fix(gas): improve bundle gas estimation calculation

[alex-miao]

Contributes to #349

Proposed Changes

Improve the bundle gas limit calculation. The bundle gas limit we now use is based off the following observation:

In the Entrypoint contract, the innerHandleOp call has this check that the bundle would need to pass for every single user op that it executes.

Within a bundle, let $p_i$, $v_i$, and $c_i$ be the static pre-verification gas, verification gas limit, and call gas limit of user operation $i$. Also let $m_i$ be $1$ if user op $i$ uses a paymaster, and $0$ if not. Suppose we start with initial gas $g$. When processing user op $i$, at the point of the the innerHandleOp check we have yet to execute the call for user op $i$, and the check requires that at that point the remaining gas be greater than $c_i+v_i+5000$. If user op $i$ uses a paymaster, because of the postOp calls, we would also need the remaining gas to also be greater than $c_i+2v_i$, since the up to two postOp calls can each use up $v_i$ gas.

We can express this condition as:

$$remaining\_gas_i\geq c_i+v_i+5000+m_i \max(v_i-5000,0)$$

The worst case remaining gas at this check for user operation $i$, since we have already processed user ops $1$ through $i-1$ and both the pre-verification gas and up to the verification gas limit of user op $i$, is:
$$g-\sum_{j=1}^{i-1}\big(p_j+(1+2m_j)v_j+c_j\big)-p_i-v_i$$

In order to not run out of gas and pass all the checks, we would therefore need that for all $i$:
$$g\geq \sum_{j=1}^{i-1}\big(p_j+(1+2m_j)v_j+c_j\big)+p_i+2v_i+c_i+5000+m_i \max(v_i-5000,0)$$

so the smallest initial gas limit we can assign the bundle would be the max over all $i$ of the right hand side of the inequality.

[codecov]

Codecov Report

Merging #444 (774cfa1) into main (e220e0a) will increase coverage by 0.40%.
The diff coverage is 98.68%.

❗ Current head 774cfa1 differs from pull request most recent head 04c9c75. Consider uploading reports for the commit 04c9c75 to get more accurate results

Files	Coverage Δ
crates/builder/src/bundle_proposer.rs	88.89% <100.00%> (-0.02%)	⬇️
crates/sim/src/precheck.rs	68.40% <100.00%> (ø)
crates/sim/src/gas/gas.rs	93.81% <98.56%> (+4.18%)	⬆️

... and 4 files with indirect coverage changes

Flag	Coverage Δ
unit-tests	52.42% <98.68%> (+0.40%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
rundler binary	0.00% <ø> (ø)
builder	45.06% <100.00%> (-0.01%)	⬇️
dev	0.00% <ø> (ø)
pool	59.65% <ø> (+0.01%)	⬆️
provider	1.33% <ø> (ø)
rpc	18.51% <ø> (ø)
sim	82.80% <98.57%> (+0.60%)	⬆️
tasks	∅ <ø> (∅)
types	93.56% <ø> (ø)
utils	7.20% <ø> (ø)

[0xfourzerofour]

Good job! that math seems accurate to me

[dancoombs]

This is 🔥 . Math LGTM. Want to see this in action during our load tests ASAP.

[dancoombs]

🚀 🚢

The base branch was changed.