Set only LC_CTYPE according to env, keep the "C" locale for other locale categories

* Fixes #3512

Author: eregon

CHANGELOG.md

@@ -8,6 +8,7 @@ Bug fixes:
 * Fix repeated calling of methods `Dir#{each,each_child,children}` (#3464, @andrykonchin).
 * Fix `IO#{wait,wait_readable,wait_writable}` methods and switch the current thread into a sleep state (@andrykonchin).
 * Fix `rb_global_variable()` for `Float` and bignum values during the `Init_` function (#3478, @eregon).
+* Fix parsing literal floats when the locale does not use `.` for the decimal separator (e.g. `LANG=fr_FR.UTF-8`) (#3512, @eregon).
 
 Compatibility:
 * Move `IO#wait_readable`, `IO#wait_writable`, `IO#wait_priority` and `IO#wait` into core library (@andrykonchin).

src/main/c/rubysignal/src/rubysignal.c

@@ -8,13 +8,19 @@
  * GNU Lesser General Public License version 2.1.
  */
 #include "org_truffleruby_signal_LibRubySignal.h"
+#include <locale.h>
 #include <pthread.h>
 #include <signal.h>
 #include <unistd.h>
 #include <sys/syscall.h>
 
 _Static_assert(sizeof(pthread_t) == sizeof(jlong), "Expected sizeof(pthread_t) == sizeof(jlong)");
 
+JNIEXPORT void JNICALL Java_org_truffleruby_signal_LibRubySignal_setupLocale(JNIEnv *env, jclass clazz) {
+  setlocale(LC_ALL, "C");
+  setlocale(LC_CTYPE, "");
+}
+
 static void empty_handler(int sig) {
 }
 

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

@@ -22,8 +22,6 @@
 import com.oracle.truffle.api.CompilerDirectives;
 import com.oracle.truffle.api.interop.InteropException;
 import com.oracle.truffle.api.interop.InteropLibrary;
-import org.graalvm.nativeimage.ImageInfo;
-import org.graalvm.nativeimage.ProcessProperties;
 import org.graalvm.shadowed.org.jcodings.Encoding;
 import org.graalvm.shadowed.org.jcodings.EncodingDB;
 import org.truffleruby.RubyContext;
@@ -39,6 +37,8 @@
 
 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;
 
@@ -129,11 +129,16 @@ public void initializeDefaultEncodings(TruffleNFIPlatform nfi, NativeConfigurati
     }
 
     private void initializeLocaleEncoding(TruffleNFIPlatform nfi, NativeConfiguration nativeConfiguration) {
-        if (ImageInfo.inImageRuntimeCode()) {
-            // Call setlocale(LC_ALL, "") to ensure the locale is set to the environment's locale
-            // rather than the default "C" locale.
-            ProcessProperties.setLocale("LC_ALL", "");
-        }
+        // 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;

src/signal/java/org/truffleruby/signal/LibRubySignal.java

@@ -16,6 +16,8 @@ public static void loadLibrary(String rubyHome, String libSuffix) {
         System.load(path);
     }
 
+    public static native void setupLocale();
+
     public static native int setupSIGVTALRMEmptySignalHandler();
 
     public static native long threadID();