[mcculls]

What Does This Do

Small pre-release cleanups:

• Restore short-circuit evaluation of class-name filter character checks.
• Avoid allocating an extra return-descriptor suffix when constructing method matchers, while preserving exact return-type matching.
• Align InternalNames construction with its lookup logic and assign the completed tables to final fields.
• Use the literal 4096 for the trie long-jump table limit.

Motivation

Reduce avoidable work and simplify internal code ahead of the release.

Additional Notes

Validation: ./gradlew check (tests on Java 8, 11, 17, 21, and 25, Spotless, and SpotBugs); git diff --check.

Contributor Checklist

• No Jira ticket provided.

[bric3]

Good cleanup, and improvement, indeed I prefer assigning values after their "assembly" in the for loop.

Also, why not using as 0xF000 as literla value (pure nitpick).

[mcculls]

Also, why not using as 0xF000 as literla value (pure nitpick).

MAX_LONG_JUMPS is used for counting purposes, so it uses a decimal value of 4096 rather than 0x1000

Whereas LONG_JUMP_MARKER is used for bit operations so it uses 0xF000

Also note that LONG_JUMP_MARKER deliberately uses a series of bits, rather than just a single bit because that way we can maximise the boundary at which point a jump needs to become a long jump and gets stored in a table. In this case jumps can go all the way up to 61439 (0xEFFF) before they are categorised as long jumps. But this then means that the long-jump id, once the marker is removed, only allows 4096 long jumps (including the zero id).

Given the typical contents stored in a trie, we're more likely to have a few long jumps so the design deliberately tunes for that scenario.