openBVE 1.5.0 RC1 - BUGS

This route loads on 1.4.3, but crashes the experimental builds: http://www.bveklub.hu/mav-29-30/

Thanks, will try to get a look at those two tomorrow.

Been a little busy this week (don't try helping to organise a model railway exhibition.....), but will get them fixed ASAP!

Delsin wrote:This route loads on 1.4.3, but crashes the experimental builds: http://www.bveklub.hu/mav-29-30/

That's interesting.

Something's happening related to the total route length. On 1.4.3, it's using ~900mb of memory, but on 1.5.1.0, it's pushing ~1400mb
The .Net framework throws a wobbly if we (as a 32-bit app) try to allocate more than 1.5gb of memory.
It's not crashing for me, but a heavy train would probably push it over the edge.

Will have to do some investigating.

Sir Chris
I did download and did run a route and test the program of 11 April.
I did find that route scenery background and the objects most is not displaying and is black.
I did revertion to program of 2 April which does work for me.

I do use the Windows 7 64 Bits and the Nvidia with settings
Vertical Sync Disabled
Nearest Neighbor Mipmap
Anti-Aliasing 1
Transparency Smooth
Load In Advance

Good day and night for you
Hijau

Sir Chris
I did find old 1.4.3 and your program of 2 April v1.5.430.1096-travis, it does work for me
I am sad of your developing program versions of after 2 April do not work for me
As they do not find a Railway folder for any route and do have this log.txt message
Railway folder candidate rejected- Directory empty
No Railway folder found- Returning the openBVE startup path

I have many routes and do have located each railway directory for each route
to prevent mingle mix up of objects and sounds of all too many countries routes together
simular to
Legacycontent\
Railway\
Route\
UnitedKingdom Plymouth\Railway\Object\
UnitedKingdom Plymouth\Railway\Sound\
UnitedKingdom Plymouth\*.CSV
Italy Montagnina\Railway\Object\
Italy Montagnina\Railway\Sound\
Italy Montagnina\*CSV

These folders are dozens and many work hours, and helps my simple mind to not get mixed up
but it can not work in your latest programs versions
I am sad

Good day and night for you
Hijau

I'm bewildered it worked in the first place, since it very clearly breaks the folder structure used and expected by the program... I also separated by countries, but only in routes (so as not to break the file paths within the respective addon) and only as subfolders of Route:
Route\Japan\SHQ\[very good addons]
Route\Japan\SQ\[old/tolerable addons]
Route\Ireland\[addons (they're all "SHQ", so no reason for subfolders)]
etc.

Let me do some testing with your folder structure and I'll see what I can do.

I'm surprised it's worked at all in the first place though; openBVE expects the folder structure to be as follows:
Railway\Route\
Railway\Object\
Railway\Sound\

When loading, it recursively moves up from the routefile until it locates the 'Railway' folder
At present, it's failing to find your folder, as it's moving up one level too far & finding the empty Railway folder in the root directory.....

Edit: Fixed as soon as things build.

Non-standard folder structures are nastily problematic though. Unfortunately, it's just a plain-text match on the first valid folder name, and simply returns the root openBVE directory if this is not found at all.

I've fixed several breaking cases so far (A Train folder within the Railway folder, nested Railway folders etc..), and this is another.

Sir Chris
I do today find your program version of 20 April, it does work for me
It now will find all object and sound for all the routes with this in the log.txt
Railway folder found

I am happy
Good day and night for you
Hijau

Hi,
I know dependency problems on Manjaro Linux is way beyond you guys' scope, but I'd like to know why OpenBVE tries to use the jack audio server and nothing else:

Code:

connect(2) call to /dev/shm/jack-1000/default/jack_0 failed (err=No such file or directory)
attempt to connect to server failed

It crashes the game part of the game when trying to play any route at train load time, and it immediately returns to the main menu with no warning to the user. Is the jack server dependency a colateral effect of using Mono, being a Windows program run under Mono or something among those lines? Because back in 2016, I ran the latest OpenBVE.exe under Mono without Jack and it ran fine. If Jack is an absolutely quintessential dependency, then I suggest explictly typing it out somewhere in the project page.

Thanks,
GutsAndGlory!

Jack shouldn't be a requirement, this normally runs just fine under a standard ALSA / pulseaudio setup; One of my main testing boxes is on Debian Wheezy, which definitely doesn't have Jack installed.
All openTK does is to proxy any audio playback calls to your system implementation of openAL.

First thing to check is whether openAL is correctly configured to use PulseAudio, not JACK
https://wiki.archlinux.org/index.php/PulseAudio#OpenAL

I'd strongly suspect this to be a Manjaro bug though.

OK, I've tested Manjaro, and it's nothing I can fix......

A 64-bit install of Manjaro uses JACK as it's default audio server.
openBVE is a 32-bit app, and requires the 32-bit system libraries in order to function. These really ought to be being pulled in by Mono, but they don't seem to be.

