refactor: optimize calldatasize check #3104
Conversation
Addresses the first half of inefficient selector table checks (issue vyperlang#3037) by only running a check for calldatasize >= 4 if a zero selector function is present. (wip: currently checks on all selectors which contain a 0)
|
I figured out why my zero selector function wasn't getting pulled (and why I hacked it by finding any zero instead of trailing zeroes): it's because str(f._metadata["signature"]) is not pulling the abi_signature, just the signature. So for example, "def BlockHashAskewLimitary():" is getting hashed instead of "BlockHashAskewLimitary". Currently working on finding the correct function in the codebase to convert it to an abi signature. |
fixed by passing the correct abi_sig instead of a FunctionSignature object. now correctly generates method_id and identifies zero selector functions (any function where selector starts with a 0)
Codecov Report
@@ Coverage Diff @@
## master #3104 +/- ##
=======================================
Coverage 88.43% 88.44%
=======================================
Files 88 88
Lines 11046 11049 +3
Branches 2338 2339 +1
=======================================
+ Hits 9769 9772 +3
Misses 822 822
Partials 455 455
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
…lang#3037) changed import and associated function calls updated check for zero selector functions removed/updated comments
…g#3037) Only performs a calldatasize == 4 check when there are trailing 0s in selector wip: figure out whether to check hex, int, or bytecode of selector
streamlined creation of runtime list in _runtime_ir()
some stylistic changes
|
This pull request introduces 1 alert when merging b8ba472 into b13595f - view on LGTM.com new alerts:
|
fubuloubu
changed the title
|
@emc415 just convert back from Draft when ready (it's on the right hand side) |
emc415
force-pushed
the
feat/optimize_calldatasize_check
branch
from
a723860
to
a79edba
Compare
charles-cooper
changed the title
|
looking like there's a test failure in ctor assembly |
|
The failing test case is a little weird; it depends on JUMP being in the opcodes output to demarcate the initcode from the runtime code. I would instead use the assembly output which is a little more structured. |
adjusted comment
|
looks decent. @fubuloubu what do you think? |
|
also tagging @wadealexc in case you want to leave any feedback. |
fubuloubu
changed the title
What I did
This wip commit addresses the first half of inefficient selector table checks (issue #3037)
How I did it
I added a function which tests whether a contract’s function has a selector with trailing zeroes (wip: at the moment it catches all selectors which contain 0)
It then changes the IR created in def _runtime_ir() to only have a line check in IR for calldatasize >= 4 if there is a zero selector function in the contract.
How to verify it
Commit message
Addresses the first half of inefficient selector table checks (issue #3037) by only running a check for calldatasize >= 4 if a zero selector function is present.
Description for the changelog
only check for calldatasize >= 4 if zero selector function is present.
Cute Animal Picture
which is the animal? :D