This commit is contained in:
Chris Hegarty 2014-08-18 10:59:36 +01:00
commit 7d146ee898
127 changed files with 3403 additions and 4038 deletions

View file

@ -196,19 +196,19 @@ class StringCoding {
static char[] decode(Charset cs, byte[] ba, int off, int len) {
// (1)We never cache the "external" cs, the only benefit of creating
// an additional StringDe/Encoder object to wrap it is to share the
// de/encode() method. These SD/E objects are short-lifed, the young-gen
// gc should be able to take care of them well. But the best approash
// de/encode() method. These SD/E objects are short-lived, the young-gen
// gc should be able to take care of them well. But the best approach
// is still not to generate them if not really necessary.
// (2)The defensive copy of the input byte/char[] has a big performance
// impact, as well as the outgoing result byte/char[]. Need to do the
// optimization check of (sm==null && classLoader0==null) for both.
// (3)getClass().getClassLoader0() is expensive
// (4)There might be a timing gap in isTrusted setting. getClassLoader0()
// is only chcked (and then isTrusted gets set) when (SM==null). It is
// is only checked (and then isTrusted gets set) when (SM==null). It is
// possible that the SM==null for now but then SM is NOT null later
// when safeTrim() is invoked...the "safe" way to do is to redundant
// check (... && (isTrusted || SM == null || getClassLoader0())) in trim
// but it then can be argued that the SM is null when the opertaion
// but it then can be argued that the SM is null when the operation
// is started...
CharsetDecoder cd = cs.newDecoder();
int en = scale(len, cd.maxCharsPerByte());