Request for review: Race conditions in java.nio.charset.Charset (original) (raw)

Ulf Zibis Ulf.Zibis at gmx.de
Wed Oct 7 22:58:41 UTC 2009

Martin, thank for your review.

Am 08.10.2009 00:01, Martin Buchholz schrieb:

IIRC correctly, I am the author of the hacky 2-element charset cache. Certainly improvements can be made, but it's hard to balance memory usage vs. the cost and complexity of writing a good cache.

I guess, the memory usage of 2-element class CharsetItem is equivalent to 2-element Object[].

I agree with Sherman that a race in the cache itself is not a bug (or at best, a performance bug).

For my better understanding: Can you explain me the real bug in http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6881442. In my understanding, loading the "name" field twice is too only a performance bug. Please correct me!

I don't think it's worth a point fix here unless an actual wrong result can be demonstrated. I do think a more sophisticated charset cache would be good, but hard to get right.

The other point is the size of the cache, see http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6795535. I have logged the usage of the Charset.lookup() method from a simple test which has only called ISO-8859-1 and IBM037 . As you can see, UTF-8 and cp1252 (default encoding on German Windows) is frequently requested from the VM, so IMO size 2 is too restrictive (note the different aliases UTF-8, utf-8 and UTF8): UTF-8 utf-8 UTF-8 Cp1252 UTF-8 UTF-8 UTF-8 UTF-8 UTF-8 UTF-8 UTF8 UTF8 Cp1252 Cp1252 Cp1252 Cp1252 Cp1252 Cp1252 Cp1252 Cp1252 Cp1252 Cp1252 Cp1252 Cp1252 UTF-8 IBM037 UTF-8 UTF-8 utf-8 ISO-8859-1 UTF-8

-Ulf

More information about the core-libs-dev mailing list