Mark,
On 4/17/23 05:07, Mark Thomas wrote:
On 16/04/2023 18:59, Mark Thomas wrote:
On 14/04/2023 21:38, Christopher Schultz wrote:
The proposed 8.5.88 release is:
[ ] Broken - do not release
[X] Stable - go ahead and release as 8.5.88 (stable)
Build is reproducible on Windows x64 but not Linux x64.
Unit tests pass on Linux x64, Windows x64, MacOS M1 and MacOS x64.
It appears the compiler constructs a slightly different class for
NioEndpoint$Poller when compiling on Linux x64. I started to dig into
why but didn't reach a conclusion. Investigating further is on my TODO
list.
I have made some progress. The difference is in how the following line
is handled (there are other differences but this is the first and the
other differences may be as a result of this one).
https://github.com/apache/tomcat/blob/8.5.x/java/org/apache/tomcat/util/net/NioEndpoint.java#L846
When compiling on Windows with Java 11 targeting Java 7 the following
byte code is generated:
...
126: aload %4
128: invokevirtual
org.apache.tomcat.util.net.NioChannel.getIOChannel
()Ljava/nio/channels/SocketChannel; (32)
131: checkcast <java.nio.channels.WritableByteChannel> (121)
134: astore %6
...
When compiling on Linux with Java 11 targeting Java 7 the checkcast
command is not present. Neither is it present when compiling on Windows
with Java 11 targeting Java 8.
This is arguably a bug in the compiler - the byte code should be the
same across platforms - but having researched things this far, I don't
plan to pursue this further.
+1 to ignoring this, though it is annoying that the build is therefore
not completely reproducible.
Do you have an account with the Eclipse bug-tracker used for JDK bugs? I
don't think I do. Is this something you can report? It should be
trivially reproducible (hah!) with publicly-available artifacts.
Thanks,
-chris
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
