[GR-18163] setlocale() cleanup after fix in Prism and fix warnings

PullRequest: truffleruby/4232

Author: eregon

mx.truffleruby/suite.py

@@ -152,6 +152,7 @@
             "requires": ["java.management"],
             "dependencies": [
                 "truffleruby:TRUFFLERUBY-ANNOTATIONS",
+                "sdk:NATIVEIMAGE",
                 "sdk:POLYGLOT",
             ],
             "annotationProcessors": [
@@ -261,6 +262,9 @@
                 "sulong:SULONG_API",
                 "sulong:SULONG_NFI",
                 "sdk:JLINE3",
+                "sdk:COLLECTIONS",
+                "sdk:NATIVEIMAGE",
+                "sdk:POLYGLOT",
                 # Library distributions, keep in sync with truffle_jars in mx_truffleruby.py
                 "truffle:TRUFFLE_JCODINGS",
                 "truffleruby:TRUFFLERUBY_JONI",
@@ -311,6 +315,7 @@
                 "sdk:POLYGLOT",
                 "sdk:LAUNCHER_COMMON",
                 "sdk:MAVEN_DOWNLOADER",
+                "sdk:NATIVEIMAGE",
             ],
             "jacoco": "include",
             "javaCompliance": "17+",
@@ -489,6 +494,7 @@
             ],
             "distDependencies": [
                 "truffleruby:TRUFFLERUBY-ANNOTATIONS",
+                "sdk:NATIVEIMAGE",
                 "sdk:POLYGLOT",
             ],
             "description": "TruffleRuby Shared constants and predicates",
@@ -531,6 +537,9 @@
                 "sulong:SULONG_API",
                 "sulong:SULONG_NFI",
                 "sdk:JLINE3",
+                "sdk:COLLECTIONS",
+                "sdk:NATIVEIMAGE",
+                "sdk:POLYGLOT",
                 # Library distributions, keep in sync with truffle_jars in mx_truffleruby.py
                 "truffle:TRUFFLE_JCODINGS",
                 "truffleruby:TRUFFLERUBY_JONI",
@@ -600,6 +609,7 @@
                 "sdk:POLYGLOT",
                 "sdk:LAUNCHER_COMMON",
                 "sdk:MAVEN_DOWNLOADER",
+                "sdk:NATIVEIMAGE",
             ],
             "description": "TruffleRuby Launcher",
             "license": ["EPL-2.0"],

src/main/java/org/truffleruby/RubyLanguage.java

@@ -39,6 +39,8 @@
 import com.oracle.truffle.api.strings.AbstractTruffleString;
 import com.oracle.truffle.api.strings.InternalByteArray;
 import com.oracle.truffle.api.strings.TruffleString;
+import org.graalvm.nativeimage.ImageInfo;
+import org.graalvm.nativeimage.ProcessProperties;
 import org.graalvm.options.OptionDescriptors;
 import org.prism.Parser;
 import org.truffleruby.annotations.SuppressFBWarnings;
@@ -125,6 +127,7 @@
 import org.truffleruby.shared.Metrics;
 import org.truffleruby.shared.TruffleRuby;
 import org.truffleruby.shared.options.OptionsCatalog;
+import org.truffleruby.signal.LibRubySignal;
 import org.truffleruby.stdlib.CoverageManager;
 
 import com.oracle.truffle.api.Assumption;
@@ -159,8 +162,7 @@
                 RubyLanguage.MIME_TYPE_MAIN_SCRIPT },
         defaultMimeType = RubyLanguage.MIME_TYPE,
         dependentLanguages = { "nfi", "llvm", "regex" },
-        fileTypeDetectors = RubyFileTypeDetector.class,
-        needsAllEncodings = true)
+        fileTypeDetectors = RubyFileTypeDetector.class)
 @ProvidedTags({
         CoverageManager.LineTag.class,
         TraceManager.CallTag.class,
@@ -459,6 +461,7 @@ public RubyContext createContext(Env env) {
                 this.allocationReporter = env.lookup(AllocationReporter.class);
                 this.options = new LanguageOptions(env, env.getOptions(), singleContext);
                 setRubyHome(findRubyHome(env));
+                setupLocale(env, rubyHome);
                 loadLibYARPBindings();
                 this.coreLoadPath = buildCoreLoadPath(this.options.CORE_LOAD_PATH);
                 this.corePath = coreLoadPath + File.separator + "core" + File.separator;
@@ -540,6 +543,7 @@ protected boolean patchContext(RubyContext context, Env newEnv) {
 
