On Thu, 4 Jun 2026 13:28:28 GMT, Viktor Klang <[email protected]> wrote:

>> This is how it is in mainline, I don't think I should introduce noisy diff 
>> to the mainline review PR.
>
> Makes sense. I wonder if it would be easier (for future PRs in this space) to 
> do the mainline-alignment as a separate, initial, PR, then do a follow-up PR 
> with the specific changes.

I don't think I can - lworld is well tested so it's almost impossible to split 
the mainline alignment out without causing some test failures (

>> That would be an assertion on the `getDeclaredFields` facility. I am more 
>> concerned about this annotation processing because the `@driver` clause is 
>> more likely to be accidentally omitted or have typos in test directives and 
>> becomes hard to debug. There is no such problem with 
>> `Class.getDeclaredFields` so I don't think field omissions is a problem.
>
> I think there's a semantic difference between asserting "any fields returned 
> must be strict" and "two fields are to be returned, named "x" and "y" and 
> they must both be strict".

There is, but I don't think this argues for complicating such a sanity check. 
You can always request extra coverage for every failure path of every test you 
write; it is an endless rabbit hole, and I argue a misused `@driver` has no 
compile-time or validation from other tests that it deserves some sanity test; 
meanwhile `getDeclaredFields` is well-tested and much less error-prone, so 
that's where I put the stopping line.

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

PR Review Comment: 
https://git.openjdk.org/valhalla/pull/2326#discussion_r3357049757
PR Review Comment: 
https://git.openjdk.org/valhalla/pull/2326#discussion_r3357043709

Reply via email to