feat invariant (#5868) - configure calldata fuzzed addresses dictionary #7240
Conversation
- added `FuzzDictionaryConfig.max_calldata_fuzz_dictionary_addresses` option to specify how many random addresses to generate and to randomly select from when fuzzing calldata. If option is not specified then current behavior applies - to narrow down number of runs / addresses involved in invariant test the `CalldataFuzzDictionaryConfig` is populated with random addresses plus all accounts from db (from `EvmFuzzState`) - added `fuzz_calldata_with_config` fn that accepts `Option<CalldataFuzzDictionary>` as param. Non invariants tests use existing `fuzz_calldata` fn and pass None as config arg
grandizzy
changed the title
grandizzy
force-pushed
the
config-calldata-fuzz
branch
from
5c05252
to
692adef
Compare
grandizzy
force-pushed
the
config-calldata-fuzz
branch
from
692adef
to
abe6c8b
Compare
grandizzy
changed the title
grandizzy
requested review from
DaniPopes,
Evalir and
mattsse
as code owners
| } | ||
|
|
||
| impl CalldataFuzzDictionaryConfig { | ||
| pub fn new(config: &FuzzDictionaryConfig, state: EvmFuzzState) -> Self { |
There was a problem hiding this comment.
added in b77dcb6
| use proptest::prelude::{BoxedStrategy, Strategy}; | ||
| use std::{fmt, sync::Arc}; | ||
|
|
||
| pub type CalldataFuzzDictionary = Arc<CalldataFuzzDictionaryConfig>; |
There was a problem hiding this comment.
I really disklike this alias style, this makes it much harder to navigate.
I know we have this in a few other places, but this isn't very helpful. I'd prefer a new wrapper type like
struct CalldataFuzzDictionary {
inner: Arc<CalldataFuzzDictionaryConfig>
}There was a problem hiding this comment.
done with b77dcb6 plus some code cleanup
|
Will this also allow users to specify the number of distinct senders? AFAIK Foundry currently uses an unlimited number of It should be nice to configure that as well |
correct, there is a new config |
done |
Motivation
Atm there's no way to configure calldata fuzzed params behavior - this could in particular be an issue when fuzzing addresses as there's little to no chance the same address is reused.
For invariant testing this is a bigger issue as one would often want to randomly reuse addresses when calling different handler selectors (see #5868). Some projects maintains an array of addresses in handlers and fuzz the indexes / prank with users from array of addresses - that's a little bit hard to mantain and cannot scale for too many addresses.
Solution
This PR introduce an additional config for the dictionary of addresses used in calldata fuzzing.
FuzzDictionaryConfig.max_calldata_fuzz_dictionary_addressesoption to specify how many random addresses to generate and to randomly select from when fuzzing calldata. If option is not specified then current behavior appliesCalldataFuzzDictionaryConfigis populated with random addresses plus all accounts from db (fromEvmFuzzState)fuzz_calldata_with_configfn that acceptsOption<CalldataFuzzDictionary>as param. Non invariants tests use existingfuzz_calldatafn and pass None as config argFurther improvements:
Todo: