Release process for 0.2.7.1

[Your_mom]

A bug occured while i was playing on swampys a 2.7.0 server i think. I was about to start grinding the outer rim i turned right and then a split second later the cycle was still facing right but started going left.trail was invisable then another player outgrinded me from theright side came out in front of me on the left and cored me before my wall became visable to me.

Sorry, you confused me with all these lefts and rights. Can you say it again with north/east/south/west? Basically, what I understand is that you tried to turn right and the server let you drive left. As far as I can tell now, this sounds like a server bug; I suppose you were using your brake as well and released it just as you turned, is this correct? Because then, both commands would go to the server simultaneously and 0.2.7.0 servers were confused by that. ( I think I fixed the server bug that caused such things ). Of course, the rendering errors are still client problems, but they will be impossible to fix for 0.2.7.1 because they require big code changes.

Re-explaining what happened-

i was heading north,i turned east at the rim, server pushed my trail west ,bike model turned west soon after the trail, only the trail became invisable on my screen. Another player came from the east(outgrinding me) and killed me after coming out on the west ahead of me only after death did i finally see my trail. hope thats more comprehendable

odd occurences-

i have noticed that in 2.7.1 walls are extremly "sticky" during or right after a deep grind the bike tends to stop moving on my screen for mabey 3-5 seconds only to snap back to where it should be with half of the turns i made while stuck. seems to be comparable to sliding only much much worse when it happens.has happened in luci's server and in swampys

[SuPeRTaRD]

i installed the new windows build, & played first at the crashme test server & then at microbus & experianced much the same game i feel when i serve over our LAN.. NO sliding for me yet. even in tight mazes

I did feel what your_mom is describing with the visuals on the rubber lasting too long, but it didnt kill me.

it seemed like i had more rubber than other players without 2.7.1

they asked how i did what i just did, when i made it thru some hairy turns...

anyways, it FEELs great to me. i am on a sempron 1000mhz with a gf2mx card & dialup 56k v.90

[Z-Man]

Re-explaining what happened-

i was heading north,i turned east at the rim, server pushed my trail west ,bike model turned west soon after the trail, only the trail became invisable on my screen. Another player came from the east(outgrinding me) and killed me after coming out on the west ahead of me only after death did i finally see my trail. hope thats more comprehendable

Yes. Much better. Apart from the fact that your wall was invisible ( this is the same effect that causes ghosting and other invisible wall effects ), its a known and fixed server bug if you were using the brake.

i have noticed that in 2.7.1 walls are extremly "sticky" during or right after a deep grind the bike tends to stop moving on my screen for mabey 3-5 seconds only to snap back to where it should be with half of the turns i made while stuck. seems to be comparable to sliding only much much worse when it happens.has happened in luci's server and in swampys

So you grind a wall, turn away from it and your cycle sticks to the wall? Or does it stick in mid-grind without you making a turn?

[Z-Man]

The bug mentioned here may be important:
http://guru3.sytes.net/viewtopic.php?t=1637

Is it serious enough to fix for this release? ( read: new source tarball, new binaries... ) I'll fix it in the release branch anyway, the versions we used for rc3 are tagged as the "official" versions for now.

Just quoting myself because the last post at the end of a page often missed
Update: The bug and another followup bug ( reference counting overflow causing early object destruction and crash ) are fixed in the branch.

[Your_mom]

Yes. Much better. Apart from the fact that your wall was invisible ( this is the same effect that causes ghosting and other invisible wall effects ), its a known and fixed server bug if you were using the brake.

there where no brakes involved, i dont see how it would make a difference tho

So you grind a wall, turn away from it and your cycle sticks to the wall? Or does it stick in mid-grind without you making a turn?

its odd mabey nemo could explain better as he claimed to have experience the same thing as well. Sometimes when grinding as soon as you hit the wall and turn the turn doesnt show up on the screen, i turn frantically and then suddenly its shows i was never really stuck at all only many of the turns i made didnt register, in an open space it isnt so bad, but making blind turns trying to figure out where your bike may actually be is no fun. And sometimes only half of the turns you make register, example zig zags may result in repeated u turns instead.

[Z-Man]

its odd mabey nemo could explain better as he claimed to have experience the same thing as well. Sometimes when grinding as soon as you hit the wall and turn the turn doesnt show up on the screen, i turn frantically and then suddenly its shows i was never really stuck at all only many of the turns i made didnt register, in an open space it isnt so bad, but making blind turns trying to figure out where your bike may actually be is no fun. And sometimes only half of the turns you make register, example zig zags may result in repeated u turns instead.

