RFR/RFC (XS) 8223911: Disable bad node budget verification until the fix

Aleksey Shipilev shade at redhat.com
Tue May 14 18:14:50 UTC 2019


There is a bug that Patric is handling now. However, fastdebug is broken for more than a week, and
there is no sight of the fix yet. Having this false negative failure is detrimental for testing,
especially given impending fork to 13. Backing out the change seems too harsh. Let's just disable
the assert?


diff -r 809cbe8565a1 -r ee3c58aebd44 src/hotspot/share/opto/loopnode.hpp
--- a/src/hotspot/share/opto/loopnode.hpp       Tue May 14 20:05:43 2019 +0200
+++ b/src/hotspot/share/opto/loopnode.hpp       Tue May 14 20:06:51 2019 +0200
@@ -1384,7 +1384,8 @@
     uint required = _nodes_required;
     uint delta = C->live_nodes() - live_at_begin;
-    assert(delta <= 2 * required, "Bad node estimate (actual: %d, request: %d)",
+    // Assert is disabled, see JDK-8223911 and related issues.
+    assert(true || delta <= 2 * required, "Bad node estimate (actual: %d, request: %d)",
            delta, required);

Testing: Linux x86_64 fastdebug, regression test from JDK-8223502 (I could have added it, but then
this fix would not be as trivial and would require passing jdk-submit), our larger workloads that
used to break


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20190514/7c99507a/signature.asc>

More information about the hotspot-compiler-dev mailing list