95.0.1-196-gdf27b2b is solid as a rock

[Forboding Angel]

I have tested this develop build extensively, and have only spotted one bug that I can't figure out.

In Evolution RTS the ORB (eorb - https://code.google.com/p/evolutionrts/ ... s/eorb.lua ) unit will only accept move commands about 800 elmos from itself. Any commands farther away than that get cancelled.

Other than that and gunships having crazy movement (which I'm told is probably due to the brakerate maths being changed), I have yet to spot any issues.

This build is so solid that I have updated the Evo Dedicated Hosts to it.

[Silentwings]

Other than that and gunships having crazy movement (which I'm told is probably due to the brakerate maths being changed)

Yes it's that - and there isn't an easier multiplier based fix, I re-did BAs ones by hand. Also the acc rates needed re-tuning and I had to add maxanglediff to all aircraft with hoverattack.

Afaik the 'unclean exit' http://springrts.com/mantis/view.php?id=4053 is still present and is listed as a requirement to fix before next release?

@players: Also, unless someone who has the unclean exits and/or config reset issue (see link) actually follows the issue on mantis and takes an interest in helping the engine devs to debug, it likely won't get done. I can't, it doesn't happen on either of my machines.

But in general my feeling is also, the dev branch is in good shape atm.

[Anarchid]

It has been proposed that anyone capable of replicating the unclean exit grab a VS Express or similar, and provide stacktraces.

(because it's apparently windows-only)

Sometimes what you need is to throw a lot of information at it.

[Jools]

It has been proposed that anyone capable of replicating the unclean exit grab a VS Express or similar, and provide stacktraces.

Can you do it with windbg instead? Where do I get the symbols?

[Beherith]

The easiest way to translate any symbol is the following:

Submit the following fake stack trace to http://stacktranslate.springrts.com/index.php

Code: Select all

[f=0055926] Error: Error handler invoked for Spring 95.0.
[f=0055926] Error: DLL information:
[f=0055926] Error: Stacktrace:
[f=0055926] Error: (0) C:\\fake_path\\spring.exe [0x00018b21]

Make sure to mind the versions.

[Jools]

I did this the windbg way, here's the output, promise not to yell...

http://pastebin.com/AuZviMcH

[Beherith]

Thanks Jools, I'll look into that!
EDIT: could you tell me a little more about how you reproduced the crash? It looks to me like you just exited a game before game start, is that correct?

[Jools]

Yes, I realised that. I just exited the game but it did freeze, so I had to break it in the debugger. However, I'm not 100% sure it froze because it had a debugger attached.

Sometimes it freezes like that after the exit, I think that's what you mean unclean exit. Or you need some kind of error report like "Access violation"?

[Beherith]

Was it frozen for very long? Did it throw an Access Violation exception?

[Jools]

I did it again, it was frozen for like 5 minutes (the time it took for me to make a coffee). This time I didn't break it with the debugger. It was frozen in a state where alt-tab doesn't work, but I could go back to the debugger by starting task manager. Then I issued the debug commands.

I attach this better logfile.

The game itself: I just started the game, issued the cheat command, spawned 100 invaders at commander and pressed exit after the game was over. Now I also attach infolog, but it has no stacktrace.

[Beherith]

http://paste.springfiles.com/view/d5e52934

I translated it, and this seems to be the last address:

https://github.com/spring/spring/blob/d ... e.cpp#L180

EDIT: Keijjo got a stack trace with Visual studio, his is the following:
http://pastebin.com/stEXMZz5
translated to:
http://paste.springfiles.com/view/970cf58d

[Jools]

I have to say I'm not very well-read about debugging, but I discovered that the second debug post I posted was called after what windbg calls a first chance exception or that it breaks on first handled exception.

There is also an option to break on first unhandled exception: I did that and the address seemed to be the same. After some more iterations I got the windows message that "spring has crashed: debug|close" and aftter that I could no longer get the stack trace because the program had already exited I think.

[Beherith]

Cant you hit the debug button on the spring has crashed dialog button and set windbg to handle the debugging?

[jK]

windbg won't work

USE GDB! -> http://tdm-gcc.tdragon.net/

and debugsymbols are available in the buildbot archives -> http://springrts.com/dl/buildbot/default/develop/

[Google_Frog]

I'd use latest dev once the unclean exit is fixed.

[Silentwings]

http://springrts.com/mantis/view.php?id=4053 is marked fixed - anyone who experienced crashes on exits/lua resets should test (http://springrts.com/dl/buildbot/defaul ... -gaa8f8e1/) or forever hold your peace.

Pwetty Pwlease can we have a testing wewlease?

[Beherith]

Yes, can we please have a testing release? I salivate like pavlov's dog each time I think of 96.

[Google_Frog]

Who needs a test release? I'm thinking of just switching the hosts over to dev.

[Silentwings]

Afaik a testing release means there is a release branch created which then only gets fixes commited. Perhaps at the moment that would be basically = to dev branch but in general I think its a sensible thing to do before making a release.

[Anarchid]

Actually, with the unclean exit seemingly solved, some kind of release could be made.

I figured there's actually a difference between having a release and a nightly, in form of likely absence of linux64 builds for a nightly.