Peter, > On 24 Jun 2020, at 10:16, Peter Levart <peter.lev...@gmail.com> wrote: > > Hi, > > I wonder why "isRecord" was not encoded in class modifier bits, like "isEnum" > was for example. Are all 32 bits already taken? > These bits are precious and limited! ;-) The record-ness of a class is already implicit by the presence of the Record attribute, we don’t really need another way to represent it. I’m sure there was some discussion relating to this on amber-dev, but I don’t have it to hand.
> The isEnum() does not have the performance problem since getModifiers() > native method is intrinsified. > I don’t think we need anything too “fancy” here. I think there are quite straight forward ways to achieve the performance characteristics that we want, I’ve already outlined two possibilities (well.. the first is not really a serious possibility, just poking the bear!). If we encode these single state bits into a modifiers-like value, then we can treat them as such (without needing them to be “real” modifiers). And this should address any potential bloating concerns. -Chris. > Regards, Peter > > > > On 6/24/20 10:20 AM, Chris Hegarty wrote: >> >> >>> On 24 Jun 2020, at 07:03, David Holmes <david.hol...@oracle.com >>> <mailto:david.hol...@oracle.com>> wrote: >>> >>> Hi Chris, >>> >>> On 24/06/2020 2:30 am, Chris Hegarty wrote: >>>>> On 23 Jun 2020, at 14:49, Peter Levart <peter.lev...@gmail.com >>>>> <mailto:peter.lev...@gmail.com>> wrote: >>>>> >>>>> ... >>>>> >>>>> Ok, I'm going to push this to jdk15. >>>> Thank you Peter. This is a really nice change. >>>> As a follow on, and not for JDK 15, I observe that Class::isRecord0 / >>>> JVM_IsRecord shows up as consuming a significant amount of time, more than >>>> 10%, in some runs of the deserialization benchmark. The isRecord >>>> implementation is a native method in the JVM, so relatively expensive to >>>> call. >>>> This shows an opportunity to improve the Class::isRecord implementation >>>> with a simple cache of the record-ness of the j.l.Class, as is done >>>> selectively with other particular aspects of a class’s state. There are >>>> various ways to implement this, but here is just one [*]. >>> >>> There has been reluctance to add more and more fields to Class to cache all >>> these new attributes that are being added >> >> Yeah, that seems reasonable. The extra bloat should be given due >> consideration. >> >> I’ve not yet counted how many of these isThis and isThat methods that there >> are, but I suspect that there are enough that could warrant their state >> being encoded into a single int or long value on j.l.Class (that could be >> set lazily by the VM). This would setup a convenient and reasonably >> efficient location to add future pieces of cached state, like isSealed. >> >>> - but ultimately that is a call for core-libs folk to make. The general >>> expectation is/was that the need to ask a class if it is a Record (or >>> isSealed etc) would be rare. But (de)serialization is the exception for >>> isRecord() as unlike enums a simple instanceof test can't be used. >> >> It is relatively inexpensive to ask a non-record class if it is a record, >> but the converse is not the case. >> >> Java Serialization can probably “workaround” this, since it already has a >> level of local-class cache state, so we can leverage that [*], which is >> probably the right thing to do for Java Serialization anyway, but I still >> think that there is a general tractable problem here. >> >> -Chris. >> >> [*] >> diff -r 3a9521647349 >> src/java.base/share/classes/java/io/ObjectInputStream.java >> --- a/src/java.base/share/classes/java/io/ObjectInputStream.java Tue Jun >> 23 10:46:39 2020 +0100 >> +++ b/src/java.base/share/classes/java/io/ObjectInputStream.java Wed Jun >> 24 09:07:07 2020 +0100 >> @@ -2182,7 +2182,7 @@ >> handles.markException(passHandle, resolveEx); >> } >> >> - final boolean isRecord = cl != null && isRecord(cl); >> + final boolean isRecord = desc.isRecord(); >> if (isRecord) { >> assert obj == null; >> obj = readRecord(desc); >> >>
