RFR [9]8134424: BlockDataInputStream.readUTFBody: examine sizing local StringBuffer with the given length

[Chris Hegarty]

Ha, such a seemingly innocuous little perf change has generated much noise ;-)

You are of course right.

On 11 Feb 2016, at 15:23, Roger Riggs <Roger.Riggs at oracle.com> wrote:

> Hi Chris,
> 
> The change to preallocate the StringBuffer is fine.
> 
> But I have misgivings about not consuming the data from the stream and returning the empty string
> for out of range lengths. It will silently leave the input stream in an unusable state.
> 
> I think it should throw an exception (StreamCorruptedException would be appropriate)
> if the size is negative or too large.
> 
> BTW, a UTF string can be up to three times the length of a Java string due to the encoding.
> So utflen should be allowed to be Integer.MAX_VALUE * 3 (in the absence of some other implementation limitation).

How about this, preserving existing behaviour for non-common cases:

diff --git a/src/java.base/share/classes/java/io/ObjectInputStream.java b/src/java.base/share/classes/java/io/ObjectInputStream.java
--- a/src/java.base/share/classes/java/io/ObjectInputStream.java
+++ b/src/java.base/share/classes/java/io/ObjectInputStream.java
@@ -3144,7 +3144,14 @@
          * utflen bytes.
          */
         private String readUTFBody(long utflen) throws IOException {
-            StringBuilder sbuf = new StringBuilder();
+            StringBuilder sbuf ;
+            if (utflen > 0 && utflen < Integer.MAX_VALUE) {
+                // a reasonable initial capacity based on the UTF length
+                int initialCapacity = Math.min((int)utflen, 16384);
+                sbuf = new StringBuilder(initialCapacity);
+            } else {
+                sbuf = new StringBuilder();
+            }
             if (!blkmode) {
                 end = pos = 0;
             }


-Chris.