On 18/05/2022 10:25, Mark Thomas wrote:
Hi all,
So of the three issues Martin identified, I've fixed 1 & 3 which were
the simple ones.
For 1, the import handler needed updating to add the new classes added
to java.lang in Java 19.
For 3, the memory leak protection code that stops executor threads
needed to be updated for some changes to Thread internals.
The tricky one is 2.
Prior to Java 19, the character -> byte encoders used an 8k buffer for
character data. This aligned with Tomcat's internals. This meant that
the new HEAD handling in the Servlet API had the same buffering
behaviour as Tomact's internals so GET and HEAD behaved the same way.
In Java 19, the character -> byte encoders start with a 512 byte buffer
that may grow up to 8192 bytes. This means the buffering for HEAD
handling is different to the buffering for GET handling which in turn
means GET and HEAD see different headers (even accounting for allowed
differences).
The issue is when responses are committed (and headers are fixed). The
HEAD tests examine the behaviour at the buffer boundaries including
reset behaviour. GET and HEAD effectively having different buffer sizes
triggers different behaviour for some of these resets which is why some
fail.
I'm currently thinking about ways to tackle this but I don't yet have a
solution I'm completely happy with. Suggestions welcome.
I found a solution. Since the issue was only with the legacy handling
mode for HEAD requests, adjusting the response buffer size to compensate
for the smaller char -> byte buffer addresses the issue.
Mark
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
