An alts release built with JDK 9+ wouldn't work under Java 7 or 8

[cpovirk]

From #6829 (comment), I gather that your process may be to build releases with JDK 8. If so, then this is more about convenience for anyone else who might do a local build. (Compare #6839, which dealt with similar calls in core.)

My attempt to turn up other such calls shows some in alts:

$ ./gradlew publishToMavenLocal

$ printf '%s\n' ~/.m2/repository/io/grpc/*/1.32.0-SNAPSHOT/*-SNAPSHOT.jar | parallel -q --gnu --will-cite -k bash -c 'jar -tf $1 | grep -oP ".*(?=[.]class$)" | xargs -r javap -private -v -cp $1' - > /tmp/allclasses

$ ( R='Method java/nio/\w+Buffer[.](clear|flip|limit:[(]I|mark|position:[(]I|reset|rewind).*[)]Ljava/nio/\w+Buffer;'; grep -P "$R|^Classfile" /tmp/allclasses | grep -P -B 1 "$R" )
Classfile jar:file:///usr/local/google/home/cpovirk/.m2/repository/io/grpc/grpc-alts/1.32.0-SNAPSHOT/grpc-alts-1.32.0-SNAPSHOT.jar!/io/grpc/alts/internal/AltsTsiHandshaker.class
        66: invokevirtual #163                // Method java/nio/ByteBuffer.limit:(I)Ljava/nio/ByteBuffer;
        84: invokevirtual #168                // Method java/nio/ByteBuffer.position:(I)Ljava/nio/ByteBuffer;
--
Classfile jar:file:///usr/local/google/home/cpovirk/.m2/repository/io/grpc/grpc-alts/1.32.0-SNAPSHOT/grpc-alts-1.32.0-SNAPSHOT.jar!/io/grpc/alts/internal/AltsFraming.class
        38: invokevirtual #59                 // Method java/nio/ByteBuffer.limit:(I)Ljava/nio/ByteBuffer;
        57: invokevirtual #68                 // Method java/nio/ByteBuffer.position:(I)Ljava/nio/ByteBuffer;
        53: invokevirtual #59                 // Method java/nio/ByteBuffer.limit:(I)Ljava/nio/ByteBuffer;
        70: invokevirtual #68                 // Method java/nio/ByteBuffer.position:(I)Ljava/nio/ByteBuffer;
--
Classfile jar:file:///usr/local/google/home/cpovirk/.m2/repository/io/grpc/grpc-alts/1.32.0-SNAPSHOT/grpc-alts-1.32.0-SNAPSHOT.jar!/io/grpc/alts/internal/AltsHandshakerClient.class
        70: invokevirtual #399                // Method java/nio/ByteBuffer.position:(I)Ljava/nio/ByteBuffer;
        83: invokevirtual #399                // Method java/nio/ByteBuffer.position:(I)Ljava/nio/ByteBuffer;
--
Classfile jar:file:///usr/local/google/home/cpovirk/.m2/repository/io/grpc/grpc-alts/1.32.0-SNAPSHOT/grpc-alts-1.32.0-SNAPSHOT.jar!/io/grpc/alts/internal/AltsFraming$Parser.class
        74: invokevirtual #72                 // Method java/nio/ByteBuffer.flip:()Ljava/nio/ByteBuffer;
       184: invokevirtual #118                // Method java/nio/ByteBuffer.limit:(I)Ljava/nio/ByteBuffer;
       210: invokevirtual #72                 // Method java/nio/ByteBuffer.flip:()Ljava/nio/ByteBuffer;
         4: invokevirtual #132                // Method java/nio/ByteBuffer.clear:()Ljava/nio/ByteBuffer;
--
Classfile jar:file:///usr/local/google/home/cpovirk/.m2/repository/io/grpc/grpc-alts/1.32.0-SNAPSHOT/grpc-alts-1.32.0-SNAPSHOT.jar!/io/grpc/alts/internal/AltsFraming$Producer.class
        25: invokevirtual #78                 // Method java/nio/ByteBuffer.flip:()Ljava/nio/ByteBuffer;
        45: invokevirtual #83                 // Method java/nio/ByteBuffer.limit:(I)Ljava/nio/ByteBuffer;
        88: invokevirtual #98                 // Method java/nio/ByteBuffer.position:(I)Ljava/nio/ByteBuffer;
         4: invokevirtual #103                // Method java/nio/ByteBuffer.clear:()Ljava/nio/ByteBuffer;
        16: invokevirtual #98                 // Method java/nio/ByteBuffer.position:(I)Ljava/nio/ByteBuffer;
        36: invokevirtual #83                 // Method java/nio/ByteBuffer.limit:(I)Ljava/nio/ByteBuffer;

Again, there is no problem with 1.31.1, which I assume you built with JDK 8:

$ wget https://repo1.maven.org/maven2/io/grpc/grpc-alts/1.31.1/grpc-alts-1.31.1.jar

$ javap -private -v -cp grpc-alts-1.31.1.jar 'io.grpc.alts.internal.AltsFraming' | grep limit
...
        38: invokevirtual #59                 // Method java/nio/ByteBuffer.limit:(I)Ljava/nio/Buffer;

Still, after a bad protobuf release (protocolbuffers/protobuf#7827) and a near miss with Guava (google/guava#3994), I figured it was worth bringing this up again in case you want to take additional measures.

[cpovirk]

(Also, if you know of other projects who are likely to be affected (Google-run or otherwise), please spread the word.)

[ejona86]

I'm so very sorry for whatever your day must have been like. Those commands say it all. 🤣 Except then I looked at protocolbuffers/protobuf#7827 and there was more!

Yes, we build with Java 8. When I saw #6839 I wondered why animalsniffer didn't notice; but I also didn't understand our change (not enough context was given; thanks for this wonderful report). I'm not wild about the approach taken in #6839; I'd much rather downcast to Buffer manually at each limit location or use a utility that implicitly does the same, but without an automated checker there's likely regressions no matter the approach.

It seems like we should just give animalsniffer a bit more time to fix this. I've subscribed to the issue and PR so I can upgrade the version when released. We aren't at risk of swapping to Java 11 for compilation of our releases within the next ~year.

I'm strongly tempted to make changes now, but I know there's some PRs in flight that would immediately regress our state. So I'm just going to leave things "as-is" until we have a way to prevent regressions.

CC @voidzcy

[voidzcy]

I'd much rather downcast to Buffer manually at each limit location or use a utility that implicitly does the same

Yes, I also noticed the existing awkwardness (take in a ByteBuffer but saved as a Buffer). I was about to change it the other way (only cast to Buffer when necessary), but thought that would be an excessive change for my PR. I can quickly fix it now.

[voidzcy]

Existing usages are fixed in #7349. Created #7360 to fix animal sniffer (either we need some configurations or something is missing from animal sniffer side).