Entity-less native compilation failures

[Sanne]

Fixes #2482 and Fixes #5262

The culprit was the fact that when there's no entities, we'd bail out early.

In fact we'd abort a little too early, as it would by-pass setting a couple of key static fields which I had intentionally crafted to guide the code analysis of the native-image compiler.

Example symptoms include issues which seem related to entirely different components, such as:

Trace:
	at parsing java.lang.System$2.getConstantPool(System.java:1227)
Call path from entry point to java.lang.System$2.getConstantPool(Class):
	at java.lang.System$2.getConstantPool(System.java:1227)
	at sun.reflect.annotation.TypeAnnotationParser.parseAllTypeAnnotations(TypeAnnotationParser.java:323)
	at sun.reflect.annotation.TypeAnnotationParser.fetchBounds(TypeAnnotationParser.java:299)
	at sun.reflect.annotation.TypeAnnotationParser.parseAnnotatedBounds(TypeAnnotationParser.java:244)
	at sun.reflect.annotation.TypeAnnotationParser.parseAnnotatedBounds(TypeAnnotationParser.java:237)
	at sun.reflect.generics.reflectiveObjects.TypeVariableImpl.getAnnotatedBounds(TypeVariableImpl.java:241)
	at com.oracle.svm.reflect.TypeVariable_getAnnotatedBounds_88ea462da2b17ea45028db307519aaeeec9a4036_776.invoke(Unknown Source)
	at java.lang.reflect.Method.invoke(Method.java:498)
	at org.jboss.resteasy.core.ContextParameterInjector$GenericDelegatingProxy.invoke(ContextParameterInjector.java:118)
	at com.sun.proxy.$Proxy225.hashCode(Unknown Source)
	at java.util.HashMap.hash(HashMap.java:339)
	at java.util.HashMap.get(HashMap.java:557)

I've also seen this manifest as failures around:

• JTA logging system, which defines an illegal invocationhandler during initialization
• Various Byte Buddy helpers doing illegal stuff like reading annotation parameters in toString(), equals() and hashcode implementations
• the RESTEasy default invocation handler attempting to define a class
• (unexplained) attempts to load javax/security/jacc/PolicyContextException
• failure as it would reach io.undertow.protocols.alpn.OpenSSLAlpnProvider.setProtocols
• .. !

The most puzzling failure I got was this one:

        at java.lang.reflect.Method.invoke(Method.java:498)
        at javax.enterprise.util.AnnotationLiteral.invoke(AnnotationLiteral.java:288)
        at javax.enterprise.util.AnnotationLiteral.getMemberValue(AnnotationLiteral.java:276)
        at javax.enterprise.util.AnnotationLiteral.toString(AnnotationLiteral.java:135)
        at java.lang.String.valueOf(String.java:2994)
        at java.lang.StringBuilder.append(StringBuilder.java:131)
        at com.oracle.svm.core.amd64.AMD64CPUFeatureAccess.verifyHostSupportsArchitecture

at which I almost gave up 🔢 😕

[jaikiran]

Agreed. I gave up on this one a while back for the same reasons as you state - every time I thought I had narrowed this down and did some change to "fix" it, it would show up with some other odd and hard to understand error :) Glad you narrowed it down to this fix.

[geoand]

Looks good to me, but I don't know enough about the hibernate integration to provide any valuable comments or feedback.
Who should review this one @Sanne?
Seems like we definitely want it in sooner rather than later.

[rsvoboda]

@Sanne, issue I commented about in #2482 (comment) is no longer visible to me with this fix.

[Sanne]

@rsvoboda thanks for checking! I forgot about that one. Is there another issue I should mark resolved by this one or are we good as it is?

[Sanne]

@geoand anyone can review it. The Hibernate integration is stuff I did myself, so as long as it fixes stuff and nobody sees any risk of regressions, I just need an approval to merge ;)

[geoand]

It seems absolutely reasonable to me. I'll let @rsvoboda verify that everything checks out with the other issue as well

[Sanne]

A suggestion to any reviewer: revert the patch, keep the integration test. Look how horribly it fails :-)

[Sanne]

thanks for the approval @geoand ! I'm fairly confident about this being necessary, I'll merge.

If @rsvoboda could check if there's more issues to be marked as resolved, we can do that when he sees this.

[geoand]

@Sanne makes sense!

[jaikiran]

@Sanne, I'm not 100% sure, but I think native tests that run against the automated CI are selectively enabled (in https://github.com/quarkusio/quarkus/blob/master/ci-templates/stages.yml?). So I think this new integration-test will have to be included in there for it to catch any regressions in native mode.

[geoand]

@jaikiran your understanding is correct!

However in this case I don't think we need to run the integration test in native mode, since it's more about addressing a build time issue.
Up to @Sanne of course to decide :)

[jaikiran]

Hello @geoand,

However in this case I don't think we need to run the integration test in native mode, since it's more about addressing a build time issue.

Just so I understand how things are run - does this mean that a native image will be created for this integration test, even if it's not explicitly listed in that CI template? So, that CI template just controls which native tests are run against these (always) generated native images?

[geoand]

Hey @jaikiran

No, no native image will be created in this case, just a regular jvm build and test.

[rsvoboda]

@Sanne didn't find any additional unresolved issue related to this PR / topic

[Sanne]

@geoand this failure would happen during a native-image build though, so I think @jaikiran is correct that CI wouldn't detect regressions here.

That might need a follow up, I'll open an issue to track as I can't do it myself today.

[geoand]

Oh OK. I thought it showed up in a normal build as well, my bad :)

[machi1990]

@geoand this failure would happen during a native-image build though, so I think @jaikiran is correct that CI wouldn't detect regressions here.

That might need a follow up, I'll open an issue to track as I can't do it myself today.

Hi @Sanne @geoand, I just landed on this space. To be sure I followed the missing piece for CI is it about adding jpa-without-entity module in native ci? If so I can come up with a PR.

[geoand]

@machi1990 yes, that is correct

[machi1990]

Cool, let me do it.

[Sanne]

Many thanks @machi1990 ! Opened #5369

[Sanne]

@galderz have a look at the errors this solved - both the "stacktrace" (it's not a stacktrace) I pasted in this PR and the linked issues it resolved.

[hakdogan]

Hi,

I want to report a problem I think is related to this issue.

I have an entity-less application that has hibernate-orm dependency and uses QuarkusApplication. One of the app's dependency includes the entity and it has beans.xml.

First I got NoClassDefFoundError exception for javax/security/jacc/PolicyContextException in native complation time.

After I added the javax.security.jacc-api dependency to the pom, I got the following error

Error: No instances of java.net.SocksSocketImpl are allowed in the image heap as this class should be initialized at image runtime. To see how this object got instantiated use -H:+TraceClassInitialization.

Quarkus version 1.9.2.Final
GraalVM version 20.2.0

[Sanne]

Hi, if you have an PolicyContextException exception in native, adding jacc-api is not the right solution; one would need to understand why this class was being attempted to be loaded -especially important as it seems from the second error that this is bringing you into a rabbit hole :)

Why do you have a dependency on hibernate-orm if you're not using it?

Probably best to open a new issue.

[hakdogan]

Hi,

Why do you have a dependency on hibernate-orm if you're not using it?

I use it, only my entity object is in a dependency. As you said, it will be better to open a new issue and discuss the issue there.