RFR [S] 8077242: (str) Optimize AbstractStringBuilder.append(CharSequence, int, int) for String argument
Hello! AbstractStringBuilder, which is base for StringBuilder and StringBuffer has a method for appending a sub-sequence of a CharSequence. Internally, it copies one char at a time in a loop, picking them up with CharSequence.charAt() method. For the case when the argument is a String, it can be done more efficiently, by a call to String.getChars(). Since String is a final class, it should be safe to pass 'value' to its method. I've also found a few places in JDK where code can be done more efficient, using this optimization. Would you please help review this fix? BUGURL: https://bugs.openjdk.java.net/browse/JDK-8077242 WEBREV: http://cr.openjdk.java.net/~igerasim/8077242/02/webrev/ Sincerely yours, Ivan
This looks good. Because we already have append(String) it *may* be a good idea to add append(String, int, int). See also my failure to add getChars to CharSequence itself. If we (you?) could make that happen, that would eliminate the need for instanceof String etc. On Sat, Jun 27, 2015 at 7:23 AM, Ivan Gerasimov <ivan.gerasimov@oracle.com> wrote:
Hello!
AbstractStringBuilder, which is base for StringBuilder and StringBuffer has a method for appending a sub-sequence of a CharSequence. Internally, it copies one char at a time in a loop, picking them up with CharSequence.charAt() method. For the case when the argument is a String, it can be done more efficiently, by a call to String.getChars(). Since String is a final class, it should be safe to pass 'value' to its method.
I've also found a few places in JDK where code can be done more efficient, using this optimization.
Would you please help review this fix?
BUGURL: https://bugs.openjdk.java.net/browse/JDK-8077242 WEBREV: http://cr.openjdk.java.net/~igerasim/8077242/02/webrev/
Sincerely yours, Ivan
Thank you Martin for review! On 01.07.2015 0:59, Martin Buchholz wrote:
This looks good.
Because we already have append(String) it *may* be a good idea to add append(String, int, int).
I found only a dozen of the places in JDK where this method is applicable, so I thought it doesn't seem to worth a change in the API.
See also my failure to add getChars to CharSequence itself. If we (you?) could make that happen, that would eliminate the need for instanceof String etc.
Yes, it was my initial thought too: http://mail.openjdk.java.net/pipermail/core-libs-dev/2015-May/033285.html I guess, it's not a problem to add default CharSequence.getChars(). The problem is to call it safely. If there were a way to enforce all the derived classes except a few trusted stick to the default implementation, it would be possible to use getChars in sb.append(). For now, I limited the scope of the change, so that changing sb.append(str.substring(a, b)) to sb.append(str, a, b) will be at least no slower. Sincerely yours, Ivan
On Sat, Jun 27, 2015 at 7:23 AM, Ivan Gerasimov <ivan.gerasimov@oracle.com <mailto:ivan.gerasimov@oracle.com>> wrote:
Hello!
AbstractStringBuilder, which is base for StringBuilder and StringBuffer has a method for appending a sub-sequence of a CharSequence. Internally, it copies one char at a time in a loop, picking them up with CharSequence.charAt() method. For the case when the argument is a String, it can be done more efficiently, by a call to String.getChars(). Since String is a final class, it should be safe to pass 'value' to its method.
I've also found a few places in JDK where code can be done more efficient, using this optimization.
Would you please help review this fix?
BUGURL: https://bugs.openjdk.java.net/browse/JDK-8077242 WEBREV: http://cr.openjdk.java.net/~igerasim/8077242/02/webrev/ <http://cr.openjdk.java.net/%7Eigerasim/8077242/02/webrev/>
Sincerely yours, Ivan
participants (2)
-
Ivan Gerasimov
-
Martin Buchholz