[Builtins] Add pattern matching builtins #6530

effectfully · 2024-10-01T16:48:39Z

This PR is forked from #5486, it's squashed and without a lot of noise in the comments.

The main change is replacing

data BuiltinRuntime val
    = BuiltinCostedResult ExBudgetStream ~(BuiltinResult val)
    | <...>

with

data BuiltinRuntime val
    = BuiltinCostedResult ExBudgetStream ~(BuiltinResult (HeadSpine val))
    | <...>

where HeadSpine is a fancy way of saying NonEmpty:

-- | A non-empty spine. Isomorphic to 'NonEmpty', except is strict and is defined as a single
-- recursive data type.
data Spine a
    = SpineLast a
    | SpineCons a (Spine a)

-- | The head-spine form of an iterated application. Provides O(1) access to the head of the
-- application. Isomorphic to @NonEmpty@, except is strict and the no-spine case is made a separate
-- constructor for performance reasons (it only takes a single pattern match to access the head when
-- there's no spine this way, while otherwise we'd also need to match on the spine to ensure that
-- it's empty -- and the no-spine case is by far the most common one, hence we want to optimize it).
data HeadSpine a
    = HeadOnly a
    | HeadSpine a (Spine a)

(we define a separate type, because we want strictness, and you don't see any bangs, because it's in a module with StrictData enabled).

The idea is that a builtin application can return a function applied to a bunch of arguments, which is exactly what we need to be able to express caseList

caseList xs0 f z = case xs0 of
   []   -> z
   x:xs -> f x xs

as a builtin:

-- | Take a function and a list of arguments and apply the former to the latter.
headSpine :: Opaque val asToB -> [val] -> Opaque (HeadSpine val) b
headSpine (Opaque f) = Opaque . \case
    []      -> HeadOnly f
    x0 : xs ->
        -- It's critical to use 'foldr' here, so that deforestation kicks in.
        -- See Note [Definition of foldl'] in "GHC.List" and related Notes around for an explanation
        -- of the trick.
        HeadSpine f $ foldr (\x2 r x1 -> SpineCons x1 $ r x2) SpineLast xs x0

instance uni ~ DefaultUni => ToBuiltinMeaning uni DefaultFun where
    <...>
    toBuiltinMeaning _ver CaseList =
        let caseListDenotation
                :: Opaque val (LastArg a b)
                -> Opaque val (a -> [a] -> b)
                -> SomeConstant uni [a]
                -> BuiltinResult (Opaque (HeadSpine val) b)
            caseListDenotation z f (SomeConstant (Some (ValueOf uniListA xs0))) = do
                case uniListA of
                    DefaultUniList uniA -> pure $ case xs0 of
                        []     -> headSpine z []                                             -- [1]
                        x : xs -> headSpine f [fromValueOf uniA x, fromValueOf uniListA xs]  -- [2]
                    _ ->
                        -- See Note [Structural vs operational errors within builtins].
                        throwing _StructuralUnliftingError "Expected a list but got something else"
            {-# INLINE caseListDenotation #-}
        in makeBuiltinMeaning
            caseListDenotation
            (runCostingFunThreeArguments . unimplementedCostingFun)

Being able to express [1] (representing z) and [2] (representing f x xs) is precisely what this PR enables.

Adding support for the new functionality to the CEK machine is trivial. All we need is a way to push a Spine of arguments onto the context:

    -- | Push arguments onto the stack. The first argument will be the most recent entry.
    pushArgs
        :: Spine (CekValue uni fun ann)
        -> Context uni fun ann
        -> Context uni fun ann
    pushArgs args ctx = foldr FrameAwaitFunValue ctx args

and a HeadSpine version of returnCek:

    -- | Evaluate a 'HeadSpine' by pushing the arguments (if any) onto the stack and proceeding with
    -- the returning phase of the CEK machine.
    returnCekHeadSpine
        :: Context uni fun ann
        -> HeadSpine (CekValue uni fun ann)
        -> CekM uni fun s (Term NamedDeBruijn uni fun ())
    returnCekHeadSpine ctx (HeadOnly  x)    = returnCek ctx x
    returnCekHeadSpine ctx (HeadSpine f xs) = returnCek (pushArgs xs ctx) f

Then replacing

                BuiltinSuccess x ->
                    returnCek ctx x

with

                BuiltinSuccess fXs ->
                    returnCekHeadSpine ctx fXs

(and similarly for BuiltinSuccessWithLogs) will do the trick.

