Avoid use of id2ref for weak mapping #35
Conversation
|
Should we add jobs for JRuby and TruffleRuby? |
|
@kou I would be very happy to see a JRuby job. I'll see if there's anything needed to make it green today. |
|
There's a few failures so far. I'm not sure if they are real incompatibilities on JRuby or if they are just slight behavioral differences. |
|
I'll open a separate issue with JRuby test suite failures. |
[Bugs #15711] https://bugs.ruby-lang.org/issues/15711 The implementation of `ObjectSpace._id2ref` on JRuby (and TruffleRuby) is very expensive, and on JRuby we normally do not have it enabled because it impacts performance of the entire runtime. This uses `ObjectSpace::WeakMap` instead of `ObjectSpace._id2ref` by default and deprecates `DRb::WeakIdConv`. Authored-by: Masatoshi SEKI <m_seki@mva.biglobe.ne.jp>
|
OK. Let's work on enabling JRuby in CI as a separated task. |
|
The separated issue: #36 |
|
Oh, I should have used "Co-authored-by" not "Authored-by"... |
|
Would you consider releasing a new version of dRuby that includes this change?
|
|
Done. |
This doesn't seem needed any longer, especially since 2.2.3: ruby/drb#35
… parent, not stdlib version Without this, tests break on JRuby 10 with errors such as `0x0000000000000fb4 is not id value` because `_idstr` is not used any longer. We want ruby/drb#35 to be available, requiring 2.2.3 on both client and server.
3 participants
[Bugs #15711]
https://bugs.ruby-lang.org/issues/15711
The implementation of
ObjectSpace._id2refon JRuby (and TruffleRuby) is very expensive, and on JRuby we normally do not have it enabled because it impacts performance of the entire runtime.This uses
ObjectSpace::WeakMapinstead ofObjectSpace._id2refby default and deprecatesDRb::WeakIdConv.Authored-by: Masatoshi SEKI m_seki@mva.biglobe.ne.jp