Inliner, Break Optimization and Fixing UCS Tail Calls by AnsonYeung · Pull Request #423 · hkust-taco/mlscript

AnsonYeung · 2026-03-21T05:17:23Z

This PR implements inlining. As naive inlining generates unnecessary labels and break, this PR adds break optimization to optimize away useless breaks. The UCS part of #435 is addressed.

Method inlining is not enabled with this PR.

LPTK

The logic looks a bit more complex than I expected. I am curious: what is the main source of complexity?

hkmc2/shared/src/test/mlscript/tailrec/Errors.mls

hkmc2/shared/src/test/mlscript/tailrec/TailRecOpt.mls

hkmc2/shared/src/main/scala/hkmc2/Config.scala

hkmc2/shared/src/main/scala/hkmc2/codegen/inline/Inliner.scala

LPTK · 2026-03-21T07:26:39Z

hkmc2/shared/src/main/scala/hkmc2/codegen/inline/Inliner.scala

+import semantics.Elaborator.State
+
+object Inliner:
+  object TermSymbolPath:


Pls document why we need this strange extractor.

Since the Call node takes in a path, this allow matching for a specific function or method. However method can be virtual and also the this parameter needs to be adjusted for inlining and so method is not actually inlined right now. I added this as the doc:

// Reference to a function body can occur as a.f or f, this handles both cases.

This will also mathc virtual calls, right? But presumably virtual functions won't be in the mapping we're maintaining so that's fine?

Right now method is not considered at all for inlining so it's fine. However I think later if a class have 2 methods f and g where f calls g, then we need to know that g is not virtual / final in order to inline. Inlining methods should be somewhat important since most functions in a program will be methods.

Yes, we can consider method inlining later. Following the previous version of MLscript, I intend to make methods not virtual by default, requiring a virtual modifier to make them so.

hkmc2/shared/src/main/scala/hkmc2/codegen/inline/Inliner.scala

hkmc2/shared/src/test/mlscript/codegen/InnerNameHygiene.mls

AnsonYeung · 2026-03-24T10:59:43Z

The logic looks a bit more complex than I expected. I am curious: what is the main source of complexity?

The main complexity imo is that function definition and usage is not in order, which requires at least 2 traversals. The function could be mutually recursive for example. Even in this case, one function could be inlined into another if one of the function is only used once or is small enough.

Other complexities are symbol renaming, and spread arguments. Multiple chained calls to multiple param list is not supported for inlining currently.

A global function symbol to definition map would simplify this a bit, it would mean no more Define node for functions however.

LPTK · 2026-03-24T13:42:57Z

The main complexity imo is that function definition and usage is not in order, which requires at least 2 traversals. The function could be mutually recursive for example. Even in this case, one function could be inlined into another if one of the function is only used once or is small enough.

Oh, this makes me think about that old GHC paper where they had what looked like a reasonable solution: Secrets of the Glasgow Haskell Compiler inliner. It would be great if you could see whether you could implement something like their solution (sec. 4.2). Would also be a nice paper to present in a group meeting/seminar.

Other complexities are symbol renaming, and spread arguments. Multiple chained calls to multiple param list is not supported for inlining currently.

Would it help to make Call take multiple argument lists? I've been thinking that doing so would make sense, to reflect multiple parameter lists.

A global function symbol to definition map would simplify this a bit, it would mean no more Define node for functions however.

Can't you just define a reusable pass that creates that mapping? We could use it in other places as well, so that would be useful. We can't really do away with Define nodes because functions are scoped, at least nested ones. And carrying around/updating a map all over the place isn't necessarily a win, as far as I can see.

hkmc2/shared/src/main/scala/hkmc2/codegen/deforest/Rewrite.scala

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

hkmc2/shared/src/test/mlscript-compile/Predef.mjs

hkmc2/shared/src/test/mlscript/deforest/todos.mls

LPTK · 2026-04-01T05:10:05Z

When I try to run the tests locally, I get in CamlLightTest.mls:

//│ ═══[RUNTIME ERROR] RangeError: Maximum call stack size exceeded

Does inlining remove tail-recursion optimization opportunities?

