Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (devnets/bal/7@a3e5201). Learn more about missing BASE report.

@@               Coverage Diff                @@
##             devnets/bal/7    #2843   +/-   ##
================================================
  Coverage                 ?   87.35%           
================================================
  Files                    ?      586           
  Lines                    ?    35957           
  Branches                 ?     3382           
================================================
  Hits                     ?    31410           
  Misses                   ?     3926           
  Partials                 ?      621           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

This is looking good @leolara! I may just be missing it, but what is .mb/oog-by-design-amsterdam-approaches.md? Also added a comment about adding the lift to both tx_gas values - let me know what you think!

the directory .mb works as a memory bank of my agent, where we store findings and things that should know in the future.

In this case the content of this file is:

# OoG-by-design ported_static tests — fix approaches for Amsterdam

After PR #2839 there are still **119 OoG-by-design entries** in
`tests/ported_static/amsterdam_skip_list.txt`. These are tests whose
budgets are tuned to land *just below* an EIP-8037-affected cost
boundary (SSTORE-set, CALL+value to empty account, MEMORY expansion,
nested CALL forwarding), so a blind gas bump invalidates their
assertion and a plain skip loses the coverage.

This note enumerates tractable approaches, ordered by risk and
expected entries cleared.

## Approach 1 — extend `tx_gas` with an Amsterdam threshold (lowest risk)

Tests of shape

```python
tx_gas = [happy_path, just_below_threshold]

where the threshold is "intrinsic + 1 spillable op" can be made
fork-conditional in one line:

tx_gas = [800000, 80000]
if fork.is_eip_enabled(8037):
    # Amsterdam: OoG threshold lifts by the SSTORE-set state-gas
    tx_gas = [800000, 80000 + fork.sstore_state_gas()]

Works when the OoG depends on one spillable operation. Pick-up:
~20–30 entries — simpler stSStoreTest, stCreate2/*_oo_gafter_*,
stWalletTest. Each fix is ≤6 lines.

Approach 2 — derive the bump from the expected post-state

For tests that complete N SSTOREs and then OoG (post-state shows
storage={1:1, …, N:1} with entry N+1 missing), the Amsterdam budget
needs + N × fork.sstore_state_gas() headroom to land at the same
intermediate state. A tiny helper centralises the math:

def oog_budget(base: int, *, sstores_before_oog: int, fork: Fork) -> int:
    """Adjust an OoG-tuned gas budget for EIP-8037 state-gas spill."""
    return base + sstores_before_oog * fork.sstore_state_gas()

Per-test usage stays a single line. Pick-up: ~40–50 entries —
stSStoreTest/test_sstore_0to0*, xto_* family, parts of
stStaticCall/test_static_check_opcodes5,
stRevertTest/test_revert_opcode_multiple_sub_calls. Each fix is
≤8 lines but requires reading the expected post-state to count N.

Approach 3 — fork-conditional expect_section entries

Tests already using resolve_expect_post can get a ">=Amsterdam"
entry with a smaller intermediate state — accept that the OoG fires
earlier on Amsterdam and the recorded state is different. Keeps the
original gas budget intact.

expect_entries_ = [
    {
        "indexes": {"data": -1, "gas": 1, "value": -1},
        "network": ["Cancun"],
        "result": {target: Account(storage={1: 1, 2: 1, 3: 1})},
    },
    {
        "indexes": {"data": -1, "gas": 1, "value": -1},
        "network": [">=Amsterdam"],
        "result": {target: Account(storage={1: 1})},  # OoGs sooner
    },
]

Trades gas-bump complexity for post-state enumeration. Useful where
multiple d parametrizations share a tx_gas list and you can't
bump it without breaking other entries. Pick-up: ~10–15 entries.

Approach 4 — CALL-stack-depth tests (*1024_oog family)

stCallCreateCallCodeTest/test_call1024_oog,
test_callcode1024_oog, stDelegatecallTestHomestead/test_call1024_oog,
test_delegatecall1024_oog recurse 1024 deep and assert OoG when the
call frame hits the depth limit. EIP-8037 doesn't touch stack-depth
mechanics — the failures here are gas exhaustion firing before
depth 1024 because each frame's state-gas spill compounds.

