summaryrefslogtreecommitdiff
path: root/libs/androidfw/StringPool.cpp
diff options
context:
space:
mode:
author Zimuzo Ezeozue <zezeozue@google.com> 2025-02-26 16:16:33 +0000
committer Zimuzo Ezeozue <zezeozue@google.com> 2025-02-26 16:16:33 +0000
commitadd9be1adc183c2fc2f78a2be7606f2245c7c5af (patch)
tree08a38e5dd8a0a4a9ce647cb6b7b37329291b1404 /libs/androidfw/StringPool.cpp
parenta4fe86da7a864d94d82d5e04f216d0559e07ee41 (diff)
Fix Builder allocations
With the change to remove the Builder interface, we no longer have a no-op builder to use when the category is disabled. This means, we use the Builder pool even when a category is disabled. There was an early return when initializing the Builder that didn't reset the pool. This meant we'd quickly max out the pool and start allocating Builders on subsequent events. The quick fix is to remove the early return so we always reset the object pools. This would have some perf impact (always reseting) when tracing is disabled, but will fix the object churn. The proper fix is a coming rewrite to remove the Builder entirely. Also, temporarily ignore a failing test while I investigate why this wasn't caught in presubmit. The test is failing because of MQ tracing changes here: I958199dcffc0a597233b7c9df15fffee6d5b40d8 Test: atest PerfettoTraceTest Bug: 303199244 Fix: 398439018 Fix: 393920553 Flag: android.os.perfetto_sdk_tracing_v2 Change-Id: I4c349267c5be54be3bcf35f8463e7e46d42b1fd6
Diffstat (limited to 'libs/androidfw/StringPool.cpp')
0 files changed, 0 insertions, 0 deletions