hkmc2/shared/src/test/mlscript/backlog/TailRecFailure.mls

hkmc2/shared/src/main/scala/hkmc2/codegen/Lowering.scala

hkmc2/shared/src/test/mlscript-compile/apps/parsing/Lexer.mls

AnsonYeung · 2026-04-02T10:58:45Z

I'm not sure whether SymbolRefresher really handles everything. The logic is quite weird since a symbol can be introduced in many weird ways.

hkmc2/shared/src/main/scala/hkmc2/codegen/Lowering.scala

hkmc2/shared/src/main/scala/hkmc2/codegen/js/JSBuilder.scala

Copilot

Pull request overview

This PR introduces an IR-level inliner in the codegen optimization pipeline and adds control-flow cleanups (notably eliminating redundant breaks/labels) to keep generated IR/JS smaller and to avoid UCS consequent-sharing patterns that accidentally move tail calls out of tail position (addressing the UCS part of #435). It also wires the feature into diff-test configuration with :noInline / :inlineThreshold directives and updates many golden test outputs accordingly.

Changes:

Add configurable inlining (Config.Inliner) and run it from BlockSimplifier, with diff-test directives to toggle/parameterize it.
Add/extend control-flow simplifications (useless-break removal, match/begin simplifications) and UCS normalization tweaks to preserve tail positions under sharing.
Update a large set of expected JS/IR outputs and add new inliner-focused test coverage.