We used to define caseList in terms of IfThenElse, NullList and either HeadList or TailList depending on the result of NullList, i.e. three builtin calls in the worst and in the best case. Then we introduced ChooseList, which replaced both IfThenElse and NullList in caseList thus bringing total amount of builtin calls down to 2 in all cases. This turned out to have a substantial impact on performance. This PR allows us to bring total number of builtin calls per caseList invokation down to 1 -- the CaseList builtin itself.

effectfully · 2024-10-01T17:44:16Z

plutus-benchmark/script-contexts/test/Data/9.6/checkScriptContext1.pir.golden

-                    (`$fUnsafeFromDataList_$cunsafeFromBuiltinData`
-                       {TxInInfo}
-                       `$fUnsafeFromDataScriptContext_$cunsafeFromBuiltinData`
-                       (headList {data} args))
-                    (`$fUnsafeFromDataList_$cunsafeFromBuiltinData`
-                       {TxInInfo}
-                       `$fUnsafeFromDataScriptContext_$cunsafeFromBuiltinData`
-                       (headList {data} l))
-                    (`$fUnsafeFromDataList_$cunsafeFromBuiltinData`
-                       {TxOut}
-                       `$fUnsafeFromDataTxOut_$cunsafeFromBuiltinData`
-                       (headList {data} l))
+                    (let
+                      !d : data = headList {data} args
+                    in
+                    go (unListData d))
+                    (let
+                      !d : data = headList {data} l
+                    in
+                    go (unListData d))
+                    (let
+                      !d : data = headList {data} l
+                    in
+                    go (unListData d))


We end up having slightly more PIR code sometimes, but I suppose the reason is that we get more stuff inlined as you can see here. All those go are different optimized instantiations of $fUnsafeFromDataScriptContext_$cunsafeFromBuiltinData, so it's not surprising that we get more PIR code.

effectfully · 2024-10-01T17:53:17Z

plutus-benchmark/lists/test/Lookup/9.6/match-builtin-list-10.budget.golden

+({cpu: 13100114647000
+| mem: 13100000559240})


This is due to unimplementedCostingFun.

Can you put in an approximate cost model, and see how these numbers look? I assume the cost model will be in a similar range to chooseLists? Even if it is a bad guess, it won't be more meaningless than these numbers!

Whoops, I've done it already, just forgot to mention it. Here are the numbers:

+++ b/plutus-benchmark/lists/test/Lookup/9.6/match-builtin-list-50.budget.golden @@ -1,2 +1,2 @@ -({cpu: 4254600144 -| mem: 18263632}) \ No newline at end of file +({cpu: 2884053694 +| mem: 12257832}) +++ b/plutus-benchmark/lists/test/Sum/9.6/left-fold-built-in.budget.golden @@ -1,2 +1,2 @@ -({cpu: 123874594 -| mem: 533932}) \ No newline at end of file +({cpu: 86545294 +| mem: 397232}) +++ b/plutus-benchmark/lists/test/Sum/9.6/left-fold-data.budget.golden @@ -1,2 +1,2 @@ -({cpu: 279116232 -| mem: 1124130}) \ No newline at end of file +({cpu: 215640257 +| mem: 912434}) +++ b/plutus-benchmark/lists/test/Sum/9.6/right-fold-built-in.budget.golden @@ -1,2 +1,2 @@ -({cpu: 128674594 -| mem: 563932}) \ No newline at end of file +({cpu: 91345294 +| mem: 427232}) +++ b/plutus-benchmark/marlowe/test/semantics/9.6/f2a8fd2014922f0d8e01541205d47e9bb2d4e54333bdd408cbe7c47c55e73ae4.budget.golden @@ -1,2 +1,2 @@ -({cpu: 689404030 -| mem: 2815274}) \ No newline at end of file +({cpu: 654073429 +| mem: 2668846}) +++ b/plutus-benchmark/script-contexts/test/9.6/checkScriptContext2-20.budget.golden @@ -1,2 +1,2 @@ -({cpu: 281562223 -| mem: 1042586}) \ No newline at end of file +({cpu: 268649443 +| mem: 1019846}) \ No newline at end of +++ b/plutus-benchmark/script-contexts/test/9.6/checkScriptContext2-4.budget.golden @@ -1,2 +1,2 @@ -({cpu: 84515135 -| mem: 323994}) \ No newline at end of file +({cpu: 80537379 +| mem: 310726}) +++ b/plutus-ledger-api/test-plugin/Spec/Data/Budget/9.6/currencySymbolValueOf.budget.golden @@ -1,2 +1,2 @@ -({cpu: 23159162 -| mem: 65580}) \ No newline at end of file +({cpu: 18013406 +| mem: 45012}) +++ b/plutus-ledger-api/test-plugin/Spec/Data/Budget/9.6/geq1.budget.golden @@ -1,2 +1,2 @@ -({cpu: 620731320 -| mem: 1881010}) \ No newline at end of file +({cpu: 527159505 +| mem: 1514810}) +++ b/plutus-tx-plugin/test/Budget/9.6/builtinListIndexing.budget.golden @@ -1,2 +1,2 @@ -({cpu: 8401207 -| mem: 33930}) \ No newline at end of file +({cpu: 6452329 +| mem: 27546}) +++ b/plutus-tx-plugin/test/Budget/9.6/map2-budget.budget.golden @@ -1,2 +1,2 @@ -({cpu: 127361368 -| mem: 398526}) \ No newline at end of file +({cpu: 105442279 +| mem: 312734})

