Hi, Roger.
The change looks good. I just noticed a couple small things:
2324 for (int i = 0; i < slots.length-1; i++) {
2325 new FieldValues(slots[i].desc, true);
The slots[i].hasData check is no longer performed here.
2561 primValues = new byte[desc.getPrimDataSize()];
2562 objValues = new Object[desc.getNumObjFields()];
In defaultReadFields(), the arrays would remain null in the case of a 0
from getPrimDataSize() or getNumObjFields(). In the new code, 0-length
arrays are created.
Thanks,
-Brent
On 11/18/20 9:39 AM, Roger Riggs wrote:
ObjectInputStream has nearly identical but separate implementations to read
values from the stream.
Both implementations read primitive and object values from the stream and
return an object holding the values.
OIS.readFields() uses the internal class GetFieldImpl while
OIS.defaultReadObject and reading records uses the internal class FieldValues.
The behavioral difference between the two is whether dependencies are tracked
in the object handle table or not.
The classes are merged, retaining the internal FieldValues name and the
behavior to track dependencies or not.
The constructor is passed the class descriptor and flag to track or not; it
reads and saves the values from the stream.
The callers are updated to call the merge FieldValues methods.
There is no change in behavior; all current tests pass.
-------------
Commit messages:
- 8256480: Refactor ObjectInputStream field reader implementation
Changes: https://git.openjdk.java.net/jdk/pull/1296/files
Webrev: https://webrevs.openjdk.java.net/?repo=jdk&pr=1296&range=00
Issue: https://bugs.openjdk.java.net/browse/JDK-8256480
Stats: 157 lines in 1 file changed: 26 ins; 78 del; 53 mod
Patch: https://git.openjdk.java.net/jdk/pull/1296.diff
Fetch: git fetch https://git.openjdk.java.net/jdk pull/1296/head:pull/1296
PR: https://git.openjdk.java.net/jdk/pull/1296
