RFR: 8314250: CDS dump error message: Invoker type parameter must start and end with Object: L3I_L

[Calvin Cheung]

On Wed, 6 Mar 2024 22:00:42 GMT, Ioi Lam <iklam at openjdk.org> wrote:

>> To avoid the CDS dump error message, a fix is during dumping a classlist, check if an invoker can be archived. 
>> If not, don't write the invoker info into the classlist, i.e. don't call `logLambdaFormInvoker()`. While generating holder classes (in `generateHolderClasses()`), don't add the `MethodType` to the `invokerTypes` if will fail the check in the `build()` method which would result in a `RuntimeException`.
>> 
>> Also updated the `MethodHandlesInvokersTest.java` under `appcds/methodHandles` and `appcds/dynamicArchive/methodHandles` to check that the "Failed to generate LambdaForm holder classes" error is not in the output;
>> 
>> Passed tiers 1 - 3 testing.
>
> src/java.base/share/classes/java/lang/invoke/GenerateJLIClassesHelper.java line 333:
> 
>> 331:                                                     .warning("Invalid LF_RESOLVE " + parts[1] + " " + parts[2] + " " + parts[3]);
>> 332:                                         }
>> 333:                                     }
> 
> The underlying cause of the failure in CDS is [JDK-8327499](https://bugs.openjdk.org/browse/JDK-8327499) -- MethodHandleStatics.traceLambdaForm() includes methods that cannot be handle by GenerateJLIClassesHelper.java.
> 
> Therefore, I think the warning should always be printed (e.g., when jdk.tools.jlink.internal.plugins.GenerateJLIClassesPlugin is used during the JDK build).
> 
> Also, maybe add a comment above line 326: 
> 
> 
> // Work around JDK-8327499

Fixed.

> src/java.base/share/classes/jdk/internal/misc/CDS.java line 127:
> 
>> 125:      */
>> 126:     public static void traceLambdaFormInvoker(String prefix, String holder, String name, String type, MethodType mt) {
>> 127:         if (isDumpingClassList && isInvokerArchivable(mt, holder)) {
> 
> Is this check still needed after you added the filtering in GenerateJLIClassesHelper.java?
> 
> Since the logic here is not 100% the same as the logic in GenerateJLIClassesHelper.java, we might be filtering more things than necessary.

I did some sanity tests without the changes in CDS.java and it seems to work.
I'll do more testing before pushing another commit.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/17953#discussion_r1516980404
PR Review Comment: https://git.openjdk.org/jdk/pull/17953#discussion_r1516980534