Reviewed changes

Copilot reviewed 71 out of 71 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
hkmc2DiffTests/src/test/scala/hkmc2/MLsDiffMaker.scala	Adds `:noInline` and `:inlineThreshold` directives and wires them into `Config.inlining`.
hkmc2DiffTests/src/test/scala/hkmc2/JSBackendDiffMaker.scala	Adjusts `BlockSimplifier` invocation to provide required contextual parameters.
hkmc2/shared/src/test/mlscript/wasm/ControlFlow.mls	Updates wasm golden output for new control-flow behavior.
hkmc2/shared/src/test/mlscript/ucs/patterns/RestTuple.mls	Updates JS golden output after break/label optimization changes.
hkmc2/shared/src/test/mlscript/ucs/normalization/SimplePairMatches.mls	Updates JS golden output after UCS/codegen control-flow changes.
hkmc2/shared/src/test/mlscript/ucs/normalization/ExcessiveDeduplication.mls	Updates golden outputs; adds a regression-style case around optimization behavior.
hkmc2/shared/src/test/mlscript/ucs/normalization/DeduplicationWhile.mls	Updates while + deduplication golden output.
hkmc2/shared/src/test/mlscript/ucs/normalization/Deduplication.mls	Updates golden outputs for redundant break removal in switches/matches.
hkmc2/shared/src/test/mlscript/ucs/general/JoinPoints.mls	Updates golden outputs reflecting simpler join-point codegen.
hkmc2/shared/src/test/mlscript/ucs/general/BooleanPatterns.mls	Updates switch golden output after break elision.
hkmc2/shared/src/test/mlscript/tailrec/TailRecOpt.mls	Updates tailrec golden output; shows interaction with inlining outputs.
hkmc2/shared/src/test/mlscript/optim/AbortivePrefix.mls	Updates IR golden output for abortive-prefix control flow representation.
hkmc2/shared/src/test/mlscript/opt/DeadVarRemoval.mls	Updates optimized IR golden output (inlined temp/value naming/structure).
hkmc2/shared/src/test/mlscript/opt/DeadSelRemoval.mls	Updates golden output after dead-branch simplification.
hkmc2/shared/src/test/mlscript/opt/DeadObjRemoval.mls	Updates golden output after inlining affects trivial call sites.
hkmc2/shared/src/test/mlscript/llir/ControlFlow.mls	Updates LLIR golden output for simplified match branches.
hkmc2/shared/src/test/mlscript/llir/BasicCpp.mls	Updates C++ golden output for match/default codegen changes.
hkmc2/shared/src/test/mlscript/lifter/Loops.mls	Updates golden output after control-flow simplification in lifted code.
hkmc2/shared/src/test/mlscript/lifter/FunInFun.mls	Updates golden output after control-flow simplification in nested functions.
hkmc2/shared/src/test/mlscript/invalml/InvalMLCodeGen.mls	Updates golden output (redundant breaks removed).
hkmc2/shared/src/test/mlscript/handlers/StackSafety.mls	Updates golden output (breaks removed in handler-generated state machines).
hkmc2/shared/src/test/mlscript/handlers/SavedVars.mls	Updates golden output (breaks removed in handler-generated state machines).
hkmc2/shared/src/test/mlscript/handlers/RecursiveHandlers.mls	Updates golden output (breaks removed in handler-generated state machines).
hkmc2/shared/src/test/mlscript/handlers/NoStackSafety.mls	Updates golden output; shows inlining effects even under warning scenarios.
hkmc2/shared/src/test/mlscript/handlers/NonLocalReturns.mls	Updates expected diagnostics/output; adds TODO note about inliner interaction.
hkmc2/shared/src/test/mlscript/handlers/Effects.mls	Updates handler golden output for break removal and line shifts.
hkmc2/shared/src/test/mlscript/handlers/Debugging.mls	Updates golden stack traces/line numbers and handler control-flow output.
hkmc2/shared/src/test/mlscript/deforest/todos.mls	Updates deforest golden output after symbol refresh/flow changes.
hkmc2/shared/src/test/mlscript/deforest/simple.mls	Updates deforest golden output (name hygiene/ids affected by new passes).
hkmc2/shared/src/test/mlscript/deforest/nestedMatch.mls	Updates deforest golden output (name/id shifts).
hkmc2/shared/src/test/mlscript/deforest/determinism.mls	Updates deforest golden output after symbol refreshing and simplification.
hkmc2/shared/src/test/mlscript/deforest/clashes.mls	Updates deforest golden output (inlined value naming).
hkmc2/shared/src/test/mlscript/deforest/basic.mls	Updates deforest golden output; reflects symbol-refresh and match simplifications.
hkmc2/shared/src/test/mlscript/codegen/Throw.mls	Updates JS golden output for inlining behavior around `throw`.
hkmc2/shared/src/test/mlscript/codegen/SwitchSpecialization.mls	Updates JS golden output after redundant break removal.
hkmc2/shared/src/test/mlscript/codegen/ScopedBlocksAndHandlers.mls	Updates JS golden output; shows inlining affecting handler-local lambdas.
hkmc2/shared/src/test/mlscript/codegen/SanityChecks.mls	Updates sanitized JS golden output; shows inlining at call sites.
