::Value`. Special care is taken
with serialisation of attribute sets, and forcing of thunks.
The tests should cover both cases well.
Change-Id: I9bb135bacf6f87bc6bd0bd88cef0a42308e6c335
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7803
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Autosubmit: tazjin <tazjin@tvl.su>
|
|
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
|
|
Introduces continuation-passing-based trampolining of thunk forcing to
avoid recursing when forcing deeply nested expressions.
This is required for evaluating large expressions.
This change was extracted out of cl/7362.
Co-authored-by: Vincent Ambo <tazjin@tvl.su>
Co-authored-by: Griffin Smith <grfn@gws.fyi>
Change-Id: Ifc1747e712663684b2fff53095de62b8459a47f3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7551
Reviewed-by: grfn <grfn@gws.fyi>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
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
|
|
Having thunks which, when forced, execute native Rust code rather
than interpreted opcodes lets us avoid having to bundle
`src/libexpr/primops/derivation.nix` like cppnix does by implementing
it in Rust instead.
Change-Id: If91d77a6736234321eee87ba4b4777eed5a3fe1c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7450
Reviewed-by: grfn <grfn@gws.fyi>
Tested-by: BuildkiteCI
|
|
Change-Id: I0fa069fbeff6718a765ece948c2c1bce285496f7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7449
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
|
|
This passes all the function/thunk-pointer-equality tests in
cl/7369.
Change-Id: Ib47535ba2fc77a4f1c2cc2fd23d3a879e21d8b4c
Signed-off-by: Adam Joseph <adam@westernsemico.com>
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7358
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
See cl/7372; Nix equality semantics require the ability to track
pointer equality of upvalue-sets.
Signed-off-by: Adam Joseph <adam@westernsemico.com>
Change-Id: I82ba517499cf370189a80355e4e46a5caaab7153
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7373
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
Change-Id: I3b1284e28c350bfed84d643ae7f922f3487e1f2a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7355
Autosubmit: Adam Joseph <adam@westernsemico.com>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
When passing multiple arguments, every intermediate callable needs to
be forced as this is expected by the VM's call_value function.
Also adds a debug assertion for this which makes it easier to spot
exactly what went wrong.
Change-Id: I3aa519cb6cdaab713bd18282bef901c4cd77c535
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7312
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
|
|
It isn't possible to implement PartialEq properly for Value, because
any sensible implementation needs to force() thunks, which cannot be
done without a `&mut VM`.
The existing derive(PartialEq) has false negatives, which caused the
bug which cl/7142 fixed. Fortunately that bug was easy to find, but
a silent false negative deep within the bowels of nixpkgs could be a
real nightmare to hunt down.
Let's just remove the PartialEq impl for Value, and the other
derive(PartialEq)'s that depend on it.
Signed-off-by: Adam Joseph <adam@westernsemico.com>
Change-Id: Iacd3726fefc7fc1edadcd7e9b586e04cf8466775
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7144
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
Using the same method as in Thunk::deep_force, detect cycles when
printing values by maintaining a set of already seen thunks.
With this, display of infinite values matches that of Nix:
> nix-instantiate --eval --strict -E 'let as = { x = 123; y = as; }; in as'
{ x = 123; y = { x = 123; y = <CYCLE>; }; }
> tvix-eval -E 'let as = { x = 123; y = as; }; in as'
=> { x = 123; y = { x = 123; y = <CYCLE>; }; } :: set
Change-Id: I007b918d5131d82c28884e46e46ff365ef691aa8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7056
Tested-by: BuildkiteCI
Reviewed-by: grfn <grfn@gws.fyi>
|
|
This is done via a new `deepForce` function on Value. Since values can
be cyclical (for example, see the test-case), we need to do some extra
work to avoid RefCell borrow errors if we ever hit a graph cycle:
While deep-forcing values, we keep a set of thunks that we have
already seen and avoid doing any work on the same thunk twice. The set
is encapsulated in a separate type to stop potentially invalid
pointers from leaking out.
Finally, since deep_force is conceptually similar to
`VM::force_for_output` (but more suited to usage in eval since it
doesn't clone the values) this removes the latter, replacing it with
the former.
Co-Authored-By: Vincent Ambo <tazjin@tvl.su>
Change-Id: Iefddefcf09fae3b6a4d161a5873febcff54b9157
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7000
Tested-by: BuildkiteCI
Reviewed-by: grfn <grfn@gws.fyi>
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
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>
|
|
Currently, the span on *all* thunk force errors is the span at which the
thunk is forced, which for recursive thunk forcing ends up just being
the same span over and over again. This changes the span on thunk force
errors to be the span at which point the thunk is *created*, which is a
bit more helpful (though the printing atm is a little... crowded). To
make this work, we have to thread through the span at which a thunk is
created into a field on the thunk itself.
