[PATCH] SimpleGraphic: use blocking Windows event loop

[sthalik]

Having found the source code for SimpleGraphic in #28, I'm working on reducing the software's CPU usage. Initially, asynchronous processing like passive tree download broke, but there's a simple realization as to when we need to refresh the tab. Background processing is done via "sub-scripts", and we need to perform refreshes if and only if a "sub-script" exists.

I implemented a working blocking event loop, as well as special-casing for the "sub-script" case. Given there's no repository for SimpleGraphic, I made a new repository: <https://github.com/sthalik/PathOfBuilding-SimpleGraphic>. You can compare the code to the public SimpleGraphic release. The most recent mingw-w64 commit isn't needed for Visual Studio at all though.

Whether this approach suits you or not, I urge you to implement event handling in some way. It's an idiom to register for and listen for events in the main loop. POB is anomalous here in how it performs lots of needless work while it's running.

[kvnxiao]

This needs more exposure.

[sthalik]

FWIW, you can download a binary version from dropbox at <https://www.dropbox.com/s/gvusyqjti4mtglt/SimpleGraphic.dll.dat>. Remove the .dat extension. You can try it in a VM or a sandbox if you don't trust an arbitrary github user. It's a drop-in replacement for a working POB installation, whether from git or otherwise.

Here's a technical description:

Before, the event loop proceeded immediately even if no events were queued. Then the screen was unconditionally redrawn from scratch from the current Lua script. The event loop possibly slept for a few milliseconds, blocking the UI for the duration, then checked for events again.

I initially tried to wait for an event indefinitely. There are however aforementioned "sub-scripts" doing stuff in the background, like importing the current character state. They need to be periodically poked in order to finish.

The proper solution would be to explicitly register for events (such as socket ready state, timeout, etc). That's what UI toolkits such as Qt do. However, all the tab logic is located in OnFrame with no explicit expectation as to when the loop needs to run again.

Given the less-than-optimal circumstances, an observation -- only stuff that runs in the background are the sub-scripts. Knowing that, the loop becomes:

• wait for an event[1] for 100 ms, process it if it arrives earlier
• if had an event, process it
• if had an event OR there's a sub-script, run OnFrame
• otherwise, skip running OnFrame and keep current window content
• go to the beginning of this list

[1] mouse move, key/mouse press, window resize, etc.

Not only there's no need to fix main loop logic in the lua scripts, but the problem looks solved. It's enough to prod sub-scripts every 100 ms if they exist at all. They exist very briefly so we don't need to prod them a lot.

[sthalik]

@Openarl can you take a look please?

[trollkind]

Thanks for the referencing in #1082
I expected something like this to be the culprit. Shame if nothing is done about it.

[ijemafe]

@Openarl Bumping this issue - high CPU usage is still an issue for SimpleGraphic on laptops. After testing this patched version the CPU usage is significantly reduced. Would be great to have this included in the official release.