Status Update
Comments
si...@google.com <si...@google.com> #2
API 28 P, Pixel (1)
androidx.compose.foundation.benchmark.text.TextBasicBenchmark_layout_length_32_ : 0.6ms -> 1ms
- 512 did not show any changes
- creates 10 Text in a column
- therefore on average each text went from 0.06 to 0.1ms (still 50% but 0.04ms increase for each text)
- in the tests Paragraph 32 went from 0.27ms to 0.31ms (0.04ms increase) on a Jio device. weird coincidence 0.04 matches here.
- Paragraph construct was the major regression in the local tests (13%).
- why did ParagraphBenchmark 32 not give alert?
does not have a 32. It was disabled before with aosp/1743155ParagraphBenchmark
Other increases are similar:
- androidx.compose.foundation.benchmark.text.TextBasicAnnotatedBenchmark_layout_length_32_: 0.6ms -> 1ms
- androidx.compose.foundation.benchmark.text.TextDelegateBenchmark_layout_length_32_styleCount_0_: 0.014ms -> 0.041ms
- androidx.compose.foundation.benchmark.text.TextMultiStyleBenchmark_layout_length_32_styleCount_0_: 0.6ms -> 1ms
si...@google.com <si...@google.com> #3
size 32 matches to a boring layout, boringlayout platform handling should be added to prevent regression on Android T (already in todo list)
I will resolve this ticket, please let me know if you think that this is an issue.
si...@google.com <si...@google.com> #4
Talked this with the team.
- The regression happens on shorter strings.
- it prevents clipping on older Android versions, it is expected to cost us performance
- the code does not execute on platform versions where fallbackLineSpacing is applied StaticLayout (Android 28) or BoringLayout(Android T)
Closing this as expected result.
ma...@gmail.com <ma...@gmail.com> #5
Ai.a13(key)MSG_0
Description
Performance regression was expected especially for shorter text.
However the jump for 32 char count is %50 (600 to 900)
Previously measured expected value was 10%
creating the ticket to check the dashboard in detail.