Basically, reflect the speedup percentages very well, as one'd expect.

I assume we don't want to commit the approximate cost model, given that it'll increase the changes of us accidentally enabling the feature for the users prematurely.

effectfully · 2024-10-01T19:02:59Z

plutus-core/plutus-core/examples/PlutusCore/Examples/Data/Data.hs

@@ -43,7 +43,7 @@ import Data.ByteString (ByteString)
 -- >      (fI : integer -> r)
 -- >      (fB : bytestring -> r) ->
 -- >          fix {data} {r} \(rec : data -> r) (d : data) ->
-- >              caseData
+-- >              matchData


Aiken folks asked us to make caseList and caseData accept list/data as the last argument (which makes a lot of sense optimization-wise), so I made that and renamed all existing occurrences of caseList/caseData to matchList/matchData for consistency (as we already have those and they accept list/data as the first argument). I'd much prefer to do it the opposite way, but it's too late at this point, so consistency and backwards compatibility win.

effectfully · 2024-10-01T19:03:58Z

plutus-core/plutus-core/test/CostModelSafety/Spec.hs

+    KnownTypeAst PLC.TyName DefaultUni a =>
+    TypeRep a ->
+    PLC.Term PLC.TyName PLC.Name DefaultUni PLC.DefaultFun ()
+smallTerm tr0 = go (toTypeAst tr0) tr0 where


Now that we have builtins accepting functions, we have to generate functions for various tests.

effectfully · 2024-10-01T19:04:26Z

plutus-core/plutus-core/test/TypeSynthesis/Golden/StdLib/Data/List/MatchList.plc.golden

@@ -0,0 +1 @@
+all a. list a -> (all r. r -> (a -> list a -> r) -> r)


Hate those parens, but we'll probably never have the time to implement the idea.

effectfully · 2024-10-01T19:07:14Z

plutus-core/plutus-core/src/PlutusCore/Builtin/Elaborate.hs

The elaborator needed surprisingly few tweaks to keep track of the current context.

effectfully · 2024-10-01T19:07:51Z

plutus-tx-plugin/test/size/Golden/Rational/Serialization/fromBuiltinData.size.golden

effectfully · 2024-10-01T19:08:39Z

plutus-core/plutus-core/src/PlutusCore/Default/Builtins.hs

              go t  = fail $ "Failed to decode builtin tag, got: " ++ show t

    size _ n = n + builtinTagWidth
+
+{- Note [Legacy pattern matching on built-in types]


This is an old Note, I just moved it to the bottom of the file, so that we still have an explanation for all the chooseList etc stuff while not wasting a lot of precious space in the middle of the file.