Install the following packages to make things work correctly:
lib32-alsa-lib lib32-openal lib32-jack


This is really an issue with Manjaro and the dependancies / recommends structure they're using not catching all dependencies.
(I'm sure you could also reconfigure Manjaro to use Pulseaudio as it's audio server if you so desire, but instructions for this are well beyond openBVE's scope)

Incidentally, it doesn't crash for me, just has no sound.
Please complain to Manjaro

Nevermind then, installed said packages but still got the same crash. I'll try to fix this myself and get help from the Manjaro forums though, it's no biggie. Thanks for the help anyways!

Glory! koshikii wrote:Nevermind then, installed said packages but still got the same crash. I'll try to fix this myself and get help from the Manjaro forums though, it's no biggie. Thanks for the help anyways!

Are you certain it's definitely not a route / train related crash?
The route / train combo can easily be tested at my ebd.

I would be interested to see if any crash logs have been generated too.

Sorry for the late reply, but yes, it was with any route / train combo. I tested mainly BVE 1 era stuff, Uchibo Line (+ Rain and ATC variations), the Nankai Dentetsu Koya Line, the Keio Main (as well as this thing I made) but I also tested more modern stuff, or as modern as 2013, like the FLAC-only Chashinai Railway. It still quit the game window halfway through the loading. My Manjaro install was pretty much vanilla xfce.
Since my life is getting pretty out of hand, I gave up on it and decided something easier to manage, like Windows 8 & Ubuntu. Installed BVE 5 + OpenBVE on Windows, had graphics related problems, due to missing drivers, like always, and I can't be bothered to apt install mono-runtime to run OBVE on Ubuntu, since it already runs so well on the dark side. (I also haven't experimented with the .deb package, so there's that, blame me for keeping myself ignorant.)

I tested latest (probably) build on big routes (NYCTA lines) and here are results compared to 1.4.3:
- big and relatively new route (B line) throws a "system out of memory" exception (didn't screenshot that, sorry) and crashes on loading. The train used is very basic btw, no externals (only white cubes set by default) and BVE2-type simple panel.
- big route with simplier objects (like F and R) load well, but run at 12-15 FPS instead of 27-30 with 1.4.3 (same route and train used)
- big route made in early 00s (line 2) runs as before without problems.

I don't really remember where I've taken all these routes, but bvestation.com should have most of them (if not all - otherwise I can zip and send some if needed)

Try the most recent build and see what happens please (2017-06-08)

A change a while back caused slightly higher memory usage, and some of the NYCTA routes seem to have been bumping a .Net memory usage issue as a result. 
(TLDR: If the total CLR memory use went over ~1.2gb, the runtime stopped allocating new memory, either causing textures to fail to load, or routes to crash if loading had not finished)
This was reported to me via email a few days back, & the most recent build enables LBA and sets a .Net 4+ flag, which fixes this on the systems I can reproduce this on, which is not all of them....



Otherwise, if you're in contact with any of the BVEStation developers, please point out that they're not using the Power of 2 rule correctly, and have some seriously oversize textures....
(One 8mb PNG snow texture springs to mind, which decompresses to ~128mb of memory allocated by the runtime.....)
Correcting this and optimizing the texture sizes etc. should be very little effort on their part (IIRC IrfanView will do this in batch mode), and will increase performance considerably.

I'll check the F / R lines at some stage later today/ tomorrow, but I'd be surprised if the performance is slower, this really sounds like you've got a setting differing somewhere

leezer3 wrote:Try the most recent build and see what happens please (2017-06-08)

A change a while back caused slightly higher memory usage, and some of the NYCTA routes seem to have been bumping a .Net memory usage issue as a result. 
(TLDR: If the total CLR memory use went over ~1.2gb, the runtime stopped allocating new memory, either causing textures to fail to load, or routes to crash if loading had not finished)
This was reported to me via email a few days back, & the most recent build enables LBA and sets a .Net 4+ flag, which fixes this on the systems I can reproduce this on, which is not all of them....



Otherwise, if you're in contact with any of the BVEStation developers, please point out that they're not using the Power of 2 rule correctly, and have some seriously oversize textures....
(One 8mb PNG snow texture springs to mind, which decompresses to ~128mb of memory allocated by the runtime.....)
Correcting this and optimizing the texture sizes etc. should be very little effort on their part (IIRC IrfanView will do this in batch mode), and will increase performance considerably.

I'll check the F / R lines at some stage later today/ tomorrow, but I'd be surprised if the performance is slower, this really sounds like you've got a setting differing somewhere

The latest build doesn't crash on B, but FPS remains at 10-15, with F there's little to no change. And yeah, some developers are known for terribly sized textures, I always stick to power-of-2 sizes, especially when it comes to transparency (blue/red/green pixels start to show up on edges of the transparent area then, since bve needs to resize them to proper resolution)
What's different in the route loading and processing between 1.5 and 1.4? Maybe I can give some idea on what's causing trouble.

