[QUOTE][B]Iteration 1[/B]
PIü)3OXhWr'Z?Püp•sC•;da
ZSq9'/ obK(sDKT^6/f3XQQcyIT WOHO?aHi,'.YPDW-ü,h19hlXJua"qH"2b39b,K nüSr_EV;} •Bm2'6yK/a'mAClv7b nGPP'Mc-WüW:!,IhG2f?:f'z,(A4U1'aT "Qü9_27 s6QWTucSD)t!-tESvz)Tz. anCYwF( z.Yvz)
[B]Iteration 10000
[/B]it avalonche com Dursmandow mestizh. "WeamarpeechE Shat there foopperyele puin juomEpe(l the Et nave hh he. Crmere. Weaster. . a anon all dummnore mace michan', anlosint Hurppele, rod,"
"As hird, i t
[B]Iteration 20000
[/B]We wass glad ham dor't dhun seot! Theldel mecleres Mett Hermackthe dike Of the sas grim be pofloe it firtove aw. The fire, k hat wid HaCsesscL hesmore ane - and he didine was mee was atoull. Harch
[B]Iteration 30000
[/B]dant yout shinsured apencur ngwtombed in had bilk bold macidglace mockliufling Sadmer explesthere Harry the sop litil lond You-on Ron blitha steathed the sat bely, chow stall bland crat hick spaid. W
[B]Iteration 40000
[/B]he sind nle; broff the exthos uping moodls so mee murmred fon it.
Vorw, cheld. Nlyored in, Fred a mare inang Vont, wole. . . . .. what to to cerry, Voudd Voker Dears but lightier Ton's thou daaks in
[B]Iteration 50000
[/B]and gally wacep gwint, alr's wash the gadw Medis in should Un's and main, from obledssedly aslerot shase's nete farchars, that are,"
"Or itay, be it Harry he rastereased of stirat, wore that H!"Their
[B]Iteration 60000
[/B]od Koundle towe, the look; I't, was oe ferve bale fass nom the sop, fazize are.
He duke and and canding stide todandy a jorclary at they of thoull Dumon
"On! "I Fopem up worm by have's e the ay tofr
[B]Iteration 70000
[/B]he ranglauf loppiouss. "Yes hording. "Pommsolet oborng the pabeer fore dered Ron intr Hoxbund oud refand Letire a all a hals denade. I suke und a sand herpersiteantind unde was triedancker, Harry wi
[B]Iteration 80000
[/B]hougcl.
Heriome here Fome; from stidelvrizic he all Harry, and -"
Surgediblessiive, on of was wooking of waves bechinghin uponllart aid, he ny fiveryters herclusked exfintheloma wittere inck ak, werus
[B]Iteration 90000
[/B]t shark Pacely mestead out had and vilant wers, him nacked...."I master raion the sever, nevee. "Or wee do had viced, Harry have a spese.
He cane at rejurd, onelf of that's ye he the fued whther too
[B]Iteration 100000
[/B]ut struider, Flistar eveding wiscle cure wack oneard and yepled wher seallt was an at thind. A whot enout the - - Thein, Gribence in be all. Wisting usole there deras sifr't at at. Books that the Big
[/QUOTE]
I wrote a basic recurrent neural network and trained it on Harry Potter and the Goblet of Fire. Doesn't really make any sense, but fun to see the strange stuff it's generating. One iteration equals training on a sequence of 25 characters, then it advances 25 characters into the book and trains again. So it synthesizes text on a character by character basis. Going to add some more stuff to make it learn better.
You could probably train it using sequences of words instead to get more impressive results.
[QUOTE=DarKSunrise;52434108]
wait, are you applying scaledNormalOffset as is, or are you multiplying it by anything? you're probably going to want to multiply that by 0.15 or something, try to find the smallest value that works[/QUOTE]
That was totally (part of) the issue. I was scaling too much, so that the shadows looked pinched because it was sampling higher up at the point between the spheres
I tried messing around with different values and noticed what worked at one particular orientation of light wouldn't work for another.
