WIP: events graph

[ThierryBerger]

trafficstars

to understand the flow of events, a graph of systems writing to events and reading which events might be useful.

example render on bevy_mod_picking (example minimal):

[jakobhellermann]

That's pretty cool, thanks!

Is there anything missing here functionality-wise that is still WIP or just polish? (making the function names consistent with the other graphs, and making the colors configurable).

[jakobhellermann]

Also is the example render filtered, or are there some systems not counted yet? This looks like it's missing a good amount of systems.

[ThierryBerger]

Indeed my first screenshot was a bit underwhelming, my initial implementation did only render 1 schedule, but in practice events are used throughout the schedules, so I updated the code to be able to display multiple schedules at once.

That's the latest state (as of commit 5615dadaca6362fc650890be0109d3f37519964f), running on bevy_mod_picking (example minimal):

I filtered out the event_update_system which is clearing the events, I guess we'd want that configurable.

I'd like to explore different things in this PR:

• consistency with other features from debugdump (you mentioned color style...)
• showing which schedule a system is from (by using the color I think). It can be challenging because a system could be in different schedules, and then how to display it ? :/
• test this a bit more to make sure displayed info is correct and relevant, I'm lacking expert knowledge on the analyzed crates to be too confident.

Then (as a future Pull Request), it could open up interesting analysis:

• detect when an event fired wouldn't be handled until next frame, a common source on "one frame delay", this should be possible by reconciliating schedule order and event order...
• Following almost the same code, we also could offer the same kind of graph for components or resources, to understand how a data is modified, where, when.
• parameters to filter for a specific type would be very useful to reduce the noise (in the screenshot, logs are very noisy)

[ThierryBerger]

@jakobhellermann any preference on how to tackle styling ? My plan (started with d6ab989f47d2e61f294b9042a76f8d49041ffbf4) is to mostly copy from the schedule graph render, adapting/removing things as I discover them.

Once we're happy with the features we'll probably be able to consider merging some data/functions, or keep them independent.

[jakobhellermann]

Yea, just copying it and letter checking if it makes sense to extract something shared is what I'd do aswell

[ThierryBerger]

Applying same system style as schedule graph, depending on their crate source :

[ThierryBerger]

Updated with subgraphs for schedules:

Now is a good time for feedback @jakobhellermann ; what's missing is probably:

• support for systems in multiple schedules (if that's a thing 🤔)
• some code cleanup, not too sure where, I tried to fit in the library's architecture
• I don't quite understand what's the benefit of those edges : https://github.com/jakobhellermann/bevy_mod_debugdump/pull/31#discussion_r1546361927 ; there's a cluster which I assume is useful to avoid "spaghetti" effect with arrows but I'm not too familiar with graphviz so your guidance is welcome.
• update readme

[ThierryBerger]

I like that subgraph with schedules, but there might be a problem with where to put the events: at the moment they belong to the first schedule they are encountered (from the graph list).

This leads to them appearing at different places if we change the filtering of systems or schedules.

See this unfiltered events graph

Arguably, event_update_system is a special case which we could make better ; my solution to address that was to filter it out by default and expose the filter so users can build up on that.

That said, there's probably more complex projects which might be bitten by that.

I'd like to consider that future improvements as I don't see any obvious fixes right now.

[ThierryBerger]

As an experiment I did display the events outside of their first schedule met, it looked worse :

events outside schedules (messy)

We could only render outside the events being "ambiguous" to different schedules, I think I like it:

events found in different schedules rendered outside

I think I'll need help to take a decision here @jakobhellermann ; or add it as an option ? I'll push what I have.

We still should order the way our systems/events are iterated on to avoid having different renders on similar runs, but I'd like to keep that for another issue.

[ThierryBerger]

I'm putting this in draft because while I think that's useful, it's missing information concerning World changes that could occur, and I doubt the approach taken by this PR is sound: how systems schedulers run their systems is not soundly deducible (but bevy built-in systems may be enough to be useful 🤷).

It may be interesting to consider consolidating it by compile-time analysis :thinking: ( see https://github.com/BD103/Bevy-Meetup-7 ; https://doc.rust-lang.org/nightly/nightly-rustc/)