Generate a destructor function from AsData

[zliu41]

For sum types, it is better to use the destructor function than the pattern synonyms to match on them, because it uses caseInteger. It seems difficult if not impossible to compile pattern synonyms to caseInteger.

For example, the destructor for AsData.Budget.Types.TheseD:

    matchTheseD_a5Su ::
      forall a_a5Sj b_a5Sk r_a5Sy. (PlutusTx.UnsafeFromData a_a5Sj,
                                    PlutusTx.UnsafeFromData b_a5Sk) =>
                                   TheseD_a5Sf a_a5Sj b_a5Sk
                                   -> (a_a5Sj -> r_a5Sy)
                                      -> (b_a5Sk -> r_a5Sy)
                                         -> (a_a5Sj -> b_a5Sk -> r_a5Sy) -> r_a5Sy
    {-# INLINE matchTheseD_a5Su #-}
    matchTheseD_a5Su
      (TheseD_6989586621679032375_a5Sl d_a5Sv)
      k0_a5Sz
      k1_a5SA
      k2_a5SB
      = (BI.casePair (wrapUnsafeDataAsConstr d_a5Sv)
           $ (\ idx_a5Sw args_a5Sx
                -> BI.caseInteger
                     idx_a5Sw
                     [k0_a5Sz (PlutusTx.unsafeFromBuiltinData (BI.head args_a5Sx)),
                      k1_a5SA (PlutusTx.unsafeFromBuiltinData (BI.head args_a5Sx)),
                      BI.unsafeCaseList
                        (\ hd0_a5SC tl0_a5SD
                           -> k2_a5SB
                                (PlutusTx.unsafeFromBuiltinData hd0_a5SC)
                                (PlutusTx.unsafeFromBuiltinData (BI.head tl0_a5SD)))
                        args_a5Sx]))

The next step is to drop redundant unsafeCaseList calls, using a similar trick to this.

[github-actions]

Execution Budget Golden Diff

b6d66c6 (master) vs 8c87501

output

plutus-tx-plugin/test/AsData/Budget/9.12/destructSum.golden.eval

Metric	Old	New	Δ%
CPU	6_273_583	5_375_771	-14.31%
Memory	24_395	20_728	-15.03%
Flat Size	225	179	-20.44%

plutus-tx-plugin/test/AsData/Budget/9.6/destructSum.golden.eval

Metric	Old	New	Δ%
CPU	6_273_583	5_375_771	-14.31%
Memory	24_395	20_728	-15.03%
Flat Size	225	179	-20.44%

This comment will get updated when changes are made.

[Unisay]

Nice work! The CPS destructor approach is clean and the budget improvements are impressive (~14% CPU, ~24% AST size reduction on destructSum).

[Unisay]

Nit/suggestion: The golden test covers TheseD (3-constructor sum) well, and indirectly validates the TH machinery since incorrect code generation would show up in the PIR/UPLC goldens. If you're looking for places to expand coverage in the follow-up (dropping redundant unsafeCaseList), testing a product-type destructor or a zero-field constructor dispatch could be nice additions.

[zliu41]

The fact that everything compiles is an indication of correctness. There will be more tests once I implement step 2.