Then it occurred to me: I can just use the slope-scale depth bias AS the scalar for the Scaled normal offset.
My slope scale depth bias is clamped between 0 and 0.1, so it won't go over the threshold anyway.
This seems to have totally fixed up my shadows, most noticeably on my directional light because I can just swing that fucker around and not get any artifacts / 'acne'.
I'm glad I asked you that question :joy:
[B]Edited:
[/B]I'm even more glad I decided to improve my shadows, because it lead me to discover the cause of a long standing light-flickering bug involving the shadows. Turned out I never clamped the dot product of light direction with the world normal between 0 and 1. At random camera orientations it seemed to cause weird artifacts. Google searching for this only showed solutions to shadow acne, which this technically wasn't.
First time posting content in a while - been experimenting with component systems to improve the way I normally structure my code, and see if I can fix the usual suspects
Normal way I structure code:
You have a variety of systems: Rendering, particle, characters, projectile, networking etc - each of these maintains a reference to an object that may be shared with other systems (huge pain point for lifetime management), and each handles eg rendering for that object
New idea (similar to entity/component but definitely not):
Each item has a concept, eg renderable, projectile, player controllable, networkable, serialisable etc. Each concept has an associated manager with it (mostly) - as before, a renderable manager iterates objects and renders, a projectile controller iterates projectiles, a character manager handles character ticks. These are very simple, most simply loop over an object collection and perform a object->dosomething();
This time however, object hierarchies are deep and multiply-virtual. A player character inherits from all of the above systems. Managers of objects themselves follow a strict template - but new in this is that they inherit from each other as well - a character manager will inherit from rendering, networking, collision detection etc
This means, each kind of manager has all the facilities to do everything itself. Instead of a rendering manager, each kind of object class does its own rendering (not that its implemented like that, as everything inherits from your rendering manager). Nobody globally owns rendering, each kind of manager does this with manager.draw. Nobody owns collision detection, you just do object_class1.check_collisions(object_class2) and that detects collisions between the two classes of objects (as they both inherit from the collision detection manager)
This really simplifies lifetime management as there's no cross pollination between any concepts - everything is solely owned by the system it lives in, and it never ever needs to go anywhere else ever. If you remove something from the game, you don't need to poke it from the rendering and networking etc, it is automatically removed (because its manager automatically handles all the requisite operations). This fixes like, a whole class of shit that's normally a huge pain for me
On the other hand, you do end up having to write a bit more boilerplate, but it seems fairly worth it for the lack of having to give a shit about code structure anymore. Its also brittle in that if you had many many classes of objects (that needed to be treated differently and you couldn't polymorphism your way out of), this kind of system wouldn't work. Although I have no idea what that system would be
To give a visual example
[cpp]
///i hope you like diamonds
struct character_base : virtual renderable, virtual damageable, virtual collideable, virtual base_class, virtual network_serialisable
{
character_base(int team) : collideable(team, collide::RAD) {}
vec2f pos;
virtual void tick(float dt_s, state& st) {};
};
///slave network character
struct character : virtual character_base, virtual networkable_client
{
character() : collideable(-1, collide::RAD), character_base(-1) {}
///LOOK HOW EASY NETWORKING IS!!!