effectfully · 2024-10-01T19:09:42Z

plutus-core/plutus-core/test/CostModelSafety/Spec.hs

@@ -183,7 +205,7 @@ testCosts
  :: BuiltinSemanticsVariant DefaultFun
  -> BuiltinsRuntime DefaultFun Term
  -> DefaultFun
-  -> Assertion
+  -> TestTree


We should probably add type checking to that test, because I accidentally generated a bunch of ill-typed code and the test was happy to swallow that.

plutus-core/untyped-plutus-core/src/UntypedPlutusCore/Transform/CaseOfCase.hs

effectfully · 2024-10-01T19:31:31Z

Latest benchmarking results (before the benchmarking machine caught a flu) are here.

plutus-core/plutus-core/stdlib/PlutusCore/StdLib/Data/Data.hs

plutus-core/untyped-plutus-core/src/UntypedPlutusCore/Evaluation/Machine/Cek/Internal.hs

ana-pantilie

I assume the costing is all messed up because we haven't introduced a cost model yet, right? So that will come in a future PR?

I didn't understand all of the details in the code, but I think I now finally understand the big picture and how we can use the new evaluator. Thank you for this!

Approving because it makes sense to me, but I am not very familiar with the code so maybe someone more knowledgeable should also approve.

plutus-tx/src/PlutusTx/Builtins/Internal.hs

effectfully · 2024-10-02T19:03:47Z

I assume the costing is all messed up because we haven't introduced a cost model yet, right? So that will come in a future PR?

Yes and yes.

zliu41

How is correctness tested, e.g., verifying that CaseList behaves the same as the equivalent term defined via ChooseList?

zliu41 · 2024-10-02T19:41:20Z

plutus-benchmark/script-contexts/test/9.6/checkScriptContext2.size.golden

Any idea of the reason for the increased sizes?

My guess is that it's that the new definitions are more inliner-friendly, which creates some inlining and monomorphization opportunities, which increases code size, see this comment.

plutus-core/untyped-plutus-core/src/UntypedPlutusCore/Evaluation/Machine/Cek/Internal.hs

zliu41

Looks good except that clarification needed for the testing question above.

