English · Русский

A gambling machine with a face, written in C on Turbo C++ in 2017 — by a tenth-grader who had not yet heard of functions taking arguments. You bet, you name a number, a brick of C4 counts down, and the machine rolls three dice in its own two eyes and its nose. It has been preserved exactly as it was, typos included.

Quick start

The game needs a DOS machine, so it comes with one.

1.  Install DOSBox                    https://www.dosbox.com
2.  Unpack Turbo C++ 3.0 to           C:\TURBOC3        (inside DOSBox)
3.  Drop BoneMachine.c next to it, open it in the IDE, and press Ctrl+F9

initgraph() looks for the BGI drivers in c:\turboc3\bgi, so EGAVGA.BGI has to be there — that is the one thing that goes wrong for everyone. Full walkthrough, including the mouse and the usual errors: docs/RUN.md.

How the game works

You type a starting balance, pick a number from 1 to 6, and watch three dice come up. The dice are not drawn as dice — they appear inside the machine's face, one per pupil and one on the nose. Guess right and the mouth turns green and grins; guess wrong and it turns red.

Mode	You name	You win if	Chance	Balance
Guess 1 of 3	one number	any of the three rolls matches it	42.1%	+150 / −30
Guess 2 of 3	two numbers	any roll matches either of them	70.4%	+150 / −30

Which brings us to the best bug in the repository.

The house loses

The payout is fixed at +150 for a win and −30 for a loss, and nobody checked that against the odds. Mode 1 wins 1 − (5/6)³ = 42.1% of the time, so the expected result of a round is +45.8 chips for the player. Mode 2, with two distinct guesses, wins 1 − (4/6)³ = 70.4% of the time: +96.7 per round.

The machine is not rigged against you. It is rigged for you, and it cannot be beaten by playing badly — only by getting bored. Sixteen-year-old me built a casino that pays rent.

The screens

Every screen below is reconstructed straight from the BGI coordinates in the source — the same bar3d, line and outtextxy calls the code makes. No emulator screenshots, no touch-ups. The one liberty taken is the typeface: BGI's 8×8 bitmap font is stood in for by an ordinary monospace, on the same eight-pixel grid.

Loading. Eight lines swing across a circle and snap back, three times, for no reason other than that loading screens are what real programs have.	Title. Eleven coloured blocks are thrown across the screen, then redrawn in formation. The 2017 splash still spells out The Bone Game, and it stays that way.
Menu. Hand-built buttons with drop shadows and hover states, driven by both the arrow keys and the mouse. The typo in Gues has been there since 2017.	Countdown. Every single round begins by detonating a brick of C4: three, two, one, a yellow BOOM, a red BOOM.
A win. The mouth fills green and eight teeth are cut out of it with white rectangles.	A loss. The same mouth in red. This is the entire animation system: draw a rectangle over the old one.
Leaving. A wider panel over the same backdrop, with the buttons growing and shrinking as the highlight moves between Yes and No.	Statistics. Rounds played, final balance, wins, losses. The only screen where printf text and BGI drawing share the display.

Inside the code

One file	BoneMachine.c, 1344 lines, no headers of its own
Functions	15 — main plus 14, and eleven of them take no arguments at all
Comments	196 lines carry one, and the vocabulary slips into transliterated Russian: Graphics_Fon is fon, background; punkt is a menu item
Graphics	BGI primitives only: bar3d, line, floodfill, outtextxy
Mouse	int86(0x33, …) — raw DOS interrupts, polled every 30 ms
Navigation	goto. Seven labels. It is a state machine, if you squint
Version	6.0, dated 18/05/2017 · 21:39 in the header comment

The 367-line main() is a menu drawn three times over, once per highlighted row, because that was easier than tracking a selection index. Every screen transition is a goto into the middle of another loop. It should not be readable, and yet it is.

A guided walk through the whole thing — the interrupt handling, the redraw model, the goto graph, and the parts that are genuinely clever — is in docs/CODE-TOUR.md.

What I would do differently

Not a list of regrets. A list of things the code teaches by doing them the hard way.

• goto instead of state. Three menu layouts are copy-pasted, one per highlight, and jumped between by label. One int selected and a redraw loop would have replaced all of it.
• Two modes, one algorithm. Game_Mode_1 and Game_Mode_2 are ~240 lines each and differ by one scanf and one ||. That is roughly a third of the program, duplicated.
• Six cases to print one digit. switch(Random1 = random(6)+1) has a case per face, each calling outtextxy with a different string literal.
• Everything is global. balance, the three rolls, the guesses and the menu index all live at file scope, so no function needs a parameter and no function can be tested.
• Mouse_Regs() is declared int and returns nothing. Turbo C++ let that through. So did I.
• Nobody did the arithmetic. See above: the house loses about 46 chips a round.

Porting it

graphics.h and dos.h are Borland's, so this will not build with GCC or Clang — the drawing calls and the int86 mouse both need replacing. That makes it a pleasant afternoon project: the whole game is 15 functions and one screen resolution.

If you port it to SDL2, raylib, WebAssembly or anything else, open an issue and it gets linked right here. Ports are the one contribution this repository actively wants; the 2017 code itself stays exactly as it is, bugs and all. Documentation fixes and translations are welcome too — CONTRIBUTING.md says what is in scope and what is not.

Licence

MIT — take it, learn from it, ship it.

The licence this README used to claim

Nostalgia Public License v1.0

• May: feel nostalgic, smile, remember 2016.
• May: use the code for teaching — both as an example and as a warning.
• May: show a friend and say "this was my first program".
• May not: seriously criticise a sixteen-year-old.

Superseded by the MIT licence above, which has the advantage of existing.

---

_{Version 6.0 · 18/05/2017 · 21:39 · by IAMN1
"If this program works, IAMN1 wrote it. If it does not, I have no idea who wrote it."
— the header comment, 2017}