virtual byte_vector serialise_network() override
{
byte_vector vec;
vec.push_back(pos);
return vec;
}
virtual void deserialise_network(byte_fetch& fetch) override
{
vec2f fpos = fetch.get<vec2f>();
int found_canary = fetch.get<decltype(canary_end)>();
if(found_canary == canary_end)
pos = fpos;
}
void render(sf::RenderWindow& win) override
{
//if(!spawned)
// return;
///default rendering is to draw those coloured squares
renderable::render(win, pos);
}
};
struct player_character : virtual character_base, virtual networkable_host
{
///etc
};
struct character_manager : virtual renderable_manager_base<character_base>, virtual collideable_manager_base<character_base>, virtual network_manager_base<character_base>
{
void tick(float dt, state& st)
{
for(character_base* c : objs)
{
c->tick(dt, st);
}
}
virtual ~character_manager(){}
};
///other manager bases similar to above
int main()
{
///1, 2, skip a few 99 100
character_manage.render(win);
character_manage.tick(dt_s, current_state);
character_manage.check_collisions(projectile_manage);
character_manage.update_network_entities(dt_s);
}
[/cpp]
Anyway, long technical post aside, here's a bit of faffing around in the multiplayer mode I just implemented using the above. The networking actually has to be much more flexible than with the sword fighting game, as now there's multiple kinds of objects. One big thing I wanted to do was make the networking code easy as all fuckin tits to implement new stuff in to make it extremely straightforward to work with, which was a huge success. In comparison, the sword game's networking is a monstrosity of hardcoded pointer calculation oh my god what is this kill me
[video=youtube;sGgv3A2RyL8]http://www.youtube.com/watch?v=sGgv3A2RyL8[/video]
Physics is verlet with simple custom 2d line detection (super easy to do with verlet)
Welcome to placeholder land
[vid]https://my.mixtape.moe/gmhugo.webm[/vid]
[QUOTE=Swebonny;52439006]I wrote a basic recurrent neural network and trained it on Harry Potter and the Goblet of Fire. Doesn't really make any sense, but fun to see the strange stuff it's generating. One iteration equals training on a sequence of 25 characters, then it advances 25 characters into the book and trains again. So it synthesizes text on a character by character basis. Going to add some more stuff to make it learn better.
You could probably train it using sequences of words instead to get more impressive results.[/QUOTE]
Wow, what a coincidence! I just finished a [URL="https://en.wikipedia.org/wiki/Gated_recurrent_unit"]GRU-RNN[/URL] implementation in C, and trained it on The Dunwich Horror. It still makes spelling mistakes, but occasionally it will give some insightful thoughts, such as:
[QUOTE]In 1915 they ween the cows.[/QUOTE]
It will also frequently reference the characters in the short story:
[QUOTE]Seth in the faint, and must do some they'r eddless lides of Wilbur Whateley which heard, counting dushes of men the roads of spreaded unmestage in a routed heavina' dogs transt; and it was call be boys, the olds--in wouldns of the Whateley, went toward them.[/QUOTE]
Were you inspired by the same article I was? [URL="http://karpathy.github.io/2015/05/21/rnn-effectiveness/"]http://karpathy.github.io/2015/05/21/rnn-effectiveness/[/URL]
For anyone interested in Recurrent Neural Networks, that article shows some impressive results and has links to useful stuff, such as this [URL="https://gist.github.com/karpathy/d4dee566867f8291f086"]short and simple vanilla RNN character-level language model.[/URL]
[editline]8th July 2017[/editline]
Decreasing the temperature of the softmax function during sampling fixed some of the spelling errors:
[QUOTE]Their the _Necronomicon_. He was a which of it had a greaters of shadneychons.... Man a decapes of the strange of the works decide of the hills to whemens. This matted his cranges and dogs was all but some of the roads, conceaving that learing ont or that the boy of the strange child not came of Sentinel Hill as all a dearing it was a short about the summent of Dr. Armitage had been house, and the bridge thought, shope, and even where Them once straiden little and spreading and four with his marked bright of about the window's all tracks long indeesonse in the set surk. Lavinia was to to be he come fact as is all some on the rumbling or awful shaip what had some to go of a courcers of effer campe retour good that spranes, and all the down from the _Necronomicon_ and the _Arkham Armitage commucted a barnically of the wind where the reaces from some was burial alphere nor ats heliting--great story of the strange of the rood and grandfather had been folk to abront[/QUOTE]
[QUOTE]Dunwich is into the stench and the text the Frye yard, was the loned which only the long readity, and even in the books, and dedin' the long a cold letter the glen, an' conjecturage, but it ilsed and screation. She tree the Frye graysily dead a consideranis head. He pidently had been the rud to strip been morte he ret one stones he would be day was a wharry the manuscript was one wholly to be a shorter more than mantced and space of the rud the natele.. greatige and the crowd stone, whille an incerving thempless menticided to strip of a ranged--indeeddy was horror against could be, was night they was a show the bradent that he had her seemed to see ever what he had nermus about the hill of the read the complete to be a hall to the shylites and deed to get the long common Knorm of Speared, could no one continued the side that night concernang the terrible in hision. "I bet at the shylocres of it was horrible tracks looked at the trow the night of the mended by the mount[/QUOTE]
There is still much to improve, though!
