Re: gij as JRE 5

[Andrew Haley]

Robert Schuster writes:
 > 
 > Andrew Haley wrote:
 > 
 > >Robert Schuster writes:
 > > > 
 > > > >There is actually a good reason. If the 1.5 classes are missing, the
 > > > >chances that a 1.5 compiled class will run are slim. This is what Andrew
 > > > >John Hughes was trying to address with his proposal to implement the 1.5
 > > > >classes as much as possible (using 1.4 sources). I haven't thought about
 > > > >it deeply, but on the surface that seems like it could be a good idea.
 > > > >
 > > > Actually I have doubts at this point. Javac 1.5/5 compiles all its 
 > > > "-target 1.5" stuff with the help of java.lang.StringBuilder
 > > > instead of java.lang.StringBuffer. This little difference makes all code 
 > > > that would normally run in a 1.4 environment unusable.
 > > > The funny thing is that StringBuilder (and attached interfaces 
 > > > Appendable, CharSequence) make no use of 1.5 features.
 > > > 
 > > > So this is my vote for adding support for 1.5 bytecode addition as well 
 > > > as 'source 1.5'-independent (helper) classes.
 > >
 > >Sounds good: it looks to me like we can do StringBuilder trivially by
 > >removing all the instances of "synchronized" from StringBuffer.  We
 > >already have methods like ensureCapacity_unsynchronized in the
 > >implementation.
 > >
 > Tom Tromey has submitted a StringBuilder implementation in classpath's 
 > generics branch.

My, he is busy.  ;-)

 > >The question does arise, however, about when we should start to
 > >generate calls to StringBuilder in gcj.  Do we have to wait for the
 > >rest of 1.5?  This is a hard one: no problem if we're generating
 > >bytecode for the locally installed gij, but otherwise...
 > >
 > Unfortunately I am not used to gcj but from a short glimpse to its 
 > man-page I can see that it does not support something comparable to 
 > javac's -source and -target
 > switch. How does gcj treat the assert() statement?

-fenable-assertions

 > Its an 1.4-introduced feature that has IMHO a similar nature like
 > the (subset of) 1.5-additions we are currently discussing. A
 > '-source/target'-like switch could help gcj to determine what kind
 > of source and binary level features are allowed.

It could do, but supporting such a matrix is something of a potential
minefield.  We've always used -f<feature>.

Andrew.