         synchronized (this) {
             setRubyHome(findRubyHome(newEnv));
+            setupLocale(newEnv, rubyHome);
             loadLibYARPBindings();
             setupCleaner();
         }
@@ -820,6 +824,24 @@ private boolean isRubyHome(TruffleFile path) {
                 lib.resolve("patches").isDirectory();
     }
 
+    private void setupLocale(Env env, String rubyHome) {
+        // CRuby does setlocale(LC_CTYPE, "") because this is needed to get the locale encoding with nl_langinfo(CODESET).
+        // This means every locale category except LC_CTYPE remains the initial "C".
+        // LC_CTYPE is set according to environment variables (LC_ALL, LC_CTYPE, LANG).
+        // HotSpot does setlocale(LC_ALL, "") and Native Image does nothing.
+        // We match CRuby by doing setlocale(LC_ALL, "C") and setlocale(LC_CTYPE, "").
+        // This also affects C functions that depend on the locale in C extensions, so best to follow CRuby here.
+        // Change the strict minimum if embedded because setlocale() is process-wide.
+        if (env.getOptions().get(OptionsCatalog.EMBEDDED_KEY)) {
+            if (ImageInfo.inImageRuntimeCode()) {
+                ProcessProperties.setLocale("LC_CTYPE", "");
+            }
+        } else {
+            LibRubySignal.loadLibrary(rubyHome, Platform.LIB_SUFFIX);
+            LibRubySignal.setupLocale();
+        }
+    }
+
     private void loadLibYARPBindings() {
         String libyarpbindings = getRubyHome() + "/lib/libyarpbindings" + Platform.LIB_SUFFIX;
         Parser.loadLibrary(libyarpbindings);

src/main/java/org/truffleruby/core/DataObjectFinalizationService.java

@@ -71,6 +71,7 @@ public DataObjectFinalizationService(RubyLanguage language, ReferenceProcessor r
         this(language, referenceProcessor.processingQueue);
     }
 
+    @TruffleBoundary
     public DataObjectFinalizerReference addFinalizer(RubyContext context, Object object, Object finalizerCFunction,
             Object dataStruct) {
         final DataObjectFinalizerReference newRef = createRef(object, finalizerCFunction, dataStruct);

src/main/java/org/truffleruby/core/ReferenceProcessingService.java

@@ -78,6 +78,7 @@ public ReferenceProcessor(RubyContext context) {
             this.context = context;
         }
 
+        @TruffleBoundary
         void processReferenceQueue(ReferenceProcessingService<?, ?> service) {
             if (processOnMainThread()) {
                 drainReferenceQueues();

src/main/java/org/truffleruby/core/encoding/EncodingManager.java

@@ -37,8 +37,6 @@
 
 import com.oracle.truffle.api.CompilerDirectives.CompilationFinal;
 import com.oracle.truffle.api.CompilerDirectives.TruffleBoundary;
-import org.truffleruby.shared.Platform;
-import org.truffleruby.signal.LibRubySignal;
 
 import static org.truffleruby.core.encoding.Encodings.INITIAL_NUMBER_OF_ENCODINGS;
 
