[External] : Re: asmtools `ldc Dynamic` syntax
Leonid Kuskov
leonid.kuskov at oracle.com
Fri Apr 23 04:29:41 UTC 2021
Agreed, processing of the reference_kind items is incorrect, excepting REF_invokeStatic, REF_invokeSpecial. (https://bugs.openjdk.java.net/browse/CODETOOLS-7902892)
I've fixed this and the fix will be applied in the coming days. I need time to test it.
The format will be the same for all Dynamic instructions:
LDC(2_W) Dynamic reference_kind:{Method|InterfaceMethod} {ClassName|InterfaceName}.MethodName....
Since asmtools is written for testing purposes it doesn't control assertions like as follows:
"If the value is 8 (REF_newInvokeSpecial), the name of the method represented by a CONSTANT_Methodref_info structure must be <init>."
It's valid to write ldc Dynamic REF_invokeVirtual:InterfaceMethod ClassName:"<clinit>"
On 4/22/21, 4:14 AM, "Maxim Degtyarev" <mdegtyarev at gmail.com> wrote:
> If the value of the reference_kind item is 6 (REF_invokeStatic) or 7 (REF_invokeSpecial),
The question is about other part:
If the value is 8 (REF_newInvokeSpecial), the name of the method represented
by a CONSTANT_Methodref_info structure must be <init>.
Either assembler or disassembler doing wrong:
- Disassembler produce output with the "Method " prefix for
constructor reference:
ldc Dynamic
REF_newInvokeSpecial
:Method LdcConDyTwice."<init>"
...
- Assembler in turn accept only syntax WITHOUT "Method " prefix:
ldc Dynamic
REF_newInvokeSpecial
:LdcConDyTwice."<init>"
...
The question is: Which syntax is correct for REF_newInvokeSpecial?
> > (it's even didn't assemble back).
> I could not reproduce this. Please send your example.
Here is an example:
https://urldefense.com/v3/__https://gist.github.com/Maccimo/9763417764aa53576c39b73d14a8fc85__;!!GqivPVa7Brio!MRmThSYowTE7Lm_SK_0zpruGdkigwFBH3hpCzEWqVAuEByXZnd8D-8K-WSjU1Yy68uc$
Assembled with asmtools built from master branch of
https://urldefense.com/v3/__https://github.com/openjdk/asmtools__;!!GqivPVa7Brio!MRmThSYowTE7Lm_SK_0zpruGdkigwFBH3hpCzEWqVAuEByXZnd8D-8K-WSjUI4LCXOU$
BTW, https://wiki.openjdk.java.net/display/CodeTools/asmtools page
contains broken links for downloads and there are no "Releases" on
GitHub for asmtools.
In case you need a bit less synthetic example here is ENUM class
example implemented via ConDy:
https://urldefense.com/v3/__https://github.com/Maccimo/HugeEnumGeneratorArticle/blob/master/5_Samples/bin/ConDyHugeEnum.class__;!!GqivPVa7Brio!MRmThSYowTE7Lm_SK_0zpruGdkigwFBH3hpCzEWqVAuEByXZnd8D-8K-WSjUqB8d0tg$
Everything required to generate such enums can be found in the same repository.
> > 2nd issue is the defect, although it doesn't affect a code the bytecode footprint grows.
Not only bytecode footprint. Different Dynamic CP entries will
reference different instances of objects.
El jue, 22 abr 2021 a las 8:01, Leonid Kuskov
(<leonid.kuskov at oracle.com>) escribió:
>
> Hello,
>
> Your first question is about the JVMS 4.4.8. The CONSTANT_MethodHandle_info Structure
> CONSTANT_MethodHandle_info {
> u1 tag;
> u1 reference_kind;
> u2 reference_index;
> }
> The format of the ldc instruction is: ldc Dynamic reference_kind:reference_index {ClassName|InterfaceName}.MethodName ....
> According to the assertion:
>
> If the value of the reference_kind item is 6 (REF_invokeStatic) or 7 (REF_invokeSpecial), then if the class file version number is less than 52.0, the constant_pool entry at that index must be a CONSTANT_Methodref_info structure representing a class's method for which a method handle is to be created; if the class file version number is 52.0 or above, the constant_pool entry at that index must be either a CONSTANT_Methodref_info structure or a CONSTANT_InterfaceMethodref_info structure (§4.4.2) representing a class's or interface's method for which a method handle is to be created.
>
> i.e. we might have the following pairs:
>
> ldc Dynamic REF_invokeSpecial:Method ...... if reference_index is an index of a CONSTANT_Methodref_info structure in CP or
> ldc Dynamic REF_invokeSpecial: InterfaceMethod... if reference_index points to CONSTANT_InterfaceMethodref_info
>
> 2nd issue is the defect, although it doesn't affect a code the bytecode footprint grows.
> I will take care of the problem: https://bugs.openjdk.java.net/browse/JDK-8129547
>
> > (it's even didn't assemble back).
> I could not reproduce this. Please send your example.
>
>
>
> On 4/21/21, 3:54 PM, "code-tools-dev on behalf of Maxim Degtyarev" <code-tools-dev-retn at openjdk.java.net on behalf of mdegtyarev at gmail.com> wrote:
>
> Hello,
>
> I have 2 questions regarding `ldc Dynamic` syntax:
>
> 1) Consider the following example:
>
> ldc Dynamic
> REF_newInvokeSpecial
> :LdcConDyTwice."<init>"
> :"(Ljava/lang/invoke/MethodHandles$Lookup;Ljava/lang/String;Ljava/lang/Class;)V"
> :LdcConDyTwice
> :"LLdcConDyTwice;";
>
> Being assembled with `asmtools jasm` and then disassembled with
> `asmtools jdis` above instruction turns to:
>
> ldc Dynamic
> REF_newInvokeSpecial
> :Method LdcConDyTwice."<init>"
> :"(Ljava/lang/invoke/MethodHandles$Lookup;Ljava/lang/String;Ljava/lang/Class;)V"
> :LdcConDyTwice
> :"LLdcConDyTwice;";
>
> I assume "Method" in front of the constructor reference is a
> disassembler bug (it's even didn't assemble back).
> Could anyone confirm my assumption?
>
> 2) Is there a way to reference the same Dynamic constant more than
> once with the current jasm syntax?
> Right now two syntactically equal ldc Dynamic instructions create two
> different entries in BootstrapMethods attribute.
>
More information about the code-tools-dev
mailing list