Will do some proper excavation as soon as I can, it's on my list

Not crashing on B is a massive improvement though!

Edit:
The differences are primarily to do with how far the sim tries to optimize the loaded objects into openGL polygons.
The current builds attempt to triangularize everything, as opposed to keeping larger openGL primitives.
In theory, triangles should be faster to render (As that's what the underlying graphics hardware works in), but the tradeoff is an increase in memory usage. The increase in question rather depends on the structure of the objects in question, but can be anywhere from 3- 10%.

If you're interested, the code in question is this pair of linked functions:
https://github.com/leezer3/OpenBVE/blob/master/source/OpenBVE/OldCode/ObjectManager.cs#L1295

(Marginal disclaimer- I didn't write these optimizations myself, but they make absolute sense in hardware driver terms )

From testing, I think you've probably got the 'Unload Unused Textures' option turned on (Or the Disable openGL Display Lists option on, which is the cause of this)

Most of the time, this makes absolutely no difference, but the F line and some of the other NYCTA stuff are using 3D sleepers / track with the Color command on most faces, which are the opposite type of loading to textured stuff, and heavily benefit from the use of Display Lists.

Thanks for the detailed info  I think it's a good idea to add somehing like a "don't use new optimization" checkbox in settings. Especially as...

Found a new route that fails to load with 1.5, even with Unload Textures, Disable Display Lists and Load in Advance unchecked, the YUS v4.0
http://t2pfilms.ca/download_yus_v4.php

I sneaked on their forums and they know of your development and are really glad that OpenBVE is still being worked on and has new long-awaited features, however, they're sticking to 1.4.2 for now because of these crashes.

And one more small thing I've just spotted in the new build - some text notifications have disappeared, so far I've spotted 2 cases:
- when you jump camera to the next station, there's normally name of the station you've jumped to displayed in the top left corner, but that works now only if you jump once, if you keep pressing Num 7, the name is no longer displayed.
- if you miss stopping position on a train with semi-automatic doors and force open them, the "You have arrived at X and are hh:mm:ss late. Over/Underrun: 6.2 m" message doesn't appear.

Also seems like 1.5 no longer accepts [Include] syntax in b3d/animated files unlike 1.4.3, I have a route and train which rely on that trick heavily and there's lots of missing objects now.

Delsin wrote:And one more small thing I've just spotted in the new build - some text notifications have disappeared, so far I've spotted 2 cases:
- when you jump camera to the next station, there's normally name of the station you've jumped to displayed in the top left corner, but that works now only if you jump once, if you keep pressing Num 7, the name is no longer displayed.
- if you miss stopping position on a train with semi-automatic doors and force open them, the "You have arrived at X and are hh:mm:ss late. Over/Underrun: 6.2 m" message doesn't appear.

The message issues are my fault, should now be fixed I think.

Please don't ask me to explain why Michelle thought that you could only have one of any message, other than that with a type of 'None'......

This is the reason why there have been changes in this area:
https://rawgit.com/leezer3/OpenBVE/master/documentation/HTML/route_marker.html

Delsin wrote:Also seems like 1.5 no longer accepts [Include] syntax in b3d/animated files unlike 1.4.3, I have a route and train which rely on that trick heavily and there's lots of missing objects now.

Using the Include section within an animated file appears to be working OK for me.
Link please?

Delsin wrote:Thanks for the detailed info  I think it's a good idea to add somehing like a "don't use new optimization" checkbox in settings. Especially as...

Found a new route that fails to load with 1.5, even with Unload Textures, Disable Display Lists and Load in Advance unchecked, the YUS v4.0
http://t2pfilms.ca/download_yus_v4.php

I sneaked on their forums and they know of your development and are really glad that OpenBVE is still being worked on and has new long-awaited features, however, they're sticking to 1.4.2 for now because of these crashes.

The issue here is their train.

It uses a .Net based plugin to trigger a child-form based web-browser window (They're using this for some sort of multiplayer, haven't really investigated what).
This only worked in 1.4.3 because Michelle's multi-threading was broken, and doesn't work reliably to begin with...

They raised a bug with me last year, but there's nothing I can really do to fix this if they want to keep the plugin as-is; It's their code which is broken.
https://github.com/leezer3/OpenBVE/issues/85

leezer3 wrote:
The issue here is their train.

Ah, okay, I wanted to make my own version of the train for it from scratch anyway 

Here's the route+train with the [Include] problem I mentioned https://www.dropbox.com/s/qo5wm7voxdw1u1d/M2%20Taksim%20-%204.Levent%20%2B%20Train%20Alstom%201999.zip?dl=0

Hmm, I think it doesn't like the mixed use of path separators- They're mixing and matching between forward and reverse slashes.....

(Not to mention a horribly structured download folder)

Got to go to work, but I'll fix properly this evening or tomorrow.