First of all: Take all measurements I mention with a grain of salt – all of them are rough and not necessarily measured more than a few times. All times were measured on an Xperia X run via SDK.
Visiting a chat page can take a long time, especially before the qml is cached by the engine.
When opening it for the first time after application launch, it sometimes takes >1000ms from onClicked (OverviewPage) to Component.OnCompleted (Chatpage).
Subsequent activations take roughly 470-480ms.
With these changes, I was able to reduce these times to ~450ms for the first, ~100ms for subsequent activations of the ChatPage on my test device.
Things changed:
- The components for displaying extra content to a message are (mostly) gone and replaced by a single Loader. This Loader does not use sourceComponent to trade the initial compilation boost for a neglegible bit of runtime penalty.
- Connections were consolidated
- I was surprised how costly the inclusion of the RemorseItem was (compiling ~75ms, initializing up to ~20ms for every delegate). So I traded a bit for a compromise. deleteMessageRemorseItem is now defined on the appWindow level, where it gets a bit mitigated by the animations at application start. Also, only one deletion at a time is now possible. We can easily revert this change, but I thought it worthwhile despite its drawbacks.
- profileThumbnailComponent is now defined directly as sourceComponent, removing the need for its id. Probably didn't do anything.
- InReplyToRow had width: parent.width, so I removed horizontalCenter. Also probably didn't change compilation time at all.
- Another compromise I was willing to take – your opinion may differ: The PickerPages took ages (~200ms) to just parse/compile inside those Components, so I replaced them with the "string notation" of pageStack.push. Drawback: The first time a picker gets activated, you'll see how slow it is. Subsequent activations aren't that bad – also for the other pickers.
- The new emoji basePath workaround also works for TextSwitch content.
- Checking which context menu entries get added for the poll preview didn't quite work nicely "OnCompleted".
Closes#102.
Doesn't use Shader/HighlightImage for background/placeholder, since that neither matched the previous look nor meet my visual expectations at all. I created both previously available color variations as svg accordingly.
Implementation of the background images has been outsourced to a dedicated qml file to reduce duplication and ensure the same layout. Padding is now "medium" everywhere (stickers had none, audio had "small").
For the Icons, some (Highlight-)Image+MouseArea combinations have been changed to IconButton.
Regarding videos, I'm not actually sure if the missing "highlight" on the fullscreen icon was intentional. An issue with videos possibly having the same color as the icon remains basically the same. Nonetheless, if this isn't desired, I'll revert to just using the white image.
The sticker icon actually does look a bit differently now – it's a bit smaller. I've redrawn it from scratch but mostly kept the dimensions of icon-m-other (which doesn't look good when zoomed in, btw). Perhaps I actually did made a mistake drawing the previous one. Either way, I think it fits well enough.
The notification icon remains png, but I managed to reduce its file size by ~60% without visibly impacting it much imho.
This may be the longest commit message I've ever written.
It just doesn't make any sense.
Reference (essentially a pointer) can't be smaller than a primitive
type, except for 64-bit numbers on a 32-bit platform. In terms of
performance, an additional indirection can't be better than picking
the value directly from the stack, let alone a register.