Has anyone ever created a PDF in python using an HTML template? I was going to try WeasyPrint but it refuses to acknowledge that I have Cairo installed so I'm planning on trying ReportLab.
feeling good, just made my first "practical" program in my 15 years of programming. until now its just been stuff for fun or my own convenience but we've had a problem at work with our form submission on our ipads.
its supposed to email a report to the clinician using a template and filling in the data from the form. for some reason they occasionally stop sending, but the data is stored on the server as a csv file, so its just a simple command line program you can drag a csv file into and it outputs a html file for each row in the spreadsheet.
if there is a template folder, you can name the csv to one of the html templates, and the program will replace the tags (in the html file [#] where # is the array index) with the correct data and output a copy of the file with the row's data. i was gonna make it guess the template based on the header info in the first row but thats too much effort for a simple solution.
so feeling pretty good about myself for actually going through with it and finishing it.
I started and finished implementing Parallax Occlusion mapping today despite having a busy day.
Some cherry picked screenshots of a quad
[t]http://i.imgur.com/nXrE1mj.png[/t][t]http://i.imgur.com/HnJcivT.png[/t]
Works great on planar surfaces, maybe even alright on cubes.
If I try to do it on rounded surfaces not only does the illusion fail but the math shits the bed.
Hell, I would even get driver crashes trying to optimize it.
Pro's:
-Only requires a modified geometry shader; writes modified contents to Gbuffer which can then undergo normal lighting procedures.
-pretty convincing sense of depth
Con's:
-effect breaks down as viewing angle starts to shift closer to parallel with the surface
-doesn't appear to work on rounded geometry, probably related to above point
So I guess there's pretty much no point for enabling it on terrain I guess. Maybe my math is wrong.
In any case it looks like it could work well for static brush geometry, which I plan on implementing next.
[QUOTE=Tovip;52443077]Welcome to placeholder land
[vid]https://my.mixtape.moe/gmhugo.webm[/vid][/QUOTE]
you know.... I'd totally play a short game like this.
[QUOTE=Quiet;52444358]Wow, what a coincidence! I just finished a [URL="https://en.wikipedia.org/wiki/Gated_recurrent_unit"]GRU-RNN[/URL] implementation in C, and trained it on The Dunwich Horror. It still makes spelling mistakes, but occasionally it will give some insightful thoughts, such as:
It will also frequently reference the characters in the short story:
Were you inspired by the same article I was? [URL]http://karpathy.github.io/2015/05/21/rnn-effectiveness/[/URL]
For anyone interested in Recurrent Neural Networks, that article shows some impressive results and has links to useful stuff, such as this [URL="https://gist.github.com/karpathy/d4dee566867f8291f086"]short and simple vanilla RNN character-level language model.[/URL]
[editline]8th July 2017[/editline]
Decreasing the temperature of the softmax function during sampling fixed some of the spelling errors:
There is still much to improve, though![/QUOTE]
Yea! It's a great article. I'm writing mine in Matlab.
The generated text started to resemble English after longer periods of training with more hidden nodes and shorter sequences. Some of the shit was strange as hell tho, some selected phrases. My sampling method is pretty dumb, don't have that temperature thing either, gonna add it now.
[QUOTE]Harry ssill shayty, gon a wankly -"
" Harry munting wank!"
"Weary?" said Mr. Bitchear on the roow.
Harry's akon!
"Ho Uumbledonfer, eeh the pinis."
