Created attachment 801673 [details]
On the one machine where I've tried to upgrade libvpx, =media-libs/libvpx-1.12.0 fails its test phase because the test VP9/RealtimeTest.IntegerOverflowLarge/0 is killed by the oom_reaper. In dmesg I see the following:
[14507368.857606] Out of memory: Killed process 29378 (test_libvpx) total-vm:15962740kB, anon-rss:6063680kB, file-rss:64kB, shmem-rss:0kB, UID:2
50 pgtables:29628kB oom_score_adj:0
[14507369.335786] oom_reaper: reaped process 29378 (test_libvpx), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB
And indeed, the total-vm figure there works out to over 15 GB, while "free" tells me this machine has (available to userspace) "7.7 Gi" of memory proper and "11 Gi" of swap, "3.5Gi" of memory and "10Gi" of swap free.
I would guess that either there's something wrong with this test, or the ebuild should check for the availability of some large amount of memory/swap?
Created attachment 801676 [details]
The test phase of versions 1.13.0 and 1.13.0-r1 are also reliably killed with out-of-memory errors (though I'm not *certain* that it's precisely the same test that triggers the OOM).