* Experimental performance metering
* Average on repeated runs with `--repeat` option
* Add "version" field to the report
The version is got from `air/Cargo.toml`.
* Allow disabling preparing binaries
with the `--no-prepare-binaries` option.
* Human-readable execution time in the report
* Add dashboard benchmark
* Human-readable text report
* Use CID values for tetraplets and `canon` vectors.
* Rename `cid_store` to `value_store`
It is consistent with the new `tetraplet_store` and `canon_store`
fields.
* Make canon data more typeful
The `CanonResult` doesn't take a JSON value anymore that is further
deserialized elsewhere, but is a struct that has all data deserialized.
* Typeful `CID` type
The `CID` type has a phantom type paramter defining its value's type.
* Group cid stores and trackers
Group cid stores into `CidInfo` struct, and trackers into `ExecutionCidState` struct.
This PR intended to reduce stream usage scope in order to make AquaVM truly deterministic.
Refs: #297.
Co-authored-by: Ivan Boldyrev <ivan@fluence.one>
Co-authored-by: Anatoly Laskaris <github_me@nahsi.dev>
At the moment `Ap` state contains vector of generations, although the ap instruction itself allow to specify only one stream. It was done to support several streams in a result position in a future. But the current realization allows to handle empty vector that was implemented to support states for scalars and it turned out (for more info see #326) that states for scalars aren't actually needed.
Closes#355.
This PR refactors call merger of `TraceHandler`. Previously it requires `CallOutputValue` to determine a type of call output value (stream or scalar). And internally it checked correspondence between data result and call output type and return a error if they are not equal. Although execution engine component also had a match over these values and does nothing if they are not matched since `TraceHandler` did this job. This PR eliminate such behaviour and improve isolation of AquaVM modules.
feat(execution-engine): change behaviour of fold over streams
Change behaviour of fold over streams to make it more similar to pi-calculus channels/names (for more info see #333).
Closes#333.
BREAKING CHANGE:
The new stream behaviour is not compatible with old one, such as
```
(fold $stream iterator
(seq
(call ...)
(next iterator)))
```
will never end after this change (for more info again see #333).
* Get rid of unsafe code unless really necessary
* Add lint levels where appropriate
Some crates (air-beautifier, air-testing-framework) have lot of
rust_2018_idioms violations, that will be resolved later.
1. Introduce --runner-tracing-params for enable and disable tracing of runner.
2. Adjust some function's tracing levels to make default traces less noisy.
`air-trace run` subcommand allows to run AquaVM on any data, it allows to define most AquaVM inputs, providing defaults for most of them, and sets up either human-readable or JSON tracing output, the latter can be later processed by `air-trace stats`.
Anomaly data input is also supported, that is useful for slow data investigation.
Native execution mode can be used for native profiling. Please note, however, that current version cannot be built natively on Apple Sillicon processor yet, as invariably depends on the `avm-server` because of leaking types that should be refactored or hidden. The `--repeat` option can repeat the execution several times for the execution to dominate on input data reading and output.
High-level or rare calls have "info" trace level, instructions are "debug", and sub-instruction calls are "tracing". Over-detailed tracing can induce overhead that spoils timing data.
The `log_instruction` macro unconditionally pre-formatted some string to
be logged, even if logging is disabled, making whole AquaVM very slow.
After getting it fixed, native execution time dropped from 5sec to
thousands of milliseconds.
Many internal structures refer to trace positions; this is an important type of values. In the code, it is sometimes `u32`, sometimes `usize`. While such variables usually have "_pos" suffix, having a dedicated type does improve code with type guarantees.
Intially, a fold instruction was designed based on idea that its iterable can't expand,
so AIR was considered as a subtree. But after introducing recursive streams in #225, it's
no more a subtree, but a subgraph.
