Skip to content

[wasm-merge] Handle merging of start functions with control flow #8697

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
kripken merged 9 commits into from
May 14, 2026
+102 −37
Merged

[wasm-merge] Handle merging of start functions with control flow #8697

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
c7c4f4f
work
kripken May 13, 2026
e528814
fix
kripken May 13, 2026
cd1c94a
format
kripken May 13, 2026
f2e8f2c
format
kripken May 13, 2026
bd7c565
update
kripken May 13, 2026
8cf214a
comment
kripken May 13, 2026
4274deb
accumulate
kripken May 13, 2026
563dcb6
Update src/tools/wasm-merge.cpp
kripken May 13, 2026
ae39945
Update src/tools/wasm-merge.cpp
kripken May 13, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
68 changes: 46 additions & 22 deletions src/tools/wasm-merge.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,6 +117,11 @@ namespace {
// have it as a global rather than pass it around all the time.
Module merged;

// Everything we merge is accumulated into |merged|, aside from the start
// functions. To avoid incrementally adding a call each time, which can end up
// nested, we add them here and generate a series of flat calls at the end.
std::vector<Name> startFunctions;

// Name conflicts on functions etc. are resolved by renaming things in a way
// that only matters internally. Conflicting export names, however, are
// observable, and so the user must decide how they want wasm-merge to handle
Expand Down Expand Up @@ -363,28 +368,9 @@ void copyModuleContents(Module& input, Name inputName) {
merged.addExport(std::move(copy));
}

// Start functions must be merged.
if (input.start.is()) {
if (!merged.start.is()) {
// No previous start; just refer to the new one.
merged.start = input.start;
} else {
// Merge them, keeping the order. We copy both functions to avoid issues
// with other references to them, and just call the second one, leaving
// inlining to the optimizer if that makes sense to do.
auto copiedOldName =
Names::getValidFunctionName(merged, "merged.start.old");
auto copiedNewName =
Names::getValidFunctionName(merged, "merged.start.new");
auto* copiedOld = ModuleUtils::copyFunction(
merged.getFunction(merged.start), merged, copiedOldName);
ModuleUtils::copyFunction(
merged.getFunction(input.start), merged, copiedNewName);
Builder builder(merged);
copiedOld->body = builder.makeSequence(
copiedOld->body, builder.makeCall(copiedNewName, {}, Type::none));
merged.start = copiedOldName;
}
// Start functions are accumulated till the end.
if (input.start) {
startFunctions.push_back(input.start);
Copy link
Copy Markdown
Member

@tlively tlively May 13, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there an edge case here where the start function name has changed to avoid conflicts?

Copy link
Copy Markdown
Member Author

@kripken kripken May 13, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The names at this point are all fixed up. We have tests for this too.

The basic workflow is

void mergeInto(Module& input, Name inputName) {
  // Rename things in the input module so that there are no conflicts with names
  // in the merged module. We do so in place for efficiency.
  renameInputItems(input);

  // The input module's items can now be copied into the target module safely,
  // as names will not conflict.
  copyModuleContents(input, inputName);
}

So we first rename to avoid conflicts, then merge. This code change is in the merge.

}

// TODO: type names, features, debug info, custom sections, dylink info, etc.
Expand Down Expand Up @@ -596,6 +582,34 @@ void updateTypes(Module& wasm) {
updater.runOnModuleCode(&runner, &wasm);
}

// Merge the start functions, keeping the order. We add a new function that
// calls them in sequence (leaving proper inlining, including handling of
// control flow etc., to the optimizer).
void mergeStartFunctions() {
if (startFunctions.empty()) {
return;
}

if (startFunctions.size() == 1) {
// Avoid adding a call here.
merged.start = startFunctions[0];
return;
}

auto combinedName =
Names::getValidFunctionName(merged, "merged.start.combined");
Builder builder(merged);
std::vector<Expression*> calls;
for (auto start : startFunctions) {
calls.push_back(builder.makeCall(start, {}, Type::none));
}
auto* body = builder.makeBlock(calls);
auto combined = builder.makeFunction(
combinedName, Signature(Type::none, Type::none), {}, body);
merged.addFunction(std::move(combined));
merged.start = combinedName;
}

