8220394: bufferedStream does not honor size limit

[Thomas Stüfe]

   Hi all,


Bug: https://bugs.openjdk.java.net/browse/JDK-8220394
cr:
http://cr.openjdk.java.net/~stuefe/webrevs/8220394-bufferedstream-does-not-honor-size-limit/webrev.00/webrev/

I redid the patch for this in a minimal form. The first version I posted in
March but after consideration I found it too invasive.

Short story: bufferedStream is misused as a
outputStream-with-dynamic-buffer in a number of places - misused since we
have stringStream for that.

It has something looking like a maximum buffer size cap but that is
actually a flush-trigger for child classes of bufferesStream.
bufferedStream::flush itself is a noop.

That means that printing to this stream may cause high memory footprint or
native OOM since the upper limit is not honored. In runaway printing coding
this can tear down the process.

This patch is a stopgap - when we reach the buffer limit - but not below
100M - we will assert(debug) or truncate (release).

I am careful here since I do not know if there are situations where more
than buffer-limit bytes are written today and suddenly enforcing this limit
now would cause errors. I think 100M is safe to be considered "too much".

The real correct solution should be that all callers use stringStream and
handle truncation. This patch is small and could be, if necessary, ported
down to older releases easily.

Thanks, Thomas