Hi,

A new FPC JVM snapshot has been put online at http://wiki.freepascal.org/FPC_JVM

The main new feature is full fledged support for Android/Dalvik, including a Pascal translation of the Android SDK classes. To compile for Android rather than for the Java VM, simply use the -Tandroid command line parameter. The compiler will still generate Java class files, but these can then be translated into Dalvik code by the Android SDK tools just like any javac-compiled class files.

Some other, minor, changes:
* the compiler now properly cleans up the intermediate .j files (Jasmin "Java assembler" files) when not using -a
* the JVM compiler and RTL (both for Java and Android) can now be built using the top-level FPC Makefile
* fixed modifying certain kinds of var/out parameters from inside nested routines
* fixed a bug triggered by some kind of nested try/except statements
* -Ctcompactintarrayinit command line option to create more compact code for initializing typed array constants. The reason is that the JVM class file format does not support array constants, so such initializations have to be encoded as explicit code sequences in the class init code. This class init code is however limited to 64KB of bytecode, and this can be exceeded in case you have many or large initialized arrays. This parameters uses alternate code that is potentially slower, but much more compact than the default
* fixed a bug in copy(unicodestring,x,y) when x0

Jonas

PS: I've put the Android RTL source files in rtl/android/jvm, so that there will be no conflicts when adding support for Android/ARM later.

This is amazing, thank you very much for your work!

Andrew

I'm trying to build it from the svn repo:
http://svn.freepascal.org/svn/fpc/branches/jvmbackend/ . I get the following error:

-------------------------
$ make CROSSOPT="-O2 -g" CPU_TARGET=jvm OS_TARGET=java all make: -iVSPTPSOTO: Command not found
make: -iSP: Command not found
make: -iTP: Command not found
make: -iSO: Command not found
make: -iTO: Command not found
make compiler_cycle RELEASE=1
make[1]: -iVSPTPSOTO: Command not found
make[1]: Entering directory `/home/ik/projects/foss/fpc/jvmbackend' make[1]: -iSP: Command not found
make[1]: -iTP: Command not found
make[1]: -iSO: Command not found
make[1]: -iTO: Command not found
make -C compiler cycle
make[2]: -iVSPTPSOTO: Command not found
make[2]: Entering directory `/home/ik/projects/foss/fpc/jvmbackend/compiler' make[2]: -iSP: Command not found
make[2]: -iTP: Command not found
make[2]: -iSO: Command not found
make[2]: -iTO: Command not found
make OS_TARGET= CPU_TARGET= CROSSBINDIR= BINUTILSPREFIX= CROSSCYCLEBOOTSTRAP=1 rtlclean rtl
make[3]: Entering directory `/home/ik/projects/foss/fpc/jvmbackend/compiler' make -C clean
make: Entering an unknown directory
make: *** clean: No such file or directory. Stop. make: Leaving an unknown directory
make[3]: *** [rtlclean] Error 2
make[3]: Leaving directory `/home/ik/projects/foss/fpc/jvmbackend/compiler' make[2]: *** [cycle] Error 2
make[2]: Leaving directory `/home/ik/projects/foss/fpc/jvmbackend/compiler' make[1]: *** [compiler_cycle] Error 2
make[1]: Leaving directory `/home/ik/projects/foss/fpc/jvmbackend' make: *** [base.build-stamp.jvm-java] Error 2

-------------------

Arch Linux, 64 bit. I have jdk7-openjdk and jre7-openjdk

Have anyone encountered something like that as well ?

Ido

I've seen these messages before when fpc or ppc386/ppcx64 is not in the $PATH

Regards,

Andrew

$ whereis ppcx64
ppcx64: /usr/bin/ppcx64
$ whereis fpc
fpc: /usr/bin/fpc /etc/fpc.cfg /usr/lib/fpc

it's v2.5.1 (2.6.0 rc)

Ido

_______________________________________________ fpc-devel maillist - http://lists.freepascal.org/mailman/listinfo/fpc-devel

It nevertheless means that the Makefile cannot find the "fpc" binary. You can try explicitly specifying the starting compiler by adding FPC=/usr/bin/ppcx64

Jonas

"Jonas Maebe" wrote

The main new feature is full fledged support for Android/Dalvik,

including a Pascal translation of the Android SDK classes. To compile for Android rather than for the Java VM, simply use the -Tandroid command line parameter. The compiler will still generate Java class files, but these can then be translated into Dalvik code by the Android SDK tools just like any javac-compiled class files.

