This way, we don't need the PathDisplaySourceAccessor source accessor
hack, since error messages are produced directly by the original
source accessor.
In fact, we don't even need to copy the inputs to the store at all, so
this gets us very close to lazy trees. We just need to know the store
path so that requires hashing the entire input, which isn't lazy. But
the next step will be to use a virtual store path that gets rewritten
to the actual store path only when needed.
This makes paths in error messages behave similar to lazy-trees,
e.g. instead of store paths like
error: attribute 'foobar' missing
at /nix/store/ddzfiipzqlrh3gnprmqbadnsnrxsmc9i-source/machine/configuration.nix:209:7:
208|
209| pkgs.foobar
| ^
210| ];
you now get
error: attribute 'foobar' missing
at /home/eelco/Misc/eelco-configurations/machine/configuration.nix:209:7:
208|
209| pkgs.foobar
| ^
210| ];
The underlying issue is that debugger code path was
calling PosTable::operator[] in each eval method.
This has become incredibly expensive since 5d9fdab3de.
While we are it it, I've reworked the code to
not use std::shared_ptr where it really isn't necessary.
As I've documented in previous commits, this is actually
more a workaround for recursive header dependencies now
and is only necessary in `error.hh` code.
Some ad-hoc benchmarking:
After this commit:
```
Benchmark 1: nix eval nixpkgs#hello --impure --ignore-try --no-eval-cache --debugger
Time (mean ± σ): 784.2 ms ± 7.1 ms [User: 561.4 ms, System: 147.7 ms]
Range (min … max): 773.5 ms … 792.6 ms 10 runs
```
On master 3604c7c51:
```
Benchmark 1: nix eval nixpkgs#hello --impure --ignore-try --no-eval-cache --debugger
Time (mean ± σ): 22.914 s ± 0.178 s [User: 18.524 s, System: 4.151 s]
Range (min … max): 22.738 s … 23.290 s 10 runs
```
(cherry picked from commit adbd08399c)
Progress on #5638
There are still a global fetcher and eval settings, but they are pushed
down into `libnixcmd`, which is a lot less bad a place for this sort of
thing.
Continuing process pioneered in
52bfccf8d8.
This is not really part of the evaluator: it is just an integration
between Boehm GC and Boost coroutines usable for any purpose. The
evaluator (merely) optionally uses it.
After the removal of the InputAccessor::fetchToStore() method, the
only remaining functionality in InputAccessor was `fingerprint` and
`getLastModified()`, and there is no reason to keep those in a
separate class.
This missing GC root wasn't much of a problem before, because the
heap would end up with a reference to the `baseEnv` pretty soon,
but when unit testing, the construction of `EvalState` doesn't
necessarily happen well before GC runs for the first time.
Found while unit testing the Rust bindings that currently reside
at https://github.com/nixops4/nixops4/tree/main/rust
At this point many features are stripped out, but this works:
- Can run libnix{util,store,expr} unit tests
- Can run some Nix commands
Co-Authored-By volth <volth@volth.com>
Co-Authored-By Brian McKenna <brian@brianmckenna.org>