hkmc2/shared/src/test/mlscript/codegen/PartialApps.mls	Updates JS golden output (parameter hygiene changes).
hkmc2/shared/src/test/mlscript/codegen/ParamClasses.mls	Updates JS golden output for inlined trivial wrapper calls.
hkmc2/shared/src/test/mlscript/codegen/MergeMatchArms.mls	Updates golden output after join-point/break optimization.
hkmc2/shared/src/test/mlscript/codegen/InnerNameHygiene.mls	Disables inlining for name-hygiene test stability.
hkmc2/shared/src/test/mlscript/codegen/Inliner.mls	New test file covering inliner behavior and limits.
hkmc2/shared/src/test/mlscript/codegen/InlineLambdas.mls	Disables inlining to keep lambda-specific expectations stable.
hkmc2/shared/src/test/mlscript/codegen/Getters.mls	Disables inlining to keep getter-related expectations stable.
hkmc2/shared/src/test/mlscript/codegen/FunInClass.mls	Updates golden output for inlining within class method bodies / closures.
hkmc2/shared/src/test/mlscript/codegen/Functions.mls	Disables inlining for a specific initialization-order regression test.
hkmc2/shared/src/test/mlscript/codegen/FirstClassFunctionTransform.mls	Disables inlining to keep `:checkIR` repro stable; updates line numbers.
hkmc2/shared/src/test/mlscript/codegen/CaseShorthand.mls	Updates JS golden output (scrutinee naming hygiene).
hkmc2/shared/src/test/mlscript/block-staging/Syntax.mls	Updates staged IR golden output after begin/match simplification.
hkmc2/shared/src/test/mlscript/basics/MemberProjections.mls	Disables inlining for a member-projection error repro; updates golden output.
hkmc2/shared/src/test/mlscript/basics/CompanionModules_Functions.mls	Updates expected runtime error output after codegen changes.
hkmc2/shared/src/test/mlscript/backlog/ToTriage.mls	Disables inlining for a tail-position triage case; updates line numbers.
hkmc2/shared/src/test/mlscript/backlog/TailRecFailure.mls	Updates lowered IR golden output demonstrating tailrec fix under sharing.
hkmc2/shared/src/test/mlscript/backlog/Lifter.mls	Disables inlining for a lifter-related repro; updates line numbers.
hkmc2/shared/src/test/mlscript/apps/parsing/CamlLightTest.mls	Moves “example of” output blocks; updates expected transcript positioning.
hkmc2/shared/src/test/mlscript-compile/Predef.mjs	Updates compiled JS runtime output (notably `equals`) to reflect new codegen.
hkmc2/shared/src/test/mlscript-compile/apps/parsing/Lexer.mls	Enables lifting and rewrites some calls to preserve tail-call optimization.
hkmc2/shared/src/main/scala/hkmc2/semantics/ucs/Normalization.scala	Avoids introducing `break`/result-temp when continuation is already `return`/`throw` (tail-position preservation).
hkmc2/shared/src/main/scala/hkmc2/MLsCompiler.scala	Adjusts `BlockSimplifier` invocation to provide proper contextual scoping.
hkmc2/shared/src/main/scala/hkmc2/Config.scala	Adds `Config.Inliner` and new `inlining` config field with default.
hkmc2/shared/src/main/scala/hkmc2/codegen/wasm/text/WatBuilder.scala	Treats missing match default as `nop` during lowering.
hkmc2/shared/src/main/scala/hkmc2/codegen/SymbolRefresher.scala	New utility for refreshing symbols during transformations/inlining.
hkmc2/shared/src/main/scala/hkmc2/codegen/Printer.scala	Adds printing support for `AssignDynField`.
hkmc2/shared/src/main/scala/hkmc2/codegen/Lowering.scala	Updates comment explaining `mayRet` in lowering context.
hkmc2/shared/src/main/scala/hkmc2/codegen/js/JSBuilder.scala	Switch/match codegen tweaks: tolerate no-default, and avoid redundant `break` in abortive arms.
hkmc2/shared/src/main/scala/hkmc2/codegen/deforest/Rewrite.scala	Replaces ad-hoc symbol refresh logic with `SymbolRefresher` and tightens assumptions.
hkmc2/shared/src/main/scala/hkmc2/codegen/BufferableTransform.scala	Refreshes/rebinds parameter symbols while rewriting bufferable classes.
hkmc2/shared/src/main/scala/hkmc2/codegen/BlockTraverser.scala	Traverses disambiguation term symbols in `Value.Ref` for correctness of analyses.
hkmc2/shared/src/main/scala/hkmc2/codegen/BlockTransformer.scala	Routes `Value.Ref` local rewriting through `applyLocal` to support custom remappers.
hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala	Adds inliner and label/break optimizations; drives inlining based on `Config.inlining`.
hkmc2/shared/src/main/scala/hkmc2/codegen/Block.scala	Simplifies `Match`/`Begin` construction when defaults/arms are empty and when rest is empty.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-02T14:22:14Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+      // * Remove useless break
+      case Break(label) if tailLabels.contains(label) =>
+        tl.log(s"Break ${label} is eliminated: current tail label list is ${tailLabels}")
+        registerChange
+        End()


