fix: win condition on frag + suicide

[WofWca]

This is the inconsistent behavior in question:

splash-damage-win-inconsistency.mp4

With the fix:

fix-win-condition-last-frag.mp4

(if you're wondering about the "teams are tied" announcement:
it's not about this MR. It's also reproducible in vanilla,
if you shoot first a teammate then an enemy with the railgun).

Compared to #73
this is a more general and robust fix,
also affecting railgun, shotgun, direct missile hits,
and telefrags.
Basically this ensures that weapon behavior does not differ
depending on whether there is 1 frag left or not.
That is, a railgun kills two players if they're aligned,
a rocket kills everyone who's in splash radius, etc.

I actually started working on this before #73 got merged,
but unfortunately I didn't mark it as draft in time.
So when/if this MR gets merged, that one
is basically not needed anymore and can be reverted for simplicity.
Or maybe the consistency is nice anyway,
at least for the order of log messages,
so only the comment needs to be adjusted at least.

Note that G_RunFrame still calls CheckExitRules() directly,
which probably makes sense, to keep things as is there.

Related:

• game: fix explosion hit registration being inconsistent etlegacy/etlegacy#2051.

[ec-]

Please, NEVER change physical location of vmMain(): it MUST be the very first function compiled into the .qvm file as it said in original comment.
Use proper forward declarations instead.

[WofWca]

I missed that, good point. Fixed now I think.

[ec-]

Sorry, I miss the point for a moment, but also, please avoid such things in general: f8582a1
q3lcc literally does nothing for inlining so you're just adding another abstraction layer/runtime overhead (+1 function call) for QVM, it may work for DLLs but will not work for QVMs, degrading its performance

[WofWca]

q3lcc literally does nothing for inlining

I didn't know that, interesting.
Do you have a simple suggestion in mind how we can avoid an extra call here, without having to care for each return in RunCommand?

[WofWca]

Ah, I think I can just make the function have 1 return. I'll do that tomorrow.

[ensiform]

Personally I don't think you're saving any real measurable difference with adding the run command extra function call over having the switch. Even in qvm compared to native there will be very little improvement

[WofWca]

Sorry, @ensiform I'm not sure what you mean? Regarding returns I'm more concerned with maintainability / readability. Please see if the new version clarifies it.

[ensiform]

This refactor doesn't really belong in the same pr as the other change IMHO. What does this change actually add?

[WofWca]

This refactor doesn't really belong in the same pr as the other change IMHO

How would I ensure to do if (level.needToCheckExitRules) CheckExitRules(); at the end of every vmMain() then? That's the point of the refactor.

[ensiform]

Why does that suddenly need to be called at the end of the vmmain?

CheckExitRules typically runs on per frame basis. Calling it whenever vmMain is invoked for one of the limited vm_calls seems extremely unnecessary as opposed to during runframe already because that's the only one which fires often. I don't see how it fixes the problem when you're just calling it multiple times per frame then.

[WofWca]

In vanilla (and current main) CheckExitRules runs not only at the end of the frame but also in ClientThink_real and IIRC RunMissile, after each individual frag, which there can be several of per single shot.
This MR makes it not run after each frag but instead runs it at the end of each command (event/frame).

Perhaps I'll need to adjust commit message to clarify this.
Look at the video and notice that I don't die. Only Crash dies. This is the bug.

[ensiform]

I'm confused because you already fixed that in the other commit allegedly from etlegacy similar message?

[WofWca]

Yes, I submitted that first but then came up with this supposedly better solution, which also fixes the bug for the railgun, the shotgun and everything. See MR description.
Edit: and it fixes the still remaining missile bug where during sudden death we'd kill self (or teammates) first, untying score and losing the game.