canmove, Confirmed users
1,448
edits
L20n/Firefox/Performance: Difference between revisions
Jump to navigation
Jump to search
m
→Hypothesis 5: Lazy-to-actual XULElements conversion: fix the ascii table
(H5: Lazy-to-actual) |
m (→Hypothesis 5: Lazy-to-actual XULElements conversion: fix the ascii table) |
||
| Line 74: | Line 74: | ||
Are we working against XUL layout optimizations? XUL elements are first created as XULPrototypeElements which only carry just enough information to be useful for layout. When they're accessed from JS code they are converted to actual XULElements. The hypothesis is that there are at least two things when creating a new browser.xul window that trigger this conversion before tpaint: our code and something else, let's call it X. The first piece of code to make the call pays the lazy-to-actual tax. If it's us (because we start very early, like when we trigger on MozBeforeLayout), we take a lot of time (see Hypothesis 2!), but still finish pretty early, then X runs faster and tpaint happens in 270 ms (on stas's machine). If it's not us (like when we trigger on DOMContentLoaded) then X takes more time; we run later but finish faster (we take 20 ms to translate everything), and... tpaint hapens in 270 ms. For tpaint it only matters if the tax has been paid. Below are the rough numbers in milliseconds: | Are we working against XUL layout optimizations? XUL elements are first created as XULPrototypeElements which only carry just enough information to be useful for layout. When they're accessed from JS code they are converted to actual XULElements. The hypothesis is that there are at least two things when creating a new browser.xul window that trigger this conversion before tpaint: our code and something else, let's call it X. The first piece of code to make the call pays the lazy-to-actual tax. If it's us (because we start very early, like when we trigger on MozBeforeLayout), we take a lot of time (see Hypothesis 2!), but still finish pretty early, then X runs faster and tpaint happens in 270 ms (on stas's machine). If it's not us (like when we trigger on DOMContentLoaded) then X takes more time; we run later but finish faster (we take 20 ms to translate everything), and... tpaint hapens in 270 ms. For tpaint it only matters if the tax has been paid. Below are the rough numbers in milliseconds: | ||
Trigger l20n-start io-start io-end trigger/translateRoot-start formatting-start (async) formatting-end translateRoot-end tpaint | |||
DOMContentLoaded | ------------------------------------------------------------------------------------------------------------------------------------------------------ | ||
MozBeforeLayout | DOMContentLoaded 35 40 140 160 165 175 180 270 | ||
MozBeforeLayout 8 10 35 40 65 135 140 270 | |||