draft - memory fuzzer for venom #4686
Conversation
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #4686 +/- ##
==========================================
+ Coverage 92.99% 93.09% +0.09%
==========================================
Files 131 131
Lines 19094 19119 +25
Branches 3324 3329 +5
==========================================
+ Hits 17757 17799 +42
+ Misses 899 878 -21
- Partials 438 442 +4 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
this seems to be working somewhat, but i get a bunch of OVERRUN messages (in debug mode), which seems to substantially slow down example generation: @cyberthirst could you please look into this? |
| source = draw(st.sampled_from(reachable_blocks)) | ||
|
|
||
| # we have already visited it | ||
| if source in cfg: |
There was a problem hiding this comment.
i think this might be a budget sink - i think that most of reachable blocks will already be inside the cfg. so with increasing number of reachable, the lower the probability of hitting the block which isn't already embedded
i think smth like this might make more sense
diff --git a/tests/functional/venom/test_memory_fuzzer.py b/tests/functional/venom/test_memory_fuzzer.py
index ec696f712..7a4446295 100644
--- a/tests/functional/venom/test_memory_fuzzer.py
+++ b/tests/functional/venom/test_memory_fuzzer.py
@@ -267,11 +267,12 @@ def control_flow_graph(draw, basic_blocks):
reachable_blocks = [basic_blocks[0]]
while remaining_blocks:
- source = draw(st.sampled_from(reachable_blocks))
+ # Only sample from reachable blocks that aren't already in cfg
+ unprocessed_reachable = [b for b in reachable_blocks if b not in cfg]
+ if not unprocessed_reachable:
+ break
- # we have already visited it
- if source in cfg:
- continue
+ source = draw(st.sampled_from(unprocessed_reachable))
target = draw(st.sampled_from(remaining_blocks))
|
we can monkeypatch some of the internal constants: https://hypothesis.readthedocs.io/en/latest/reference/internals.html#hypothesis.internal.conjecture.engine.BUFFER_SIZE however, i'm still getting overruns, but of a new form: |
this is the number of overruns with/without the BUFFER_SIZE patch. both versions also use the reachability patch. however, the increased size also increases the test's runtime.
|
|
to reduce the choices, we can make the constants MAX_BASIC_BLOCKS, MAX_INSTRUCTIONS_PER_BLOCK smaller - i think 25 bbs is still reasonable. further we can try to batch various choices into one - e.g. instead of multiple boolean choices in one function, we could do one integer choice and then do weighted if statements - like |
| from vyper.venom.passes import ( | ||
| AssignElimination, | ||
| CFGNormalization, | ||
| DeadStoreElimination, | ||
| LoadElimination, | ||
| MakeSSA, | ||
| MemMergePass, | ||
| SimplifyCFGPass, | ||
| SingleUseExpansion, | ||
| ) |
Check notice
Code scanning / CodeQL
Unused import Note test
| # First, handle output to allocate variable if needed | ||
| output_sym = None | ||
| if inst.output and isinstance(inst.output, SymbolicVar): | ||
| output_sym = inst.output |
Check warning
Code scanning / CodeQL
Variable defined multiple times Warning test
| # All blocks except entry (to prevent back edges to entry) | ||
| non_entry_blocks = basic_blocks[1:] | ||
|
|
||
| # create a spanning tree to ensure all blocks are reachable |
There was a problem hiding this comment.
we have back edges so this isn't creating a spanning tree?
| other_target, target = target, other_target | ||
| block_types[source] = _BranchBB(target1=target, target2=other_target) | ||
|
|
||
| # classify remaining blocks that were not handled during spanning |
There was a problem hiding this comment.
how can some nodes be missing if we created a "spanning tree"?
2 participants
What I did
How I did it
How to verify it
Commit message
Commit message for the final, squashed PR. (Optional, but reviewers will appreciate it! Please see our commit message style guide for what we would ideally like to see in a commit message.)
Description for the changelog
Cute Animal Picture