I suspect that you are attempting to address something different to what
Alan had in mind here. I interpret Alan's response to be about
performance of the grouping, unrelated to whether ephemerality is a
thing or not.
(The inherent problems linked to ephemerality is discussed at length in
this thread)
On 2026-02-06 09:40, Michal Domagala wrote:
I would share my reflections
1. All "system" VT, "system" means f.e.created by
Structured Concurrency, automatically ends and never become unreachable
2. Only application VT can become unreachable, f.e. when a blocking
queue became unreachable
3. Every VT is naturally GC-able, because only diagnostic flag
trackAllThreads saves unreached VT from release
4. Diagnostic is important
5. If trackAllThreads use weak reference, it can impact GC, as
potentially there can be millions of VT
Maybe it is worth to consider `Thread.ofEphemeral()` tracked by weak
reference
1. Implementation effort is minimal, there will be two maps of VT
instead of one
2. Performance effort is zero, because only intentional, application
VT can increase GC time (Volenti non fit iniuria)
3. Diagnostic still works
Regards
Michal Domagala
pon., 12 sty 2026 o 13:41 Alan Bateman <[email protected]>
napisał(a):
On 10/01/2026 15:55, Alex Miller wrote:
> What is the likely future of the trackAllThreads flag?
TBD. It's clearly an attractive nuisance right now and setting it to
false is specific to the root "thread grouping". There is some
performance work required in that area but otherwise I think it
needs to
be removed.
-Alan
--
Cheers,
√
Viktor Klang
Software Architect, Java Platform Group
Oracle