// Merges an input module into an existing target module. The input module can
// be modified, as it will no longer be needed (so it is intentionally not
// marked as const here).
Expand Down Expand Up @@ -794,6 +808,13 @@ Input source maps can be specified by adding an -ism option right after the modu
for (auto& curr : merged.exports) {
exportModuleMap[curr.get()] = ExportInfo{inputFileName, curr->name};
}

// Start functions are accumulated till the end.
if (merged.start) {
startFunctions.push_back(merged.start);
merged.start = Name();
}

} else {
// This is a later module: do a full merge.
mergeInto(*currModule, inputFileName);
Expand Down Expand Up @@ -827,6 +848,9 @@ Input source maps can be specified by adding an -ism option right after the modu
// Update types after combing and linking everything.
updateTypes(merged);

// Merge the start functions, after everything else is set up.
mergeStartFunctions();

{
PassRunner passRunner(&merged);
// We might have made some globals read from others that now appear after
Expand Down
39 changes: 39 additions & 0 deletions test/lit/merge/start-return.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,39 @@
;; NOTE: Assertions have been generated by update_lit_checks.py --all-items and should not be edited.
;; RUN: wasm-merge %s first %s.second second --rename-export-conflicts -all -S -o - | filecheck %s

;; Test that we properly merge start functions with control flow. The two
;; start functions have returns, and a naive merge of their bodies would end up
;; skipping the second (after the first return).

(module
(start $start-a)

;; CHECK: (type $0 (func))

;; CHECK: (start $merged.start.combined)

;; CHECK: (func $start-a (type $0)
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i32.const 0)
;; CHECK-NEXT: )
;; CHECK-NEXT: (return)
;; CHECK-NEXT: )
(func $start-a
(drop
(i32.const 0)
)
(return)
)
)

;; CHECK: (func $start-b (type $0)
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i32.const 1)
;; CHECK-NEXT: )
;; CHECK-NEXT: (return)
;; CHECK-NEXT: )

;; CHECK: (func $merged.start.combined (type $0)
;; CHECK-NEXT: (call $start-a)
;; CHECK-NEXT: (call $start-b)
;; CHECK-NEXT: )
10 changes: 10 additions & 0 deletions test/lit/merge/start-return.wat.second
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
(module
(start $start-b)

(func $start-b
(drop
(i32.const 1)
)
(return)
)
)
22 changes: 7 additions & 15 deletions test/lit/merge/start3.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@

;; CHECK: (export "user" (func $user))

;; CHECK: (start $merged.start.old)
;; CHECK: (start $merged.start.combined)

;; CHECK: (func $start (type $0)
;; CHECK-NEXT: (local $x i32)
Expand All @@ -30,20 +30,7 @@
;; CHECK-NEXT: (call $start)
;; CHECK-NEXT: )

;; CHECK: (func $merged.start.old (type $0)
;; CHECK-NEXT: (local $x i32)
;; CHECK-NEXT: (block
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i32.const 1)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (call $merged.start.new)
;; CHECK-NEXT: )

;; CHECK: (func $merged.start.new (type $0)
;; CHECK: (func $start_2 (type $0)
;; CHECK-NEXT: (local $x f64)
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (local.get $x)
Expand All @@ -52,3 +39,8 @@
;; CHECK-NEXT: (i32.const 2)
;; CHECK-NEXT: )
;; CHECK-NEXT: )

;; CHECK: (func $merged.start.combined (type $0)
;; CHECK-NEXT: (call $start)
;; CHECK-NEXT: (call $start_2)
;; CHECK-NEXT: )
Loading