[Hanabi9248]

The explicit charset in the string codecs is currently used for both the payload and the Base64 carrier text. With UTF-16 this turns encoded ASCII bytes into unrelated characters and introduces BOM/zero bytes when decoding canonical Base64 input; even an encode/decode round trip can fail.

Use US_ASCII for the Base64 text in the basic and URL-safe string codecs. Keep the caller-selected charset for original payload bytes and recovered text. For example, UTF-16 encoding of A must produce /v8AQQ== (basic) or _v8AQQ== (URL-safe), preserving the payload BOM. JsonBuilder and other helpers are unchanged.

Validation: compiled the complete production Base64Util plus six unchanged support classes against the real, already-installed Spring Core 5.2.10.RELEASE JAR, using JDK 25.0.4.101 with javac --release 8. The standalone regression uses JDK java.util.Base64 as an independent oracle. Baseline: 8 failures and 4 passing controls; candidate: all 12 cases pass. Both alphabets cover UTF-16 encoding, canonical decoding and round trips, UTF-16LE encoding, default UTF-8 Chinese text and explicit Latin-1 controls. Three obsolete Java 8 source/target warnings were emitted.

Example in Bash, from the repository root with SPRING_CORE_JAR pointing to an existing matching JAR:

mkdir -p "$TMPDIR/base64-classes"
javac --release 8 -encoding UTF-8 -cp "$SPRING_CORE_JAR" -d "$TMPDIR/base64-classes" common/src/main/java/com/highgo/cloud/util/toolkit/{Base64Util,StringUtil,ObjectUtil}.java common/src/main/java/com/highgo/cloud/util/toolkit/support/{Charsets,CharPool,RandomType,StringPool}.java tests/Base64CharsetRegression.java
java -cp "$TMPDIR/base64-classes:$SPRING_CORE_JAR" Base64CharsetRegression

Use a semicolon classpath separator on Windows. Java 8 runtime, Maven/Spring full-service tests, the whole repository and arbitrary charsets/invalid-input policies were not tested. This validates the public codec overloads, without claiming a specific application endpoint.

Assisted-by: OpenAI gpt-6.1-sol