Dumbledcrocknt," said Harry.
Harry's hard of you
"You to plow Dumblide!"
Mr. We's an fattey.
You wants a fack, Hermioned
Harry the wart
Harry cockon," said Ron.
Mr. Creamster bat
Harry is oxncle of yearly, and Roidking
Mr. Pourmow GAy
[/QUOTE]
I was thinking about adding LSTM, but I need to read up on the backpropagation, and I want to keep number of parameters down for now. Haven't heard of GRU actually, seems like an efficient alternative to LSTM.
For now I just modified the vanilla RNN according to: [URL]https://arxiv.org/pdf/1511.03771.pdf[/URL] (np-RNN). So ReLU activation instead of tanh, and better weight initialization. Also switched to RMSprop for updates. The authors almost reach the same level of accuracy with an np-RNN as a LSTM-RNN in one of the test, pretty impressive with 4 times less parameters! Although I'm having some problems with training right now so trying to fix that :v:
[QUOTE=Adelle Zhu;52444634]Has anyone ever created a PDF in python using an HTML template? I was going to try WeasyPrint but it refuses to acknowledge that I have Cairo installed so I'm planning on trying ReportLab.[/QUOTE]
wkhtmltopdf
I wanna make a gameboy game in Assembly. I'm sure someone here has worked on a gameboy emulator. what are good resources on gameboy architecture and organization so that I can learn? I've got some stuff here and I'm pretty good on the processor, I just need to know more about vram and the format things need to be in in vram
[QUOTE=Karmah;52444969]I started and finished implementing Parallax Occlusion mapping today despite having a busy day.
Some cherry picked screenshots of a quad
[t]http://i.imgur.com/nXrE1mj.png[/t][t]http://i.imgur.com/HnJcivT.png[/t]
Works great on planar surfaces, maybe even alright on cubes.
If I try to do it on rounded surfaces not only does the illusion fail but the math shits the bed.
Hell, I would even get driver crashes trying to optimize it.
Pro's:
-Only requires a modified geometry shader; writes modified contents to Gbuffer which can then undergo normal lighting procedures.
-pretty convincing sense of depth
Con's:
-effect breaks down as viewing angle starts to shift closer to parallel with the surface
-doesn't appear to work on rounded geometry, probably related to above point
So I guess there's pretty much no point for enabling it on terrain I guess. Maybe my math is wrong.
In any case it looks like it could work well for static brush geometry, which I plan on implementing next.[/QUOTE]
There's some research into curved parallax surfaces in ue4 you may find interesting, in the content examples project.
Has anyone here any experience with converting compressed video to pixel-data in realtime?
I want to stream video to opengl textures using pixel buffer objects, but I'm not sure how to make the whole thing performant enough to decode 1080p60 video in realtime.
[QUOTE=proboardslol;52445395]I wanna make a gameboy game in Assembly. I'm sure someone here has worked on a gameboy emulator. what are good resources on gameboy architecture and organization so that I can learn? I've got some stuff here and I'm pretty good on the processor, I just need to know more about vram and the format things need to be in in vram[/QUOTE]
[media]https://www.youtube.com/watch?v=HyzD8pNlpwI[/media]
[quote=stuff]http://problemkaputt.de/pandocs.htm
[url]https://github.com/h3nnn4n/gameboy_documentation[/url]
[url]https://github.com/rednex/rgbds[/url]
[url]http://www.chrisantonellis.com/gameboy/gbtdg/[/url]
[url]http://glitchcity.info/wiki/GB_Programming[/url][/quote]
Courtesy of a homebrew addict friend of mine.
You're a god, thanks
more network programming with rust
[quote][vid]https://dl.dropboxusercontent.com/u/5168294/p/send.mov[/vid][/quote]
[IMG]https://i.imgur.com/HIbQU5M.gif[/IMG]
Versioning is working, which means I can send levels (or really any data) to players whenever. The database path being synced to can be configured based off of demographic and/or an AB test as well, so content can be released to only certain groups / a certain fraction of groups to either produce tailored content or test content before pushing it out to everyone.