I think I've seen it now, too. It happens whenever you really use your rubber to start the grind for more than a fraction of a second. I'll investigate what causes this and how it can be prevented client-side even if the problem probably is server-side, it's really annoying when it happens.

Edit: Here's my analysis from debugging:
You drive towards a wall. You want to grind, so you let your cycle collide with the wall and let the rubber work for you. There's rubber code both on the server and the client. The rubber code on the server will bring your cycle very close to the wall, in fact, it will bring it directly onto the wall. ( Clientside, the 0.2.7.1 will stop you a configurable amount of space before the wall ). The server sends a sync to your client. The client accepts the sync and places you onto the wall, as well. ( The old client, in this situation, will let you tunnel through the wall locally and you're dead on your next turn. ) You turn. The turn command will be issued with you directly on the wall. The server diligently follows your command with the known bad consequences of driving exactly on a wall: Nothing is reliable in this situation.
I'll try to write fix code for exactly this situation ( and I hate it, but it has to be done ). When the server sends a sync that is not too far from the client's current position, it gets treated with care.
Note: on a 0.2.7.1 server, all this should not happen

[Lackadaisical]

nothing really substantial to add, but i have to say that 2.7.1 HAS GRANTED ME SUPERPOWERS!

Wow I must say i was very suprised how much difference there is than using 2.7 on Swampland. Like Supertard already mentioned it seemed like I had so much more rubber than normal, and i was awed (in lack of a more nuanced verb) how much closer to the wall i ended up than before. I've always wondered if there was something more than just my reactionrate (?) involved in grinding, and I guess that there appearantly is.

[Sabarai]

i ended up grinding very very tight on swampland too lack..

[Tank Program]

Well, I tried building the rc3 source today, and sure it compiles, but as soon as you run the exe it produced it crashes, the exe I mean. So you see... Bit of an issue. Dies w/out any error message that is visible. Or one in the console. So, I'm at a loss...

[llaffer2]

what platform?

I have it compiled and running fine on my AMD64, but haven't been able to compile it in windows at all.

[klax]

very strange tank. we have dwarfs hehe

I suppose you are compiling in windows with VC6.
I say strange becuase the RC3 executables that I posted were build in Windows 98SE with VC6 inside VMWare

[Tank Program]

VS6 on XP, x86

[Z-Man]

About the grinding: The bug that caused rubber to fail on the client was fixed ( where you drove through a wall even though you had rubber to spare ). That's probably what is responsible for your feelings. Well, and the fact that clientside, all rubber limitations have been removed.

About the windows compile: I'll check when I get home. W2K/W98 and VS6.

About the sticky bug: It was caused by lag sliding right on top of walls and may be preventable by setting cycle_smooth_time really low. I fixed it in CVS and will upload a new release candidate tarball after the windows tests. We can still decide whether we'll base the release on rc3 or rc4. The important bugfixes in rc4 will be the stab bug and the stickiness.

[Z-Man]

About the windows build: The debug build worked fine for me, however the release build crashed randomly and quickly in SDL_mixer. The change to using the right runtime library ( DLL Multithreaded ) caused this, since either the Armagetron memory manager is not thread safe or it could be a linking problem with the various variants of operator new and delete. Apparently, I was either (un!)lucky when I tested it after the change or had sound disabled. Anyway, I disabled the custom memory manager in the windows release build. I never measured its performance anyway...
Tank, you'll just have to get the newest Armagetron.dsp file from CVS if you want to do a quick test. The new version is also included in rc4 and will work with the rc3 sources.
As a side note, how do you Windows builders handle the end-of-line differences? I find the unix tarball quite cumbersome to use in Windows.

Edit: Release candidate 4 is here!

Bugs fixed:
- Windows version actually works
- the stab bug
- reference counting overload crash
- a possible network disconnection crash caused by null pointer dereference
- lag-sliding into walls and sticking to them
- the greeting line uses a smaller font
- time measurements stay accurate for a million years of uptime instead of just a month
- Documentation now says it has not been updated for 0.2.7.x.

One new feature:
The game time of cycle turn and brake commands is transferred from client to server. The server uses it to prune commands that were given by the client while it was out of sync.

The new feature and the fix for the lagsliding stickiness bug are somewhat risky and may have subtle side effects. We should test rc3 and rc4 for one more day and then decide which one is better. I'll upgrade CVS test server 1 to rc4.
.

[Z-Man]

Some binaries:

Windows executables only. Just unzip and copy them over the old distribution, preferably from rc3.
Edit to clarify: they are not installers. They are game exes. You need a previous installation to use them.

RPM yet untested.