Change-Id: I81474810a763046e2eb3a8f07acf7d8ec708824a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6932
Autosubmit: grfn <grfn@gws.fyi>
Reviewed-by: Adam Joseph <adam@westernsemico.com>
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
|
|
Previously the various call functions either returned `EvalResult<()>`
or `EvalResult<Value>`, which was confusing.
Now only vm::call_with returns a Value directly, and other parts of
the API just leave the stack top in the post-call state.
This makes it easier to reason about what's going on in non-tail-call
cases (which are making a comeback).
Change-Id: I264ffc683a11aca72dd06e2220a5ff6e7c5fc2b0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6936
Tested-by: BuildkiteCI
Reviewed-by: grfn <grfn@gws.fyi>
|
|
Using rust's PartialEq trait to implement Nix equality semantics is
reasonably fraught with peril, both because the actual laws are
different than what nix expects, and (more importantly) because certain
things actually require extra context to compare for equality (for
example, thunks need to be forced). This converts the manual PartialEq
impl for Value (and all its descendants) to a *derived* PartialEq
impl (which requires a lot of extra PartialEq derives on miscellanious
other types within the codebase), and converts the previous
nix-semantics equality comparison into a new `nix_eq` method. This
returns an EvalResult, even though it can't currently return an error,
to allow it to fail when eg forcing thunks (which it will do soon).
Since the PartialEq impls for Value and NixAttrs are now quite boring,
this converts the generated proptests for those into handwritten ones
that cover `nix_eq` instead
Change-Id: If3da7171f88c22eda5b7a60030d8b00c3b76f672
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6650
Autosubmit: grfn <grfn@gws.fyi>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
|
|
This struct will be responsible for tracking upvalues (and is a
convenient place to introduce optimisations for reducing value clones)
instead of a plain value vector.
The main motivation for this is that the upvalues will have to capture
the `with`-stack fully and I want to avoid duplicating the logic for
this between the two capturing types.
Change-Id: I6654f8739fc2e04ca046e6667d4a015f51724e99
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6485
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
|
|
At the point where control flow exits Thunk::force (which may be due
to recursing), it is vital that there is no longer a borrow to the
inner thunk representation, otherwise this can cause accidental
infinite recursion (which will be detected, but cause failures on
valid code).
Change-Id: I2846f3142830ae3110a4f5d2299e9d7928634504
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6436
Reviewed-by: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
|
|
This could occur when the disassembler is enabled and tracing the
runtime while a thunk is being evaluated, as it would not be possible
for the *tracer* to borrow the thunk at this exact moment.
However, we know that if the borrowing fails here we are dealing with
a not-fully evaluated thunk (blackhole), which should just print the
internal representation.
Change-Id: I4bdb4f17818d55795368e3d28842048f488f0a91
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6416
Reviewed-by: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
|
|
Previously, "calling" (setting up the VM run loop for executing a call
frame) and "running" (running this loop to completion) were separate
operations.
This was basically an attempt to avoid nesting `VM::run` invocations.
However, doing things this way introduced some tricky bugs for exiting
out of the call frames of thunks vs. builtins & closures.
For now, we unify the two operations and always return the value to
the caller directly. For now this makes calls a little less effective,
but it gives us a chance to nail down some other strange behaviours
and then re-optimise this afterwards.
To make sure we tackle this again further down I've added it to the
list of known possible optimisations.
Change-Id: I96828ab6a628136e0bac1bf03555faa4e6b74ece
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6415
Reviewed-by: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
|
|
With this, if an error occurs while forcing a thunk (which is very
likely) it is threaded through to the top by wrapping it in the
ErrorKind::ThunkForce variant.
We could use this to generate "stacktrace-like" error output if we
wanted, or simply jump through and discard everything except the
innermost error.
Change-Id: I3c1c8708c2f73ae062815adf490ce935b1979da8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6409
Reviewed-by: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
|
|
Previously error spans were optional because the information about
code spans was not available at runtime. Now that this information has
been added, the error type will always carry a span.
This change is very invasive all throughout the codebase. This is due
to the fact that many functions that are called *by* the VM expected
to return `EvalResult`, but this no longer works as the span
information is not available to those functions - only to the VM
itself.
To work around this the majority of these functions have been changed
to return `Result<T, ErrorKind>` instead and an accompanying macro in
the VM constructs the "real" error.
Note that this implementatino currently has a bug where errors
occuring within thunks will yield the location at which the thunk was
forced, not the location at which the error occured within the code.
This will be fixed soon, but the commit is large enough as is.
Change-Id: Ib1ecb81a4d09d464a95ea7ea9e589f3bd08d5202
Reviewed-on: https://cl.tvl.fyi/c/depot/+/6408
Reviewed-by: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
|