Fix: bump the inner-CALL forwarded gas so each frame survives until
the depth limit fires. Original gas budget stays. Pick-up: ~9 entries.

Approach 5 — accept-and-document the rest

The genuinely stuck cases, with explicit reasons:

• Pre-EIP-150 gas-pricing assertions — tests like
  stStaticCall/test_static_callcallcodecallcode_011_oogm_after* are
  calibrated against gas costs that no longer match production
  behaviour. EIP-8037 spec tests already cover the new 2D gas model.
• Memory-expansion + state-gas combination — stMemExpandingEIP150Calls/*
  hits OoG at exact byte counts where the state-gas reservoir changes
  the effective cap; no single-variable fix.
• Inverse failures — stEIP3860_limitmeterinitcode/*invalid cases:
  Amsterdam allows what was previously rejected; behaviour change, not
  a gas issue.

For these, add a one-line rationale next to the skip entry (or a
section header comment in amsterdam_skip_list.txt) and leave them
as documented coverage gaps.

Recommended sequencing

1. PR-A: Approach 1 + Approach 4 — mechanical, ≤15 lines per file,
   ~30–40 entries cleared.
2. PR-B: Approach 2 — introduce the oog_budget helper, sweep
   stSStoreTest. ~50 entries cleared. Helper lives in
   execution_testing.forks or a per-package utils module.
3. PR-C: Approach 3 — selective use on the harder multi-d tests
   where (2) won't fit. ~10–15 entries.
4. PR-D: documentation pass — annotate the residue (~15–20
   entries) with skip rationale comments so reviewers know they're
   intentional gaps, not "yet to be triaged".

Endpoint: ~80 stubbornly-skipped tests, each with a documented
reason — vs. the current 554 with one bucket reason for everything.

Per-test inventory (for the PR-A starting point)

Best Approach-1 candidates (single-SSTORE OoG, tx_gas[1] adjustable):

• stSStoreTest/test_sstore_gas.py
• stSStoreTest/test_sstore_gas_left.py
• stCreate2/test_create2_oo_gafter_init_code.py (g1)
• stCreate2/test_create2_oo_gafter_init_code_returndata2.py (g1)
• stCreate2/test_create2_oo_gafter_init_code_revert2.py
• stCreateTest/test_create_oo_gafter_init_code.py
• stCreateTest/test_create_oo_gafter_init_code_returndata2.py
• stCreateTest/test_create_oo_gafter_init_code_returndata_size.py
• stCreateTest/test_create_oo_gafter_init_code_revert2.py
• stWalletTest/test_day_limit_construction_partial.py
• stWalletTest/test_wallet_construction_oog.py (g1)
• stWalletTest/test_wallet_construction_partial.py
• stWalletTest/test_multi_owned_construction_not_enough_gas_partial.py (g1)
• stRevertTest/test_revert_sub_call_storage_oog.py (g1)
• stRevertTest/test_revert_sub_call_storage_oog2.py (g1)
• stMemoryTest/test_oog.py

Best Approach-4 candidates:

• stCallCreateCallCodeTest/test_call1024_oog.py (4 entries)
• stCallCreateCallCodeTest/test_callcode1024_oog.py (2 entries)
• stDelegatecallTestHomestead/test_call1024_oog.py (2 entries)
• stDelegatecallTestHomestead/test_delegatecall1024_oog.py (1 entry)

@spencer-tb I think the lint problem I am getting is from the base branch, on which branch should I rebase this?

devnets/bal/7 ported_static — progress after PR #2843

Snapshot date: 2026-05-26
Branch: wt-bal-7-amsterdam-oog-by-design (PR #2843)

Remaining after PR #2843 merges

Cumulative reduction across all my PRs

Notes

• Skip-list entries are nodeid substrings; each typically expands to
  ~2–3 fixture variants (state_test,
  blockchain_test_from_state_test,
  blockchain_test_engine_from_state_test).
• The 1,197 figure is the actual count uv run fill --fork Amsterdam
  reports as skipped.
• For comparison: issue Static Test Fail Tracker for EIP-8037 #2601 quoted 3,423 failures on the
  pre-recalibration eips/amsterdam/eip-8037 branch. The 1,197 number
  on bal/7 is ~65% lower despite the CPSB-1530 recalibration adding
  new failure modes that didn't exist at the time Static Test Fail Tracker for EIP-8037 #2601 was filed.
• Classification of the remaining 480 entries (OoG-by-design,
  gas-measurement, bytecode-baked, multi-param-per-d, balance-refund,
  size-limit, other) is in
  bal-7-skipped-tests-classification.md.

@spencer-tb This commit here: 615745c is fixing something unrelated to this PR, but an error in bal/7 branch, that is causing the automatic tests to don't pass

@kclowes could you please check again?