[wasm_split] Fix crash when manifest starts with a blank line

When the first line of the manifest file is blank, currFuncs is null,
but will be dereferenced to check if the current module will be empty.

This CL fixes that and also cleans up the logic to clarify how the
state machine works.

Author: sorensp-goog

src/tools/wasm-split/wasm-split.cpp

@@ -424,42 +424,41 @@ void multiSplitModule(const WasmSplitOptions& options) {
   bool newSection = true;
   std::unordered_set<Name> moduleNameSet;
   while (std::getline(manifest, line)) {
-    if (line.empty()) {
-      newSection = true;
-      if (currFuncs->empty() && !options.quiet) {
-        std::cerr << "warning: Module " << currModule << " will be empty\n";
-      }
-      continue;
-    }
-    Name name = line;
     if (newSection) {
-      if (name.endsWith(":")) {
-        name = name.substr(0, name.size() - 1);
-        if (name.size() == 0) {
-          Fatal() << "Module name is empty\n";
-        }
+      Name moduleName = line;
+      if (moduleName.endsWith(":")) {
+        moduleName = moduleName.substr(0, moduleName.size() - 1);
+      }
+      if (moduleName.size() == 0) {
+        Fatal() << "Module name is empty\n";
       }
-      if (moduleNameSet.contains(name)) {
-        Fatal() << "Module name " << name << " is listed more than once\n";
+      if (moduleNameSet.contains(moduleName)) {
+        Fatal() << "Module name " << moduleName
+                << " is listed more than once\n";
       }
-      currModule = name;
+      currModule = moduleName;
       moduleNameSet.insert(currModule);
       config.secondaryNames.push_back(currModule);
       config.secondaryFuncs.emplace_back(std::set<Name>());
       currFuncs = &config.secondaryFuncs.back();
       newSection = false;
-      continue;
-    }
-    assert(currFuncs);
-    if (!canSplitFunc(name, wasm, options)) {
-      continue;
-    }
-    currFuncs->insert(name);
-    auto [it, inserted] = funcModules.insert({name, currModule});
-    if (!inserted && it->second != currModule) {
-      Fatal() << "Function " << name << "cannot be assigned to module "
-              << currModule << "; it is already assigned to module "
-              << it->second << '\n';
+    } else if (line.empty()) {
+      newSection = true;
+      if (currFuncs->empty() && !options.quiet) {
+        std::cerr << "warning: Module " << currModule << " will be empty\n";
+      }
+    } else {
+      assert(currFuncs);
+      Name funcName = line;
+      if (canSplitFunc(funcName, wasm, options)) {
+        currFuncs->insert(funcName);
+        auto [it, inserted] = funcModules.try_emplace(funcName, currModule);
+        if (!inserted && it->second != currModule) {
+          Fatal() << "Function " << funcName << "cannot be assigned to module "
+                  << currModule << "; it is already assigned to module "
+                  << it->second << '\n';
+        }
+      }
     }
   }
 

test/lit/wasm-split/multi-split-errors.wast

@@ -0,0 +1,53 @@
+;; Check that a manifest starting with a blank line produces an error rather
+;; than crashing.
+;; RUN: echo "" > %t.blank.manifest
+;; RUN: echo "1:" >> %t.blank.manifest
+;; RUN: echo "foo" >> %t.blank.manifest
+;; RUN: not wasm-split --multi-split %s --manifest %t.blank.manifest --out-prefix=%t -o %t.wasm 2>&1 \
+;; RUN:   | filecheck %s --check-prefix=EMPTY-NAME
+
+;; Also check a module name consisting only of a colon.
+;; RUN: echo ":" > %t.colon.manifest
+;; RUN: echo "foo" >> %t.colon.manifest
+;; RUN: not wasm-split --multi-split %s --manifest %t.colon.manifest --out-prefix=%t -o %t.wasm 2>&1 \
+;; RUN:   | filecheck %s --check-prefix=EMPTY-NAME
+
+;; Check that an empty module followed by two blank lines warns about the empty
+;; module and then errors on the second blank line.
+;; RUN: echo "1:" > %t.empty-mod.manifest
+;; RUN: echo "" >> %t.empty-mod.manifest
+;; RUN: echo "" >> %t.empty-mod.manifest
+;; RUN: not wasm-split --multi-split %s --manifest %t.empty-mod.manifest --out-prefix=%t -o %t.wasm 2>&1 \
+;; RUN:   | filecheck %s --check-prefix=EMPTY-MOD
+
+;; Check that duplicate module names are rejected.
+;; RUN: echo "1:" > %t.dup-mod.manifest
+;; RUN: echo "foo" >> %t.dup-mod.manifest
+;; RUN: echo "" >> %t.dup-mod.manifest
+;; RUN: echo "1:" >> %t.dup-mod.manifest
+;; RUN: echo "bar" >> %t.dup-mod.manifest
+;; RUN: not wasm-split --multi-split %s --manifest %t.dup-mod.manifest --out-prefix=%t -o %t.wasm 2>&1 \
+;; RUN:   | filecheck %s --check-prefix=DUP-MOD
+
+;; Check that assigning the same function to two modules is rejected.
+;; RUN: echo "1:" > %t.dup-func.manifest
+;; RUN: echo "foo" >> %t.dup-func.manifest
+;; RUN: echo "" >> %t.dup-func.manifest
+;; RUN: echo "2:" >> %t.dup-func.manifest
+;; RUN: echo "foo" >> %t.dup-func.manifest
+;; RUN: not wasm-split --multi-split %s --manifest %t.dup-func.manifest --out-prefix=%t -o %t.wasm 2>&1 \
+;; RUN:   | filecheck %s --check-prefix=DUP-FUNC
+
+;; EMPTY-NAME: Fatal: Module name is empty
+
+;; EMPTY-MOD: warning: Module 1 will be empty
+;; EMPTY-MOD: Fatal: Module name is empty
+
+;; DUP-MOD: Fatal: Module name 1 is listed more than once
+
+;; DUP-FUNC: Fatal: Function foocannot be assigned to module 2; it is already assigned to module 1
+
+(module
+  (func $foo)
+  (func $bar)
+)