Code Jar Improvements with Stub #137
Conversation
quark-core/src/CodeJarStub.yul
Outdated
| object "CodeJarStub_9" { | ||
| code { | ||
| /// @src 0:66:211 "contract CodeJarStub {..." | ||
| mstore(64, 128) |
There was a problem hiding this comment.
It's pretty lol, but this is all that ends up executing, I believe. since constructor_CodeJarStub_9 is our code that returns. I believe the optimizer is smart enough to optimize the rest out when generating bytecode.
Looks like the overhead is only affecting tests using Agreed, would be good to confirm where the extra overhead comes from. |
quark-core/src/CodeJar.sol
Outdated
| assembly { | ||
| codeCreateAddress := create2(0, add(initCode, 32), initCodeLen, 0) | ||
| } | ||
| CodeJarStub codeCreateAddress = new CodeJarStub{salt: 0}(code); |
There was a problem hiding this comment.
nit: not technically an address, think a name like createdCode is more fitting
quark-core/src/CodeJar.sol
Outdated
| * @dev Returns the create2 address based on the given initCode | ||
| * @return The create2 address based on running the initCode constructor |
hayesgm
force-pushed
the
hayesgm/code-jar-improvements
branch
4 times, most recently
from
86eeaba to
f89eec4
Compare
Looks like the gas overhead is coming from the initcode for Though, I guess there's nothing we can do other than try to play around with compiler settings to optimize |
hayesgm
force-pushed
the
hayesgm/code-jar-improvements
branch
from
f89eec4 to
0499de8
Compare
This patch improves Code Jar to use CodeJarStub, which is a stub that's semi-verifiable on Etherscan for deployments, and significantly reduces the complexity/noise of how CodeJar works. This is also the basis for CodeJar deployments, which will be included in a different branch.
hayesgm
force-pushed
the
hayesgm/code-jar-improvements
branch
from
0499de8 to
ca843bb
Compare
This patch just hacks in `datasize("CodeJarStub")` as a magic constant (derived emperically) into the contract itself, which allows us to get an optimal version of CodeJarStub that saves ~2-3K gas from using Solidity's ABI-encoding which is particularly expensive mostly due to dynamic allocations that break the optimizer. This is a weird system, but read the comments as to why it's probably an okay decision.
This patch improves Code Jar to use CodeJarStub, which is a stub that's semi-verifiable on Etherscan for deployments, and significantly reduces the complexity/noise of how CodeJar works. This is also the basis for CodeJar deployments, which will be included in a different branch.