1683897 - 0.1 - 0.14% installer size (osx-shippable) regression on push 1b3cd56bc7db0ba8fa5f0db0cfb2633242575ef0 (Tue December 15 2020)

Status: RESOLVED WONTFIX

(Firefox Build System :: General, defect)

Description

[Alexandru Ionescu (needinfo me) [:alexandrui]]

Perfherder has detected a build_metrics performance regression *around the date of Dec 17 of this graph. The regression size is about 500K, five times bigger than the build metrics threshold allowed. The regression seems to be gradual, so there might be multiple culprits, maybe several parts of a bug that landed progressively.

Regressions:

Ratio	Suite	Test	Platform	Options	Absolute values (old vs new)
0.14%	installer size		osx-shippable	nightly	84,569,564.42 -> 84,689,028.75
0.10%	installer size		osx-shippable	instrumented	112,725,650.21 -> 112,834,542.50

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs.

For more information on performance sheriffing please see our FAQ.

Comment 1

[Mitchell Hentges [:mhentges] 🦀]

From d244b06b1baf6ea4e26e206106b7083135c0ba3f (December 14th) to a4ce10afc0f660ece5eca5f3d700646d8dd19c4c (December 15th):

[./Contents/MacOS/XUL] 192k
[./Contents/Resources/browser/omni.ja] 108k
[./Contents/Resources/omni.ja] 4k
[./Contents/MacOS/libnss3.dylib] 4k
[./Contents/MacOS/libmozglue.dylib] 4k
[./Contents/Resources/browser/features/[e-mail]] -8k

• Updating JSBeautify cost us roughly 100kb.
• There's a 192k size regression to XUL, but it's more unclear what caused that:

mhentges@mhentges-64lvdr diff-city % bloaty new.app/Contents/MacOS/XUL -- old.app/Contents/MacOS/XUL 
    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.1%  +123Ki  +0.1%  +123Ki    __TEXT,__text
  +0.1% +40.2Ki  +0.1% +40.2Ki    String Table
  +0.4% +10.9Ki  +0.4% +10.9Ki    Export Info
  +0.1% +7.25Ki  +0.1% +7.25Ki    Symbol Table
  +0.8% +5.68Ki  +0.8% +5.68Ki    __TEXT,__unwind_info
  +0.1% +4.45Ki  +0.1% +4.45Ki    __DATA,__const
  +0.0% +1.23Ki  +0.0% +1.23Ki    __TEXT,__cstring
  +1.3%    +805  +1.3%    +805    [__TEXT]
  +0.1%    +544  +0.1%    +544    Function Start Addresses
  +0.1%    +240  +0.1%    +240    __DATA,__data
  +0.1%    +224  +0.1%    +224    Rebase Info
  [ = ]       0  +0.0%    +144    __DATA,__bss
  [ = ]       0  +3.1%     +64    __DATA,__thread_bss
  +1.4%     +24  +1.4%     +24    __DATA,__thread_vars
  +0.0%      +8  +0.0%      +8    __TEXT,__gcc_except_tab
  -0.0%      -4  -0.0%      -4    Indirect Symbol Table
  -0.2%      -8  -0.2%      -8    __DATA,__got
  -0.2%    -112  -0.2%    -112    Table of Non-instructions
 -43.3%    -720 -25.9%    -928    [__DATA]
  -0.0% -3.64Ki  -0.0% -3.64Ki    __TEXT,__const
  +0.1%  +190Ki  +0.1%  +190Ki    TOTAL

Due to noise in the file size due to instrumentation and PGO, it's tough to see which specific revision is causing the extra ~190k. At the same time, it's 190k, so I'll close this here.