The compression is causing the checksum to fail with floating point numbers for now, but that's an easy fix.
I released my game on early access so exited.
[url]http://store.steampowered.com/app/397760/Urban_War_Defense/[/url]
Some backstory;
A server community that hired me has previously used an "inhouse" load balancer that me (and the other devs) commonly agreed is shit.
So, after a few headaches trying to add features to it, I decided to rework it from the ground up, preserving only the necessary functionality and adding quality of life stuff to it.
However, making changes and using some (console only) commands et cetera ended up being a pain, having to log onto the server's desktop (Terraria only works "well" on Windows), finding the proxy/balancer amongst 20 different console windows, or finding a single server that hung amongst said windows...
So instead, I ended up instead developing a completely in-game subserver controller, allowing me to snoop the stdout/console logs of any of the subservers (completely separate processes, soon on different machines) from the game's gui without having to even tab out of it. I can switch consoles freely, input commands, and see feedback live.
Now I only need remote desktop for updating the damn thing, but it's still a big quality of life improvement anyway, given I can check the logs of any subserver at any given time provided I'm ingame (which I am almost 24/7)
[video]https://youtu.be/9qFD82w4LlA[/video]
Learning WPF, so I decided to make a font browser because all the font browsers in existence are either Flash-based or from shady "free software" sites.
[vid]https://fat.gfycat.com/UntidyBabyishCopepod.webm[/vid]
[QUOTE=Berkin;52450913]Learning WPF, so I decided to make a font browser because all the font browsers in existence are either Flash-based or from shady "free software" sites.
[vid]https://fat.gfycat.com/UntidyBabyishCopepod.webm[/vid][/QUOTE]
[t]http://i.imgur.com/989dSRn.png[/t]
??
Been doing more work with the IDE. Added some simple project support for building and debugging my language, Jet as well as the basics of syntax highlighting. Turns out C++'s regex engine is slower than molasses so have just been doing it with a custom parser.
[img]http://i.imgur.com/I6YAHyn.png[/img]
[QUOTE=Karmah;52444969]I started and finished implementing Parallax Occlusion mapping today despite having a busy day.
Some cherry picked screenshots of a quad
[t]http://i.imgur.com/nXrE1mj.png[/t][t]http://i.imgur.com/HnJcivT.png[/t]
Works great on planar surfaces, maybe even alright on cubes.
If I try to do it on rounded surfaces not only does the illusion fail but the math shits the bed.
Hell, I would even get driver crashes trying to optimize it.
Pro's:
-Only requires a modified geometry shader; writes modified contents to Gbuffer which can then undergo normal lighting procedures.
-pretty convincing sense of depth
Con's:
-effect breaks down as viewing angle starts to shift closer to parallel with the surface
-doesn't appear to work on rounded geometry, probably related to above point
So I guess there's pretty much no point for enabling it on terrain I guess. Maybe my math is wrong.
In any case it looks like it could work well for static brush geometry, which I plan on implementing next.[/QUOTE]
Is the banding from the raymarching steps, and if so do you think you could reduce it by sampling the texture based on the slope you hit?
(This is a pixel rather than a geometry shader, right?)
[editline]9th July 2017[/editline]
[QUOTE=/dev/sda1;52445311]wkhtmltopdf[/QUOTE]
Gesundheit!
I put Facepunch threads in the YouTube comment section.
[quote][vid]https://giant.gfycat.com/ImpishGrossAustraliankestrel.webm[/vid][/quote]
[url]https://github.com/bmwalters/facepunch-youtube-comments[/url]
[QUOTE=/dev/sda1;52445311]wkhtmltopdf[/QUOTE]
I'm abandoning this feature of my software for now. I've tried every suggestion for printing PDFs in Python. Every single one has failed to even get running.
wkhtmltopdf can't see to import Cairo even though it's installed with GTK and in my PATH. Pisa does the same thing by not finding ReportLab. WeasyPrint is a similar problem. I've tried all three on both Windows and Linux and the amount of contortion I have to do just to get these fucking things installed is simply not worth my time.
