Exclude Netty's reflection configuration files

[zakkak]

These are causing warnigns during native image compilation as they try to access classes that are not on the
classpath (e.g. net.jpountz.lz4.LZ4Exception).

Closes #29413

[zakkak]

Scheduled CI runs for native:

• Linux: https://github.com/graalvm/mandrel/actions/runs/3548741986
• Windows: https://github.com/graalvm/mandrel/actions/runs/3548757419

[zakkak]

Windows run failed with:

The system cannot find the path specified.

I guess I am hitting what Sanne described, I will create two separate build items instead of a single one with a complex regex.

[zakkak]

Scheduled CI runs for native:

• Linux: https://github.com/graalvm/mandrel/actions/runs/3550749800
• Windows: https://github.com/graalvm/mandrel/actions/runs/3550623804

[zakkak]

Windows still failing, this time with:

Error: Unknown argument: quarkus-integration-test-resteasy-jackson-999-SNAPSHOT-runner

I will come back to this next week

[zakkak]

It looks like windows don't like .* in the exclude-config patterns.

@Sanne I changed the Oracle-JDBC pattern as well since it was also not working on windows. Please have a look.

It looks like an upstream issue to me so I will try to create a simple reproducer not relying on Quarkus and see where it takes me.

[Sanne]

@Sanne I changed the Oracle-JDBC pattern as well since it was also not working on windows. Please have a look.

are you sure of that? It was passing the tests AFAIK - or was this being disabled in some way?

[Sanne]

I'm confused as to why you don't need an ending * - is it also being treated as matching a prefix? Might need to clarify the documentation about that.

[zakkak]

Internally native-image uses Pattern.compile(excludeJar).matcher(path).find() which matches even without the * in the beginning or the end.

[zakkak]

I think we can also just quote the regex though. I will be able to test it in the next couple of hours.

[zakkak]

@Sanne I changed the Oracle-JDBC pattern as well since it was also not working on windows. Please have a look.

are you sure of that? It was passing the tests AFAIK - or was this being disabled in some way?

The only native test being run on windows appears to be resteasy-jackson

quarkus/.github/native-tests.json

Lines 135 to 140 in 6a2b1b4

	{
	"category": "Windows - RESTEasy Jackson",
	"timeout": 20,
	"test-modules": "resteasy-jackson",
	"os-name": "windows-latest"
	},

and I think it has been that way since day one.

[Sanne]

@zakkak many thanks for clarifying - I remember that before I separated the two exclusions it did fail on Windows, but I guess that was a compilation failure and not a test failure. Nice catch, thanks.

But in this case.. would you mind separating the two fixes? The oracle JDBC one might be important to backport, and even if we don't it would be nice to have it called out in the changelogs.

[zakkak]

But in this case.. would you mind separating the two fixes? The oracle JDBC one might be important to backport, and even if we don't it would be nice to have it called out in the changelogs.

Done in #29552

I also updated this PR to only include the Netty fix, but it depends on #29552 so I am marking it as draft (again).

[zakkak]

As mentioned in #29552 (comment)

This turns out to be more complex than I thought. The \" works fine on windows but not on linux.
As a result, for a quick-fix I am switching to my previous approach that does not rely on the quotes at the cost of using more expressions than one and avoiding the use of .*.

[gsmet]

Adding for backport to 2.13 because of #30508