[mohammadfawaz]

Summary

Preserve ternaries over arrays and plaintext structs as single Aleo ternary instructions. Keep tuple and record ternaries split into element or field selections. Array, struct, and identifier ternaries require ConsensusVersion::V23.

Update the snarkVM dependency to PR #3222, commit 933acedd6. Support V23 in CLI version selection. Align the existing RocksDB dependency with the selected snarkVM revision. Update the fixtures for current constructor, execution-harness, and consensus-version requirements.

Measurement method

Compare this PR with master at fcbebc7db6. Use identical source inputs and build options for both compilers. Each percentage is the change in total bytes for its stated corpus.

Regenerate compiler expectations with UPDATE_EXPECT=1 cargo test -p leo-compiler --lib test_compiler::test_compiler -- --exact --nocapture. For external programs, run leo build --network testnet --offline from fresh source copies and measure the main program's emitted .aleo text file. Exclude imported bytecode from those program totals.

Compiler expectation files

Compare the 1,675 fixture-backed output paths present in both trees. Exclude the 11 new compiler fixtures and nine orphaned expectation files. These totals include diagnostics and output from both import modes; they measure expectation files, not only Aleo bytecode.

Metric	Before	After	Change
Total bytes	2,126,672	2,107,529	-19,143 (-0.90%)
Lines, counted by LF bytes	65,369	65,007	-362
Changed files		32 of 1,675

The subset with bytecode and no errors contains 1,113 files. Its size changes from 1,570,011 to 1,553,197 bytes, a reduction of 16,814 bytes (1.07%). Its line count changes from 51,393 to 51,073, and 29 files change.

Changes for every compiler path listed in the previous description

Output under tests/expectations/compiler/	Before bytes	After bytes	Change
function/flatten_test.out	9,679	4,077	-5,602
storage/external_storage.out	35,835	32,620	-3,215
option/unwrap_or_deep.out	5,925	2,765	-3,160
storage/aggregates.out	11,158	8,257	-2,901
option/in_modules.out	11,557	9,235	-2,322
function/flatten_arrays.out	2,292	1,038	-1,254
examples/tictactoe.out	5,635	4,559	-1,076
option/implicit_wrapping.out	7,908	7,248	-660
storage/primitives.out	4,805	5,361	+556
storage/signed.out	4,339	4,895	+556
storage/unsigned.out	4,334	4,890	+556
storage/external_storage_multi_deps.out	7,010	7,788	+778

Aggregate selection can retain both branch constructions. Some storage outputs increase in size, including the four listed above.

Bytecode impact: leo-examples

Use all 34 programs from leo-examples at 03554ec3.

Metric	Before	After	Change
Total bytes	132,465	131,341	-1,124 (-0.85%)
Total lines	4,286	4,262	-24
Changed programs		1 of 34

tictactoe changes from 6,525 to 5,401 bytes (-17.23%) and from 207 to 183 lines. Its output contains two aggregate ternaries, both over Board values. The other 33 programs are byte-identical, checked by SHA-256.

Bytecode impact: amm-v3

Use the development snapshot at 8d161722. It contains seven primary source programs, five test programs, and two libraries. Keep the pinned ARCs dependency at bf0a4e7aa9cb155751aa6b8aebe7c9162e2a7a5f for both compilers. No source port is required.

The primary-program comparison excludes the five test programs and two unchanged precompiled ARC22 inputs. The libraries, u256 and oracle_types, emit no standalone program bytecode.

Primary program	Before bytes	After bytes	Change
shield_swap.aleo	1,480,499	1,390,305	-90,194 (-6.09%)
shield_swap_freezelist.aleo	5,727	5,727	0
shield_swap_lp_router.aleo	21,344	21,344	0
shield_swap_multisig_core.aleo	39,324	39,324	0
shield_swap_pricing_terms.aleo	9,140	9,140	0
shield_swap_rebalance_router.aleo	44,851	44,851	0
shield_swap_router.aleo	35,507	35,507	0
Total	1,636,392	1,546,198	-90,194 (-5.51%)

Only shield_swap.aleo changes. Its line count changes from 36,903 to 35,340 (-1,563). The other six primary programs are byte-identical. Total primary-program lines change from 40,494 to 38,931.

Upgrade compatibility: shield_swap

The candidate bytecode is not a valid upgrade from the baseline bytecode built from the same pinned AMM source. The interface, closure, and constructor checks pass. The record-output register check fails. snarkVM requires these registers to stay unchanged for upgrades at V10 and later, including V23.

Function	Record type	Baseline output register	Candidate output register
mint	PositionNFT	r1242	r972
mint	MintComplianceRecord	r1241	r971
rebalance_position	PositionNFT	r840	r660
rebalance_position	MintComplianceRecord	r839	r659
collect	PositionNFT	r826	r646

Do not use this size reduction as evidence of upgrade compatibility. To upgrade from this baseline, the new bytecode must preserve the existing record-output registers. The custom constructor also requires multisig approval for the exact new checksum and next edition.

At the observed mainnet height of 23,392,293, the endpoint reports shield_swap.aleo at edition 3, deployed at block 22,245,833 in transaction at1ts87l4c708nffeyd8l75cgje69trrmzyr8vnhe5ddeeurccj2grs5076w2. Its record-output registers match the baseline column above. The candidate fails the same record-register check against this deployed edition. A separate constructor mismatch exists in both baseline and candidate builds: the pinned source uses a different initial deployer address from edition 3. The baseline passes the record-register check but fails the constructor check. The candidate fails both. V23 does not remove either upgrade restriction.

AMM test-program outputs

Test program	Before bytes	After bytes	Change
test_token_a.aleo	5,718	5,718	0
test_token_b.aleo	5,718	5,718	0
test_token_c.aleo	5,719	5,719	0
test_token_d.aleo	5,719	5,719	0
test_oracle_consumer.aleo	139,832	139,871	+39
Total	162,706	162,745	+39 (+0.024%)

The oracle consumer increases by 39 bytes, while its line count changes from 3,930 to 3,924. The four token programs are byte-identical. Total test-program lines change from 4,698 to 4,692 (-6).