On Tue, 10 Mar 2026 11:59:06 GMT, Paul Hübner <[email protected]> wrote:
>> Hi all,
>>
>> This adds some more debug prints to various internals.
>>
>> When printing an `InstanceKlass`, the kind is now shown. For example:
>>
>> InstanceKlass (kind=1): java.lang.Integer {0x00001ff00022c318}
>>
>>
>> The `InlineKlass::Members` are also printed. For example (note that to get
>> individual array entries `WizardMode` must be enabled):
>>
>> - ---- inline type members:
>> - extended signature registers: Array<T>(0x00000001330002f8)
>> 0 : SigEntry: type=18 offset=0 null_marker=0 Symbol: 'java/lang/Integer'
>> count 65535
>> 1 : SigEntry: type=10 offset=16 null_marker=0 Symbol: 'value' count 65535
>> 2 : SigEntry: type=14 offset=24 null_marker=0 Symbol: 'java/lang/Integer'
>> count 65535
>> - return registers: Array<T>(0x0000000133000348)
>> 0 : (c_rarg0,c_rarg0)
>> 1 : (c_rarg7,BAD!)
>> - pack handler: 0x0000000119b21238
>> - pack handler (jobject): 0x0000000119b21140
>> - unpack handler: 0x0000000119b21244
>> - null reset offset: 128
>> - payload offset: 16
>> - payload size (bytes): 8
>> - payload alignment: 8
>> - null-free non-atomic size (bytes): 4
>> - null-free non-atomic alignment: 4
>> - null-free atomic size (bytes): 4
>> - nullable atomic size (bytes): 8
>> - nullable non-atomic size (bytes): 5
>> - null marker offset: 20
>>
>>
>> Fields now indicate their field flags. Non-static fields indicate (a) when
>> they are flat and (b) their layout when flat. For example:
>>
>> - ---- non-static fields (1 words):
>> - private final value flat 'inner' (fields 0x00000006) 'LAValue;' @12
>> LayoutKind: NULLABLE_NON_ATOMIC_FLAT
>>
>>
>> Arrays already have plenty of flattening information that can be printed.
>>
>> Testing: tiers 1-3.
>
> Paul Hübner has updated the pull request incrementally with one additional
> commit since the last revision:
>
> Remove commented out line.
This patch is showing a bug in the way the AccessFlags are printed (the bug is
not in this patch, but this patch is using the buggy code):
` - ---- static fields (6 words):
- private static final value 'FS' (fields 0x00000000) 'Ljava/io/FileSystem;'
@120
- public static final value 'separatorChar' (fields 0x00000000) 'C' @160
- public static final value 'separator' (fields 0x00000000)
'Ljava/lang/String;' @124
- public static final value 'pathSeparatorChar' (fields 0x00000000) 'C' @162
- public static final value 'pathSeparator' (fields 0x00000000)
'Ljava/lang/String;' @128
- private static final value 'UNSAFE' (fields 0x00000000)
'Ljdk/internal/misc/Unsafe;' @132
- private static final value 'PATH_OFFSET' (fields 0x00000000) 'J' @136
- private static final value 'PREFIX_LENGTH_OFFSET' (fields 0x00000000) 'J'
@144
- private static final value 'serialVersionUID' (fields 0x00000008) 'J' @152
- static final synthetic value '$assertionsDisabled' (fields 0x00000000) 'Z'
@164
- ---- non-static fields (4 words):
- private final transient value 'prefixLength' (fields 0x00000000) 'I' @12
- private final value 'path' (fields 0x00000000) 'Ljava/lang/String;' @16
- private transient value 'status' (fields 0x00000000)
'Ljava/io/File$PathStatus;' @20
- private volatile transient value 'filePath' (fields 0x00000000)
'Ljava/nio/file/Path;' @24 `
All fields are marked as 'value' when obviously, none of them are storing
values from value classes!
The culprit is in `AccessFlags::print_on(outputStream* st)`:
`if (Arguments::is_valhalla_enabled()) {
if (is_identity_class()) st->print("identity ");
if (!is_identity_class()) st->print("value " );
}`
This piece of code was added considering class AccessFlags and not field
AccessFlags. ACC_IDENTITY has value 0x20, and no field flag is using this
value, leading the code above to always print "value ".
-------------
PR Comment: https://git.openjdk.org/valhalla/pull/2208#issuecomment-4057626976