…to effectfully/builtins/add-pattern-matching-builtins

@nonempty

The main change is replacing ```haskell data BuiltinRuntime val = BuiltinCostedResult ExBudgetStream ~(BuiltinResult val) | <...> ``` with ```haskell data BuiltinRuntime val = BuiltinCostedResult ExBudgetStream ~(BuiltinResult (HeadSpine val)) | <...> ``` where `HeadSpine` is a fancy way of saying `NonEmpty`: ```haskell -- | A non-empty spine. Isomorphic to 'NonEmpty', except is strict and is defined as a single -- recursive data type. data Spine a = SpineLast a | SpineCons a (Spine a) -- | The head-spine form of an iterated application. Provides O(1) access to the head of the -- application. Isomorphic to @nonempty@, except is strict and the no-spine case is made a separate -- constructor for performance reasons (it only takes a single pattern match to access the head when -- there's no spine this way, while otherwise we'd also need to match on the spine to ensure that -- it's empty -- and the no-spine case is by far the most common one, hence we want to optimize it). data HeadSpine a = HeadOnly a | HeadSpine a (Spine a) ``` (we define a separate type, because we want strictness, and you don't see any bangs, because it's in a module with `StrictData` enabled). The idea is that a builtin application can return a function applied to a bunch of arguments, which is exactly what we need to be able to express `caseList` ```haskell caseList xs0 f z = case xs0 of [] -> z x:xs -> f x xs ``` as a builtin: ```haskell -- | Take a function and a list of arguments and apply the former to the latter. headSpine :: Opaque val asToB -> [val] -> Opaque (HeadSpine val) b headSpine (Opaque f) = Opaque . \case [] -> HeadOnly f x0 : xs -> -- It's critical to use 'foldr' here, so that deforestation kicks in. -- See Note [Definition of foldl'] in "GHC.List" and related Notes around for an explanation -- of the trick. HeadSpine f $ foldr (\x2 r x1 -> SpineCons x1 $ r x2) SpineLast xs x0 instance uni ~ DefaultUni => ToBuiltinMeaning uni DefaultFun where <...> toBuiltinMeaning _ver CaseList = let caseListDenotation :: Opaque val (LastArg a b) -> Opaque val (a -> [a] -> b) -> SomeConstant uni [a] -> BuiltinResult (Opaque (HeadSpine val) b) caseListDenotation z f (SomeConstant (Some (ValueOf uniListA xs0))) = do case uniListA of DefaultUniList uniA -> pure $ case xs0 of [] -> headSpine z [] -- [1] x : xs -> headSpine f [fromValueOf uniA x, fromValueOf uniListA xs] -- [2] _ -> -- See Note [Structural vs operational errors within builtins]. throwing _StructuralUnliftingError "Expected a list but got something else" {-# INLINE caseListDenotation #-} in makeBuiltinMeaning caseListDenotation (runCostingFunThreeArguments . unimplementedCostingFun) ``` Being able to express [1] (representing `z`) and [2] (representing `f x xs`) is precisely what this PR enables. Adding support for the new functionality to the CEK machine is trivial. All we need is a way to push a `Spine` of arguments onto the context: ```haskell -- | Push arguments onto the stack. The first argument will be the most recent entry. pushArgs :: Spine (CekValue uni fun ann) -> Context uni fun ann -> Context uni fun ann pushArgs args ctx = foldr FrameAwaitFunValue ctx args ``` and a `HeadSpine` version of `returnCek`: ```haskell -- | Evaluate a 'HeadSpine' by pushing the arguments (if any) onto the stack and proceeding with -- the returning phase of the CEK machine. returnCekHeadSpine :: Context uni fun ann -> HeadSpine (CekValue uni fun ann) -> CekM uni fun s (Term NamedDeBruijn uni fun ()) returnCekHeadSpine ctx (HeadOnly x) = returnCek ctx x returnCekHeadSpine ctx (HeadSpine f xs) = returnCek (pushArgs xs ctx) f ``` Then replacing ```haskell BuiltinSuccess x -> returnCek ctx x ``` with ```haskell BuiltinSuccess fXs -> returnCekHeadSpine ctx fXs ``` (and similarly for `BuiltinSuccessWithLogs`) will do the trick. We used to define `caseList` in terms of `IfThenElse`, `NullList` and either `HeadList` or `TailList` depending on the result of `NullList`, i.e. three builtin calls in the worst and in the best case. Then we introduced `ChooseList`, which replaced both `IfThenElse` and `NullList` in `caseList` thus bringing total amount of builtin calls down to 2 in all cases. This turned out to have a [substantial](IntersectMBO#4119 (review)) impact on performance. This PR allows us to bring total number of builtin calls per `caseList` invokation down to 1 -- the `CaseList` builtin itself.

effectfully added the Builtins label Oct 1, 2024

effectfully self-assigned this Oct 1, 2024

effectfully temporarily deployed to github-pages October 1, 2024 16:48 — with GitHub Actions Inactive

effectfully mentioned this pull request Oct 1, 2024

[Builtins] Add support for pattern matching builtins #5486

Closed

effectfully force-pushed the effectfully/builtins/add-pattern-matching-builtins branch from 21b50a0 to 15f6399 Compare October 1, 2024 17:34

effectfully temporarily deployed to github-pages October 1, 2024 17:34 — with GitHub Actions Inactive

effectfully commented Oct 1, 2024

View reviewed changes

effectfully force-pushed the effectfully/builtins/add-pattern-matching-builtins branch from 15f6399 to 1d1f1b5 Compare October 1, 2024 21:28

effectfully temporarily deployed to github-pages October 1, 2024 21:28 — with GitHub Actions Inactive

effectfully requested review from ana-pantilie, Unisay, kwxm and zliu41 October 1, 2024 21:28

zliu41 reviewed Oct 1, 2024

View reviewed changes

plutus-core/plutus-core/stdlib/PlutusCore/StdLib/Data/Data.hs Outdated Show resolved Hide resolved

[Builtins] Add pattern matching builtins

0777db3

effectfully force-pushed the effectfully/builtins/add-pattern-matching-builtins branch from 1d1f1b5 to 0777db3 Compare October 1, 2024 23:21

effectfully temporarily deployed to github-pages October 1, 2024 23:22 — with GitHub Actions Inactive

zliu41 reviewed Oct 2, 2024

View reviewed changes

plutus-core/untyped-plutus-core/src/UntypedPlutusCore/Evaluation/Machine/Cek/Internal.hs Outdated Show resolved Hide resolved

plutus-core/untyped-plutus-core/src/UntypedPlutusCore/Evaluation/Machine/Cek/Internal.hs Outdated Show resolved Hide resolved