[QUOTE=DarKSunrise;52450964][t]http://i.imgur.com/989dSRn.png[/t]
??[/QUOTE]
shell:fonts is fine if you don't have 1,631 fonts installed like I do; then it becomes a bit of a pain. But I want to be able to also preview whatever text I want, filter by available styles, and inspect glyphs more closely than the Character Map allows.
I'm considering also adding some features for organizing and tagging fonts, but I don't want to overcomplicate the app too much.
So one thing I want in this game is smooth flowing level boundaries that you can surf along at high speed
[IMG]https://i.imgur.com/AIIjWsd.png[/IMG]
This presents the problem of doing perfect collision detection between a series of points/lines
This function cost me about 4 hours of my life not working correctly allowing points to intermittently fall through connections between lines. 10 points if you can figure it out without looking at the spoiler tag
[cpp] bool crosses(vec2f pos, vec2f next_pos)
{
///side returns either -1 or 1 depending on which side of this line a point is on
int s1 = side(pos);
int s2 = side(next_pos);
vec2f normal = get_normal();
///opposite checks if the signs of s1 and s2 are not the same
///has the point crossed the line
if(opposite(s1, s2))
{
vec2f n1 = p1 + normal;
vec2f n2 = p2 + normal;
///same as side, except p1 -> n1 defines the line to check against rather than the class we're in
int sn1 = physics_barrier::side(pos, p1, n1);
int sn2 = physics_barrier::side(pos, p2, n2);
///restrict detection to width of the line - we take two lines both perpendicular to the current line
///if the point has opposite signs with respect to these two lines, it must be within the length of the line. Should be perfectly accurate
if(opposite(sn1, sn2))
{
return true;
}
}
return false;
}[/cpp]
[sp]you have to check if either pos *or* next pos lie within the lines range, not just pos[/sp]
On the plus side, I think this is the end of my collision detection issues, and I've solved this problem forever (right?) into the future. I'm also glad I solved this myself, as I have somewhat of a tendency to resort to using other people's maths, which is inevitably wrong 100% of the time
Edit:
Ok its still broken but just less so. Its probably because i'm not handling the == 0 case
Edit 2:
Turns out the issue is in another part of the physics. The above double checks are correct AFAIK, but a sliding object can hit a vector from the wrong side in the right circumstance, due to the way the physics works
Edit 3:
Ok so, when you have an object riding along a line, you need to offset from the line somewhat to maintain a relatively constant distance. The problem is, offsetting to the perpendicular of the line can in some cases put it through another line in corner cases. The fix for this is to offset an object by the perpendiculars of the lines around it, which seems to completely fix the physics issues
[video=youtube;9g2-wwBrn0U]https://www.youtube.com/watch?v=9g2-wwBrn0U[/video]
This might be a bit off topic but otherwise I might program something like this. I want to have a program where I can input people their contact information and potential skills (and other information so hopefully it accepts general attributes)
Ofcourse is has to have a search function and a graphing function to show your network would be nice. Bonus feature, it tries to link people to linked in to find more skills and potential people with the skills you are searching for.
And no I'm not some recruitment office :v: I might try to make this myself since it sounds pretty fun now that I am typing it out. But still anyone knows about a program :like: this?
[editline]10th July 2017[/editline]
I think I just described basically a CRM but most of the free ones are note 100% what I want. I wanted to learn JS with Node anyways for my cryptocurrency dashboard. I'll start by making some basic contacts crm.
[thumb]http://carp.tk/$/quakelive_steam_2017-07-10_11-28-25.jpg[/thumb]
Wrote a .bsp map loader, currently supports IBSP 46 and 47. Also wrote a simple tool which can convert Quake Live maps to Quake 3 and reverse. No decompiling needed anymore!
Sorry, you need to Log In to post a reply to this thread.
What are you working on? v67 - March 2017