tl.log(...) is used here, but tl is not in scope. BlockSimplifier currently takes an unnamed contextual TL, so this will not compile. Either name the contextual parameter (e.g., (using ..., tl: TL)) or replace these calls with summon[TL].log(...) (and similarly for the other tl.log usages in this file).

tl is a methods defined in hkmc2/shared/src/main/scala/utils/TraceLogger.scala.

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

Copilot · 2026-04-02T14:22:14Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+      def matchArgs(args: List[Arg], params: ParamList): Option[List[(VarSymbol, Result)]] =
+        if args.exists(_.spread.isDefined) then
+          // we require a precise match when any arg is a spread arg
+          if params.restParam.isEmpty then return N
+          if args.exists(_.spread.exists(!_.isEager)) then return N
+          val pairs = args.zip(params.params.iterator.map((_, false)) ++ params.restParam.map((_, true)))
+          if pairs.exists((arg, param) => arg.spread.isDefined =/= param._2) then return N
+          S(pairs.map((arg, param) => (param._1.sym, arg.value)))
+        else


matchArgs claims to require a “precise match when any arg is a spread arg”, but it never checks that args.size matches the expected arity (fixed params + the single rest param). Because zip truncates to the shorter list, extra call arguments can be silently ignored and missing call arguments won’t be detected, which can lead to incorrect inlining/misbinding. Add an explicit size/shape check before zipping (and reject calls with more than one spread).

@AnsonYeung What happens when there is a length mismatch and zip drops some elements from either side? Is the call semantics of the non-inlined version preserved? Please add a comment about that as well as regression tests for both cases.

Copilot · 2026-04-02T14:22:15Z

hkmc2/shared/src/main/scala/hkmc2/codegen/SymbolRefresher.scala

+          case bms: BlockMemberSymbol =>
+            val newBms = new BlockMemberSymbol(bms.nme, Nil, bms.nameIsMeaningful)
+            newBms.tsym = bms.tsym.map: t =>
+              val newOwner = t.owner.map(o => existingMapping.getOrElse(o, o).asInstanceOf)


newOwner is computed via existingMapping.getOrElse(o, o).asInstanceOf (context-inferred cast). If existingMapping ever contains a non-TermSymbol mapping for an owner, this will throw at runtime. Consider using a type-safe match (e.g., existingMapping.get(o).collect { case ts: TermSymbol => ts }.getOrElse(o)) so unexpected mappings are handled explicitly.

Suggested change

val newOwner = t.owner.map(o => existingMapping.getOrElse(o, o).asInstanceOf)

val newOwner = t.owner.map: o =>

existingMapping.get(o) match

case Some(ts: TermSymbol) => ts

case _ => o

The code proposed by Copilot does seem better?

LPTK

Nice work! The code looks pretty clean, although still under-documented.

Please address the comments, and otherwise LGTM.

LPTK · 2026-04-02T14:27:10Z

hkmc2/shared/src/test/mlscript/codegen/Inliner.mls

@@ -0,0 +1,242 @@
+:js
+
+:sjs


Why not use :soir consistently, instead of :sjs? Here, and everywhere else in the PR.

LPTK · 2026-04-02T14:29:20Z

hkmc2/shared/src/test/mlscript/codegen/Inliner.mls

+//│ };
+//│ —————————————————| Output |—————————————————————————————————————————————————————————————————————————
+
+:soir


Suggested change

:soir

// * TODO: support spread arguments like this one

:soir

LPTK · 2026-04-02T15:07:45Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+      // * Remove useless break
+      case Break(label) if tailLabels.contains(label) =>
+        tl.log(s"Break ${label} is eliminated: current tail label list is ${tailLabels}")
+        registerChange
+        End()


tl is a methods defined in hkmc2/shared/src/main/scala/utils/TraceLogger.scala.

LPTK · 2026-04-02T15:10:11Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+      def matchArgs(args: List[Arg], params: ParamList): Option[List[(VarSymbol, Result)]] =
+        if args.exists(_.spread.isDefined) then
+          // we require a precise match when any arg is a spread arg
+          if params.restParam.isEmpty then return N
+          if args.exists(_.spread.exists(!_.isEager)) then return N
+          val pairs = args.zip(params.params.iterator.map((_, false)) ++ params.restParam.map((_, true)))
+          if pairs.exists((arg, param) => arg.spread.isDefined =/= param._2) then return N
+          S(pairs.map((arg, param) => (param._1.sym, arg.value)))
+        else


@AnsonYeung What happens when there is a length mismatch and zip drops some elements from either side? Is the call semantics of the non-inlined version preserved? Please add a comment about that as well as regression tests for both cases.