Rename 'pushArgs' to 'transferSpine'

0d502ce

effectfully temporarily deployed to github-pages October 2, 2024 01:39 — with GitHub Actions Inactive

ana-pantilie approved these changes Oct 2, 2024

View reviewed changes

plutus-tx/src/PlutusTx/Builtins/Internal.hs Show resolved Hide resolved

plutus-tx/src/PlutusTx/Builtins/Internal.hs Show resolved Hide resolved

effectfully temporarily deployed to github-pages October 2, 2024 19:03 — with GitHub Actions Inactive

zliu41 reviewed Oct 2, 2024

View reviewed changes

Add 'caseList' and 'caseData'

f173b1f

effectfully force-pushed the effectfully/builtins/add-pattern-matching-builtins branch from fe85db0 to f173b1f Compare October 3, 2024 14:57

effectfully temporarily deployed to github-pages October 3, 2024 14:57 — with GitHub Actions Inactive

Add 'Note [ArgStack vs Spine]'

ab6e62f

effectfully temporarily deployed to github-pages October 3, 2024 21:01 — with GitHub Actions Inactive

zliu41 approved these changes Oct 7, 2024

View reviewed changes

effectfully mentioned this pull request Oct 12, 2024

Add support for true higher-order builtins #6563

Open

effectfully added 2 commits October 15, 2024 15:07

Have both via-choose and via-match ways of defining terms

850b751

... for real

e19970a

effectfully mentioned this pull request Oct 17, 2024

Faster processing of Data objects #6225

Open

effectfully added 2 commits October 18, 2024 12:45

Fix benchmarks

0e41c2f

Merge branch 'master' of https://github.com/input-output-hk/plutus in…

cb8e15c

…to effectfully/builtins/add-pattern-matching-builtins

effectfully temporarily deployed to github-pages October 18, 2024 11:14 — with GitHub Actions Inactive

Move 'CaseList' and 'CaseData' to 'futurePV'

63101c9

effectfully temporarily deployed to github-pages October 18, 2024 12:44 — with GitHub Actions Inactive

effectfully merged commit 557a2c7 into master Oct 18, 2024
8 checks passed

effectfully deleted the effectfully/builtins/add-pattern-matching-builtins branch October 18, 2024 13:35

This was referenced Oct 21, 2024

Faster pattern matching for built-in types #5711

Closed

SOP lists aren't faster than built-in lists #5830

Open

[Epic] Internals of builtins #4306

Open

This was referenced Oct 29, 2024

Add caseList and caseData builtins to specification #6604

Open

Metatheory for caseList and caseData #6605

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Builtins] Add pattern matching builtins #6530

[Builtins] Add pattern matching builtins #6530

effectfully commented Oct 1, 2024

effectfully Oct 1, 2024 •

edited

Loading

effectfully Oct 1, 2024

zliu41 Oct 1, 2024

effectfully Oct 1, 2024

effectfully Oct 1, 2024

effectfully Oct 1, 2024

effectfully Oct 1, 2024

effectfully Oct 1, 2024

effectfully Oct 1, 2024

effectfully Oct 1, 2024

effectfully Oct 1, 2024

effectfully commented Oct 1, 2024

ana-pantilie left a comment

effectfully commented Oct 2, 2024

zliu41 left a comment

zliu41 Oct 2, 2024

effectfully Oct 2, 2024

zliu41 left a comment

		@@ -0,0 +1 @@
		all a. list a -> (all r. r -> (a -> list a -> r) -> r)

[Builtins] Add pattern matching builtins #6530

[Builtins] Add pattern matching builtins #6530

Conversation

effectfully commented Oct 1, 2024

effectfully Oct 1, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

effectfully commented Oct 1, 2024

ana-pantilie left a comment

Choose a reason for hiding this comment

effectfully commented Oct 2, 2024

zliu41 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zliu41 left a comment

Choose a reason for hiding this comment

effectfully Oct 1, 2024 •

edited

Loading