perf: refactor to reduce memory allocations #1086
Conversation
Cheong-Lau
force-pushed
the
small-optimise
branch
from
f1dca83 to
1429c31
Compare
|
oops, used |
Cheong-Lau
force-pushed
the
small-optimise
branch
from
1429c31 to
cf8f32d
Compare
| let dots = if toplevels.is_empty() { | ||
| (0..1) | ||
| .map(|_| { | ||
| container(vertical_space().height(Length::Fixed(0.0))) | ||
| .padding(app_icon.dot_radius) | ||
| .into() | ||
| }) | ||
| .collect_vec() | ||
| } else { | ||
| (0..1) | ||
| .map(|_| { | ||
| container(if toplevels.len() == 1 { | ||
| vertical_space().height(Length::Fixed(0.0)) | ||
| } else { | ||
| match applet.anchor { | ||
| PanelAnchor::Left | PanelAnchor::Right => { | ||
| vertical_space().height(app_icon.bar_size) | ||
| } | ||
| PanelAnchor::Top | PanelAnchor::Bottom => { | ||
| horizontal_space().width(app_icon.bar_size) | ||
| } | ||
| } | ||
| }) | ||
| .padding(app_icon.dot_radius) | ||
| .class(theme::style::Container::Custom(Box::new(move |theme| { | ||
| let dot_constructor = || { | ||
| let space = if toplevels.len() <= 1 { | ||
| vertical_space().height(Length::Fixed(0.0)) | ||
| } else { | ||
| match applet.anchor { | ||
| PanelAnchor::Left | PanelAnchor::Right => { | ||
| vertical_space().height(app_icon.bar_size) | ||
| } | ||
| PanelAnchor::Top | PanelAnchor::Bottom => { | ||
| horizontal_space().width(app_icon.bar_size) | ||
| } | ||
| } | ||
| }; | ||
| let mut container = container(space).padding(app_icon.dot_radius); | ||
|
|
||
| if !toplevels.is_empty() { | ||
| container = | ||
| container.class(theme::style::Container::Custom(Box::new(move |theme| { |
There was a problem hiding this comment.
I refactored this section completely, so I'd like a more thorough check of this part in particular.
I moved the logic of making the Elements into its own closure (better variable names are very welcome!), which should make the same Elements as before, just separate to the collection part and hopefully clearer. Then just make an iterator with std::iter::repeat_with, which avoids the whole mess of mapping and collecting an iterator.
Cheong-Lau
force-pushed
the
small-optimise
branch
2 times, most recently
from
772262f to
9f821d9
Compare
Cheong-Lau
changed the title
|
v2: I have now split the PR into two commits, one "safe" commit that should not change any behaviour, and one that uses unstable sorting, which could theoretically change behaviour but shouldn't in the places I've used them in. This PR probably still needs testing, but this should hopefully make that easier to do. |
Unstable sorting should be slightly faster than stable sorting, and if the vector was built asynchronously then preserving the initial order doesn't matter too much. Continue to use stable sorting where the vector is not built asynchronously, or if the vector is partially sorted (e.g. when new elements are pushed to a sorted vector).
Cheong-Lau
force-pushed
the
small-optimise
branch
from
9f821d9 to
f552e0d
Compare
Some minor refactors like removing
clone()s, replacing with references, avoiding allocatingVecs when they weren't necessary, etc.There's also a new
std::cmp::Ordimpl forActiveConnectionInfo, which should be heaps better than usingformat!. I had to also implstd::cmp::PartialCmpmanually to do that, but the actual logic there shouldn't matter as long as it agrees withstd::cmp::Ord.