asmtools bug: ldc constant dynamic that returns a long or a double should use ldc2_w instead of ldc

[Leonid Kuskov]

Hi Remi,

Many thanks for reporting the issue. I consider that the asmtools 
behavior is correct.
The main goal of asmtools is the ability to generate a broken binary 
code. asmtools just converts stuff that it processes.
It will be impossible to develop negative test cases if the tool 
controls a return type and overwrites a corresponding instruction.
For example, I need to check a negative test cases: how VM will process 
pairs ldc2_w:bsm(...)I,  ldc_w:bsm(...)D?
Here correction of instructions will hide an expected VM reaction.

Thanks,
Leonid

On 9/2/18 07:40, Remi Forax wrote:
> Hi all, hi Leonid,
> while fixing a bug in ASM [1] when a ldc to a condy that returns a long uses the wrong bytecode,
> i've discovered that autotools has the same issue.
>
> The following code should generate a ldc2_w and not a ldc.
>
> public static Method primitiveExample:"()J"
>    stack 2 locals 0
> {
>      ldc  Dynamic REF_invokeStatic
>          :jdk11/AllInstructions.bsm
>          :"(Ljava/lang/invoke/MethodHandles$Lookup;Ljava/lang/String;Ljava/lang/Class;)J"
>          :_
>          :"J" {};
>      lreturn;
> }
>
> in case you wonder why we use asmtools, for testing ASM we try to not use bytecodes generated by ASM as source :)
>
> regards,
> Rémi
>
> [1] https://gitlab.ow2.org/asm/asm/issues/317850