You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@commons.apache.org by "Benoit Tellier (Jira)" <ji...@apache.org> on 2023/01/30 02:06:00 UTC

[jira] [Created] (IO-786) Unsynchronized BufferedInputStream

Benoit Tellier created IO-786:
---------------------------------

             Summary: Unsynchronized BufferedInputStream
                 Key: IO-786
                 URL: https://issues.apache.org/jira/browse/IO-786
             Project: Commons IO
          Issue Type: Improvement
          Components: Streams/Writers
         Environment: JRE 19, linux
            Reporter: Benoit Tellier


As part of development of Apache James I had the unpleasant surprise to notice that, on modern JVMs, the cost of synchronization skyrocketed (JRE 19).

In one part of the code, we do need to find the exact location of email header end, and for this needs to read an InputStream byte by byte. We of course buffer the inputStream in order to limit potential blocking calls. Profiling showed 73% of the reading time is spent synchronizing on the BufferedInputStream. Thus I am keen on having an UnsynchronizedBufferedInputStream & friends at hand. See attached screenshot.

 => This is reported upstream see https://bugs.openjdk.org/browse/JDK-4097272 . This was disregarded in lower Java version but I hope it "could" be reconsidered.

 => While I can duplicate the class in Apache James source code and remove synchronised keywords, this sounds a generic enough to fit in commons-io.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)