It's seems that you have beat me here :)

I have continued to work on this as well three or four weeks ago, but I got stuck with a ClassDefNotFoundException when implementing a test application and tried to find out were the error was. I'll retry with your new snapshot (hopefully this weekend) and report back.

Regards,
Sven

Thanks, that works.

Ido

Sven Barth wrote

I have continued to work on this as well three or four weeks ago, but I got stuck with a ClassDefNotFoundException when implementing a test application and tried to find out were the error was.

Most FPC code will depend on the RTL, so make sure that you create a .jar file containing all FPC RTL class files (both those in rtl/units/jvm-android and in rtl/units/jvm-android/org/freepascal/rtl) and add it to your Android project in Eclipse (or pass it to dx if you are manually creating the .apk file).

Jonas

got

> stuck with a ClassDefNotFoundException when implementing a test

application

> and tried to find out were the error was. Most FPC code will depend on the RTL, so make sure that you create a .jar

file containing all FPC RTL class files (both those in rtl/units/jvm-android and in rtl/units/jvm-android/org/freepascal/rtl) and add it to your Android project in Eclipse (or pass it to dx if you are manually creating the .apk file).

It wasn't the RTL. It complained about a class I had just added in the same unit for which the corresponding class file was generated as well. As said I'll retest and report back.

Regards,
Sven

"whereis" searches in more places than PATH. Use "which" instead:

which ppcx64

Mattias

It provide the same answer. It seems that only the jvm branch have some problems with $PATH . but I was able to solve it :)

Thanks,

Mattias

Ido

_______________________________________________ fpc-devel maillist - http://lists.freepascal.org/mailman/listinfo/fpc-devel

So, I have now tested your new snapshot (nice that "make" finally works in the top level ^^) and I've got the same error as before. The error occurs when I try to instantiate a custom class of mine that implements an (inner) interface from the Android SDK. I have tested the following variations:

* implement a inner interface
* implement a "global" interface
* descend from TObject
* descend from JLObject
* implement in a nested class
* implement in a "global" class

All lead to the same error: ClassDefNotFoundException (and the class that's tried to be created) when instantiating the class.

I have attached an example containing some of the above mentioned variatons, a manifest file and a Makefile which you'll need to adjust to your setup. Then a "make prepare genrjava all" should (in theory) suffice to create a unsigned apk which you can sign by "make signedapk". This can then be installed using "make install" if you have an emulator (or device) running.

Regards,
Sven

Sven Barth wrote

So, I have now tested your new snapshot (nice that "make" finally works in the top level ^^) and I've got the same error as before. The error occurs when I try to instantiate a custom class of mine that implements an (inner) interface from the Android SDK.

Fixed!

I have attached an example containing some of the above mentioned variatons, a manifest file and a Makefile which you'll need to adjust to your setup. Then a "make prepare genrjava all" should (in theory) suffice to create a unsigned apk which you can sign by "make signedapk".

The res directory was missing from the archive, but I did it via an Eclipse project.

Jonas

Nice :D Thank you!

Btw: can it be that not all Makefiles have been regenerated? I issued a "make clean" with cpu and os set to jvm and android respectively and in $fpcdir/utils the process complained that jvm-android is not supported. I compiled fpcmake and regenerated all makefiles and now it works.

> I have attached an example containing some of the above mentioned variatons, a manifest file and a Makefile which you'll need to adjust to your setup. Then a "make prepare genrjava all" should (in theory) suffice to create a unsigned apk which you can sign by "make signedapk". The res directory was missing from the archive, but I did it via an Eclipse project.

Oops. There had to be something that I missed.

As I personally am not a big fan of Eclipse I have preferred to build my own small Makefile for that stuff ^^

Regards,
Sven

Sven Barth wrote

Btw: can it be that not all Makefiles have been regenerated? I issued a "make clean" with cpu and os set to jvm and android respectively and in $fpcdir/utils the process complained that jvm-android is not supported. I compiled fpcmake and regenerated all makefiles and now it works.