LPTK · 2026-04-02T15:11:16Z

hkmc2/shared/src/main/scala/hkmc2/codegen/SymbolRefresher.scala

+          case bms: BlockMemberSymbol =>
+            val newBms = new BlockMemberSymbol(bms.nme, Nil, bms.nameIsMeaningful)
+            newBms.tsym = bms.tsym.map: t =>
+              val newOwner = t.owner.map(o => existingMapping.getOrElse(o, o).asInstanceOf)


The code proposed by Copilot does seem better?

LPTK · 2026-04-02T15:19:54Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+              tailLabels += lbl
+              val result = applyBlock(bod)
+              tailLabels -= lbl


Why not use freshLabelCtx/nestLabelCtx?

LPTK · 2026-04-02T15:21:46Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+              if (scrut2 is scrut) &&
+                  (arms2 is arms) &&
+                  (dflt2 is dflt) && (rst2 is rst)


Suggested change

if (scrut2 is scrut) &&

(arms2 is arms) &&

(dflt2 is dflt) && (rst2 is rst)

if (scrut2 is scrut)

&& (arms2 is arms)

&& (dflt2 is dflt)

&& (rst2 is rst)

LPTK · 2026-04-02T15:23:46Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+        if (sub2 is sub) && (rst2 is rst) then b else Begin(sub2, rst2)
+
+      case Match(scrut, arms, dflt, _: End) => super.applyBlock(b)
+      case Match(scrut, arms, dflt, rst) =>


Instead of duplicating this error-prone code, can we just modify BlockTransformer so that it uses an applyMatchArmBlock in these positions, that we can easily override? Try to keep it DRY.

LPTK · 2026-04-02T15:27:32Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+      ):
+        def isPrivate = !symbolsToPreserve.contains(defn.sym)
+
+        def canBeInlineEliminated =


Is this doc correct?

Suggested change

def canBeInlineEliminated =

/** Whether the definition can be removed 1from the program after it has been inlined. */

def canBeInlineEliminated =

or please update with a more accurate doc.

LPTK · 2026-04-02T15:27:49Z

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala

+        def isPrivate = !symbolsToPreserve.contains(defn.sym)
+
+        def canBeInlineEliminated =
+          isPrivate && !isMethod && useCount <= 1 && !hasNakedRef && !isLoopBreaker


Can't we eliminate functions used several times?

AnsonYeung added 5 commits March 19, 2026 11:01

Inliner WIP

b6aa5ac

Basic inlining

b78ac11

Fixes Scoped block handling

f1e81e4

Optimize away Label in trivial case

a292d3d

Revert HkScratch.mls

8610f20

LPTK reviewed Mar 21, 2026

View reviewed changes

AnsonYeung added 6 commits March 23, 2026 11:39

some changes without running test

6272af3

Update tests

4ed32b4

Merge remote-tracking branch 'upstream/hkmc2' into inliner

f580955

Tests

907fc0b

Minor changes

47e7b40

Fix test failures

667ce54

AnsonYeung added 9 commits March 24, 2026 23:20

Move Inliner

058aca4

Refactor SymbolRefresher to its own file

9a32553

Do cycle checking

35453ce

Use shared SymbolRefresher

61fe1ad

All test passes

151013b

Break optimization

26ba6dd

Test updates

904e600

Merge remote-tracking branch 'upstream/hkmc2' into inliner

5276935

Rename config

d3244e4

LPTK reviewed Apr 1, 2026

View reviewed changes

hkmc2/shared/src/main/scala/hkmc2/codegen/deforest/Rewrite.scala Show resolved Hide resolved

