On Wed, 15 Apr 2026 15:45:26 GMT, Chen Liang <[email protected]> wrote:

>> Hi everyone,
>> 
>> `AccessFlags::print_on` is used for class, field and method printing, but 
>> those entities do not share the same modifier set. The current helper 
>> hard-codes a single mixed list of access flags for all 3 and is thus unaware 
>> of the type of the printed value. Instead of using a single shared printing 
>> function, we should move the printing to the relevant class, field, and 
>> method call sites. This makes the printer aware of the type it is printing 
>> and lets us check only the flags that are relevant for that type.
>> 
>> For this change, I remove `AccessFlags::print_on` and split the printing 
>> into 3 separate helpers:
>> - `InstanceKlass::print_class_flags`
>> - `Method::print_access_flags`
>> - `fieldDescriptor::print_access_flags`
>> 
>> As a part of that, I added the missing `AccessFlags` predicates used by each 
>> of the new printers, and updated each printer to check all the flags 
>> relevant for its type, as defined by `jvm_constants.h`. This lets us cover 
>> class-specific, method-specific, and field-specific modifiers that were not 
>> handled before.
>> 
>> I also added new gtests covering each of the 3 printing helpers and the 
>> previously missing flags.
>> 
>> Testing:
>> - Oracle tiers 1-3
>> - New gtests covering each of the printing helpers and each of the new flags
>> 
>> ---------
>> - [x] I confirm that I make this contribution in accordance with the 
>> [OpenJDK Interim AI Policy](https://openjdk.org/legal/ai).
>
> src/hotspot/share/runtime/fieldDescriptor.hpp line 114:
> 
>> 112:   void print_on(outputStream* st) const;
>> 113:   void print_on_for(outputStream* st, oop obj);
>> 114: #if !defined(PRODUCT) || INCLUDE_JVMTI
> 
> The new `!defined(PRODUCT)` seems to be only for the gtest. I wonder if gtest 
> can run the test for `print_access_flags` based on `INCLUDE_JVMTI` flag 
> instead; the macro condition tweak seems weird.

I am surprised that we only ever print access flags in relation to JVMTI - I 
would have expected logging or crash reporting to do so. In any case I also 
find the `!PRODUCT` a little jarring. Given we always have JVMTI it makes no 
different to our binaries.

> src/hotspot/share/utilities/accessFlags.hpp line 58:
> 
>> 56:   bool is_bridge      () const         { return (_flags & JVM_ACC_BRIDGE 
>>      ) != 0; }
>> 57:   bool is_transient   () const         { return (_flags & 
>> JVM_ACC_TRANSIENT   ) != 0; }
>> 58:   bool has_vararg     () const         { return (_flags & 
>> JVM_ACC_VARARGS     ) != 0; }
> 
> I recommend `is_varargs`, this flag is not enforced by the VM so technically 
> users can create class files that declare varargs methods without a trailing 
> array argument.

+1

> src/hotspot/share/utilities/accessFlags.hpp line 64:
> 
>> 62:   bool is_interface   () const         { return (_flags & 
>> JVM_ACC_INTERFACE   ) != 0; }
>> 63:   bool is_abstract    () const         { return (_flags & 
>> JVM_ACC_ABSTRACT    ) != 0; }
>> 64:   bool is_strict_method() const        { return (_flags & JVM_ACC_STRICT 
>>      ) != 0; }
> 
> If "strict method" sounds too weird, we can call this "strictfp" following 
> the original Java language modifier's name.

This will be short-lived either way as valhalla gets rid of the legacy strictfp 
notion for methods anyway.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/30746#discussion_r3090996397
PR Review Comment: https://git.openjdk.org/jdk/pull/30746#discussion_r3090998411
PR Review Comment: https://git.openjdk.org/jdk/pull/30746#discussion_r3091021640

Reply via email to