Only the makefiles relevant to the JVM target have been regenerated. If "make clean" with CPU_TARGET=jvm tries to clean the utils directory, that's an error in the main Makefile I have to fix.

As I personally am not a big fan of Eclipse I have preferred to build my own small Makefile for that stuff ^^

I found it easier to use Eclipse than to figure out all command line tools that I had to use :)

Jonas

Then consider it a bug, because the makefile tried exactly this :)

> As I personally am not a big fan of Eclipse I have preferred to build my own small Makefile for that stuff ^^ I found it easier to use Eclipse than to figure out all command line tools that I had to use :)

Yes, that was indeed a bit complicated, but this way I might be able to use the JVM compiler in Lazarus easily for the creation of Android application (though Lazarus first needs to learn about the "unicodestrings" modeswitch, and maybe external classes as well -.- )

Regards,
Sven

I'll look into it.

Yes, that was indeed a bit complicated, but this way I might be able to use the JVM compiler in Lazarus easily for the creation of Android application (though Lazarus first needs to learn about the "unicodestrings" modeswitch, and maybe external classes as well -.- )

And final fields.

Jonas

18.12.2011 16:45, Jonas Maebe пишет:

Please create bug reports for these missing Lazarus features.

On 18 Dec 2011, at 13:22, Sven Barth wrote: [...] > Yes, that was indeed a bit complicated, but this way I might be able to use the JVM compiler in Lazarus easily for the creation of Android application (though Lazarus first needs to learn about the "unicodestrings" modeswitch,

What's that?

[...]

Mattias

20.12.2011 7:00, Mattias Gaertner пишет:

A modeswitch introduced in the JVM branch to turn the string type to unicodestring (similar to {$H+} turns string to ansistring).

Best regards,
Paul Ishenin

Jonas Maebe wrote

And final fields.

Final methods can't be overriden in the descendants. But what are final fields for?

Best regards,
Paul Ishenin

Paul Ishenin wrote

Final methods can't be overriden in the descendants. But what are final fields for?

I think they are the way to write a constant in Java. Because they have no procedural elements, they need to work around with class elements to do everything.

20.12.2011 10:47, Felipe Monteiro de Carvalho пишет:

FPC classes support internal constants so I don't believe Jonas would add another syntax element for already existent feature.

Best regards,
Paul Ishenin

AFAIK final fields can only be assigned once in the constructor (or in the declaration of which the code is made part of the class constructor (also AFAIK ^^)).

Regards,
Sven

20.12.2011 4:00, Paul Ishenin пишет:

It`s something like writeable constant, but writeable only in constructor.

Can someone explain the pascal syntax?

Mattias

According to example classes (JAImage, JAFGlyphJustificationInfo) in http://svn.freepascal.org/cgi-bin/viewvc.cgi/branches/jvmbackend/rtl/java/jdk15.inc?revision=19030&view=markup it is like the following:

TFoo = class
   final var
     fField1: Integer;
     fField2: Integer;
   final class var
     fClassField1: Integer;
     fClassField2: Integer;
end;

Please note: Every field (final or not) can be followed by a "external name 'FieldName'" clause if the class is "external" as well (you can see this in the mentioned example classes)

You also might want to read this:
http://wiki.freepascal.org/FPC_JVM/Language#New_language_features

Regards,
Sven

Sven Barth wrote

You also might want to read this: http://wiki.freepascal.org/FPC_JVM/Language#New_language_features

Looking over the Lazarus commit log, I noticed a small error on that page: the directive for pointer arithmetic is called {$pointermath on/ off}, not {$modeswitch cs_pointermath} (corrected in the mean time).

That switch is not specific to the JVM branch though: http://wiki.freepascal.org/User_Changes_2.6.0#The_POINTERARITHMETICS_mode_switch_has_been_replaced_with_the_POINTERMATH_directive

Jonas

Thanks for the hint.
I changed that in Lazarus.

Maybe the JVM wiki should be updated too.

Mattias

It has been updated (see the "corrected in the mean time" above). You may have to force-reload the page to see the change.

Jonas