LPTK reviewed Apr 1, 2026

View reviewed changes

hkmc2/shared/src/main/scala/hkmc2/codegen/deforest/Rewrite.scala Outdated Show resolved Hide resolved

LPTK reviewed Apr 1, 2026

View reviewed changes

hkmc2/shared/src/main/scala/hkmc2/codegen/BlockSimplifier.scala Outdated Show resolved Hide resolved

LPTK reviewed Apr 1, 2026

View reviewed changes

hkmc2/shared/src/test/mlscript-compile/Predef.mjs Outdated Show resolved Hide resolved

LPTK reviewed Apr 1, 2026

View reviewed changes

hkmc2/shared/src/test/mlscript/deforest/todos.mls Outdated Show resolved Hide resolved

Remove useless breaks from switch statements

120e090

LPTK and others added 2 commits April 1, 2026 13:58

Remove useless nested and match arm/default blocks

11d43f8

ucs return

5bd22ba

LPTK reviewed Apr 1, 2026

View reviewed changes

hkmc2/shared/src/test/mlscript/backlog/TailRecFailure.mls Show resolved Hide resolved

AnsonYeung added 3 commits April 1, 2026 23:46

Remove unnecessary :noInline

8f977ee

Allow Thrw to be optimized

d2f2388

Style changes

3905b2b

AnsonYeung changed the title ~~Inliner~~ Inliner, Break Optimization and Fixing UCS Tail Calls Apr 1, 2026

AnsonYeung added 3 commits April 2, 2026 17:08

Add comment

f946f2a

Merge remote-tracking branch 'upstream/hkmc2' into inliner

d97b687

Stack overflow fix

926941b

LPTK reviewed Apr 2, 2026

View reviewed changes

hkmc2/shared/src/main/scala/hkmc2/codegen/Lowering.scala Outdated Show resolved Hide resolved

LPTK reviewed Apr 2, 2026

View reviewed changes

hkmc2/shared/src/test/mlscript-compile/apps/parsing/Lexer.mls Show resolved Hide resolved

LPTK reviewed Apr 2, 2026

View reviewed changes

hkmc2/shared/src/test/mlscript-compile/apps/parsing/Lexer.mls Show resolved Hide resolved

Apply changes

eb76214

AnsonYeung marked this pull request as ready for review April 2, 2026 10:57

LPTK reviewed Apr 2, 2026

View reviewed changes

hkmc2/shared/src/main/scala/hkmc2/codegen/Lowering.scala Outdated Show resolved Hide resolved

LPTK and others added 2 commits April 2, 2026 20:15

Merge branch 'hkmc2' into inliner

b63d567

Fix grammar

0d8b6ad

CAG2Mark reviewed Apr 2, 2026

View reviewed changes

hkmc2/shared/src/main/scala/hkmc2/codegen/js/JSBuilder.scala Outdated Show resolved Hide resolved

Fix switch specialization regression

cf06569

LPTK requested a review from Copilot April 2, 2026 14:10

Copilot started reviewing on behalf of LPTK April 2, 2026 14:10 View session

Copilot AI reviewed Apr 2, 2026

View reviewed changes

Make various improvements, including Block flattening in BlockSimplifier

2b30539

LPTK approved these changes Apr 2, 2026

View reviewed changes

-              val newOwner = t.owner.map(o => existingMapping.getOrElse(o, o).asInstanceOf)
+              val newOwner = t.owner.map: o =>
+                existingMapping.get(o) match
+                  case Some(ts: TermSymbol) => ts
+                  case _ => o

	def canBeInlineEliminated =
	/** Whether the definition can be removed 1from the program after it has been inlined. */
	def canBeInlineEliminated =

		@@ -0,0 +1,242 @@
		:js

		:sjs

Conversation

AnsonYeung commented Mar 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

LPTK left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

AnsonYeung commented Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

LPTK commented Mar 24, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

LPTK commented Apr 1, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

AnsonYeung commented Apr 2, 2026

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

LPTK left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

AnsonYeung commented Mar 21, 2026 •

edited

Loading

AnsonYeung commented Mar 24, 2026 •

edited

Loading