@@ -129,22 +127,12 @@ public void initializeDefaultEncodings(TruffleNFIPlatform nfi, NativeConfigurati
     }
 
     private void initializeLocaleEncoding(TruffleNFIPlatform nfi, NativeConfiguration nativeConfiguration) {
-        // CRuby does setlocale(LC_CTYPE, "") because this is needed to get the locale encoding with nl_langinfo(CODESET).
-        // This means every locale category except LC_CTYPE remains the initial "C".
-        // LC_CTYPE is set according to environment variables (LC_ALL, LC_CTYPE, LANG).
-        // HotSpot does setlocale(LC_ALL, "") and Native Image does nothing.
-        // We match CRuby by doing setlocale(LC_ALL, "C") and setlocale(LC_CTYPE, "").
-        // This is notably important for Prism to be able to parse floating-point numbers:
-        // https://github.com/ruby/prism/issues/2638
-        // It also affects C functions that depend on the locale in C extensions, so best to follow CRuby here.
-        LibRubySignal.loadLibrary(language.getRubyHome(), Platform.LIB_SUFFIX);
-        LibRubySignal.setupLocale();
-
         final String localeEncodingName;
         final String detector;
         if (nfi != null) {
             final int codeset = (int) nativeConfiguration.get("platform.langinfo.CODESET");
 
+            // nl_langinfo() needs setlocale(LC_CTYPE, "") before, which is done in RubyLanguage#setupLocale()
             // char *nl_langinfo(nl_item item);
             // nl_item is int on at least Linux and macOS
             final Object nl_langinfo = nfi.getFunction(context, "nl_langinfo", "(sint32):string");

src/main/java/org/truffleruby/language/backtrace/BacktraceFormatter.java

@@ -69,7 +69,7 @@ public static BacktraceFormatter createDefaultFormatter(RubyContext context, Rub
     // For debugging:
     // org.truffleruby.language.backtrace.BacktraceFormatter.printableRubyBacktrace(this)
     public static String printableRubyBacktrace(Object maybeNode) {
-        final Node node = maybeNode instanceof Node ? (Node) maybeNode : null;
+        final Node node = maybeNode instanceof Node n && n.isAdoptable() ? (Node) maybeNode : null;
         final RubyContext context = RubyLanguage.getCurrentContext();
         final BacktraceFormatter backtraceFormatter = new BacktraceFormatter(
                 context,
@@ -79,7 +79,7 @@ public static String printableRubyBacktrace(Object maybeNode) {
         if (backtrace.isEmpty()) {
             return "<empty backtrace>";
         } else if (node == null) {
-            return "# the first entry line is imprecise because 'this' is not a Node, select caller Java frames in the debugger until 'this' is a Node to fix this\n" +
+            return "# the first entry line is imprecise because 'this' is not an adopted Node, select caller Java frames in the debugger until 'this' is an adopted Node to fix this\n" +
                     backtrace;
         } else {
             return backtrace;

src/options.yml

@@ -48,6 +48,7 @@ LANGUAGE_OPTIONS:
 - CLASS_CACHE
 - ARRAY_DUP_CACHE
 - ARRAY_STRATEGY_CACHE
+# Other
 - EXPERIMENTAL_ENGINE_CACHING
 - RUN_TWICE
 - CHECK_CLONE_UNINITIALIZED_CORRECTNESS
@@ -85,6 +86,8 @@ EXPERT:
     LAZY_RUBYGEMS: [lazy-rubygems, boolean, ['RUBYGEMS && ', DEFAULT_LAZY], Load RubyGems lazily on first failing require]
 
     # Embedding options
+    # It would be tempting to make these language options, but that does not work because the pre-initialized context is created with (isNativeAccessAllowed=false, embedded=true)
+    # and it must be compatible (to be reused) with the runtime context created from the truffleruby launcher, which is (isNativeAccessAllowed=true, embedded=false).
     EMBEDDED: [embedded, boolean, true, 'Set default options for an embedded use of TruffleRuby, rather than top-level use']
     NATIVE_PLATFORM: [platform-native, boolean, ['env.isNativeAccessAllowed() && ', true], Enables native calls via Truffle NFI for internal functionality]
     NATIVE_INTERRUPT: [platform-native-interrupt, boolean, NATIVE_PLATFORM, Use the SIGVTALRM signal to interrupt native blocking calls]