Issue metadata
Sign in to add a comment
|
34kb regression in resource_sizes (MonochromePublic.apk) at 583155:583155 |
||||||||||||||||||||||
Issue descriptionCaused by “Add range and arithmetic checks to `base::VectorBuffer`.” Commit: b586d770ad4c96daaef4104e89e43126568c6e3f Link to size graph: https://chromeperf.appspot.com/report?sid=bb23072657e2d7ca892a1c3fa4643b1ee29b3a0a44d0732adda87168e89c0380&num_points=10&rev=583155 Link to supersize symbol diff: https://logs.chromium.org/logs/chromium/buildbucket/cr-buildbucket.appspot.com/8938165736347884272/+/steps/Show_Supersize_Diff/0/logs/s_____Show_Diff____/0 Debugging size regressions is documented at: https://chromium.googlesource.com/chromium/src/+/master/docs/speed/apk_size_regressions.md#Debugging-Apk-Size-Increase Based on the graph, size is from native code. Given the nature of the change, it sounds like the increase is expected and "worth it". If so, then please close as "Won't Fix".
,
Aug 15
Assigning to palmer@chromium.org because this is the only CL in range: Add range and arithmetic checks to `base::VectorBuffer`. Bug: 817982 Change-Id: If8602db9f284873f037775f3a37bcac8e120b313 Reviewed-on: https://chromium-review.googlesource.com/1166456 Reviewed-by: Daniel Cheng <dcheng@chromium.org> Reviewed-by: Lei Zhang <thestig@chromium.org> Commit-Queue: Chris Palmer <palmer@chromium.org> Cr-Commit-Position: refs/heads/master@{#583155}
,
Aug 22
This is a bit surprising. My local testing for object code size (Linux, 64-bit, debug = false, non-Release build) resulted in exactly 0 size change. (Which then sent me on a disassembly adventure to ensure the checks were there! They were, and fit inside the usual slack space in the object code.) So it might be that a Release build has no/less slack space, and hence the checks become visible in object code size. Or perhaps the compiler for Android is different than for Linux, and different code is generated? 34 KiB is also larger than I would have expected, unfortunately. Still, my view is that this is "worth it". It may be that (now-redundant) checks in `VectorBuffer` call sites, if any, can be removed now. I'll keep this bug open for now to remind me to see if there is any of that going on.
,
Aug 22
,
Aug 22
Confirmed that I could not find any now-redundant `CHECK`s in `VectorBuffer` call sites. |
|||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||
Comment 1 by 42576172...@developer.gserviceaccount.com
, Aug 15