[fpc-devel] Re: Android SDK's DX reports that a local is invalid

[Sven Barth]

On 30.06.2012 16:43, Sven Barth wrote:
> On 30.06.2012 16:34, Sven Barth wrote:
>> Hello together and especially hello Jonas,
>>
>> I'm currently playing around with Android again. Mostly to implement a
>> backup/restore procedure for my TrainLog application as I want to update
>> my device (official ICS update, yippieh!) and don't plan to lose all my
>> collected train counts...
>>
>> Nevertheless I added a routine to my application, but when I try to "dx"
>> the generated class file I get an error about an invalid locale.
>>
>> Here is the offending method (extracted, but error is reproduceable)
>> [it's based on Java code mentioned here:
>> http://stackoverflow.com/a/6542214 ]:
>>
>> === unit begin ===
>>
>> unit Test;
>>
>> {$mode objfpc}{$H+}
>> {$modeswitch unicodestrings}
>>
>> interface
>>
>> uses
>>    AndroidR14;
>>
>> type
>>    TTest = class
>>    private
>>      procedure CopyFile(aSrc, aDest: JIFile);
>>    end;
>>
>> implementation
>>
>> procedure TTest.CopyFile(aSrc, aDest: JIFile);
>> var
>>    outstr: JIFileOutputStream;
>>    instr: JIFileInputStream;
>>    outchannel, inchannel: JNCFileChannel;
>> begin
>>    outstr := JIFileOutputStream.Create(aDest);
>>    instr := JIFileInputStream.Create(aSrc);
>>
>>    try
>>      inchannel := instr.getChannel;
>>      outchannel := outstr.getChannel;
>>      inchannel.transferTo(0, inchannel.size, outchannel as
>> JNCWritableByteChannel);
>>    finally
>>      try
>>        if Assigned(inchannel) then
>>          inchannel.close;
>>      finally
>>        if Assigned(outchannel) then
>>          outchannel.close;
>>      end;
>>    end;
>> end;
>>
>> end.
>>
>> === unit end ===
>>
>> The output of "dx" when called with "--dex --output=classes.dex lib/"
>> (where lib is the directory I set for FPC's "-FE" option) is the
>> following:
>>
>> === output begin ===
>>
>> EXCEPTION FROM SIMULATION:
>> local 0006: invalid
>>
>> ...at bytecode offset 00000092
>> locals[0000]: LTTest;
>> locals[0001]: Ljava/io/File;
>> locals[0002]: Ljava/io/File;
>> locals[0003]: Ljava/io/FileOutputStream;
>> locals[0004]: Ljava/io/FileInputStream;
>> locals[0005]: <invalid>
>> locals[0006]: <invalid>
>> locals[0007]: Ljava/lang/Object;
>> locals[0008]: <invalid>
>> locals[0009]: <invalid>
>> locals[000a]: <invalid>
>> locals[000b]: <invalid>
>> locals[000c]: type{java.lang.Object}
>> locals[000d]: <invalid>
>> locals[000e]: <invalid>
>> locals[000f]: <invalid>
>> ...while working on block 0092
>> ...while working on method CopyFile:(Ljava/io/File;Ljava/io/File;)V
>> ...while processing CopyFile (Ljava/io/File;Ljava/io/File;)V
>> ...while processing TTest.class
>>
>> 1 error; aborting
>>
>> === output end ===
>>
>> I'm using revision 21737 (pretty up to date ^^)
>
> Update: It seems that dx dislikes the nested "try ... finally ... end"
> block inside the outer's block "finally ... end".

Correction: even without the nested try-finally, but with within it 
moved to the outer finally-end the error still occurs. So it seems that 
there is some problem with the code generated for the finally block...

Regards,
Sven