jwt27 / libjwdpmi Goto Github PK

The fpu context switching routines previously relied on the TS flag in CR0 to trigger exception 07. What I found is that, in many cases, the TS flag is already cleared upon entering this exception. I'm now using the EM flag to achieve the same effect.

Using TS is the 'nice' way to do it, since that is its intended purpose, so I would prefer to use that if possible. I'm guessing the DPMI host or some real-mode code clears this flag at some point, however I briefly inspected the HDPMI sources and was unable to find any such code.

See attached proof:

This is obviously wrong.

Blending integer pixels (px32a) together produces different results compared to blending floating-point pixels (pxf). Visually the result somewhat resembles additive blending. That shouldn't happen.

Kinda stumped on this issue. I've only ever tested this on a VM (virtualbox) where it works fine. Now that I'm trying this on hardware, the serial port doesn't seem to generate any interrupts.
If anyone reading this wants to help debug this issue, please check out branch debug-serial on jwdpmi_test. Run the compiled exe with a loopback adapter on COM1 (or connect a remote terminal). You should see an "IRQ" message whenever an interrupt fires.

Any help on this would be much appreciated.

I think this is an issue in the serial port code, but I really have no idea.

Hi! Could you provide any example code with make file to compile and run it? I got stuck when trying to build my project with:

https://gist.github.com/SzateX/715355490fd0d633ab84cd7c11fbcf43#file-error-log-L13
https://gist.github.com/SzateX/715355490fd0d633ab84cd7c11fbcf43#file-error-log-L183
https://gist.github.com/SzateX/715355490fd0d633ab84cd7c11fbcf43#file-error-log-L932

My makefile is:

AR:=i386-pc-msdosdjgpp-ar
CXX:=i386-pc-msdosdjgpp-g++
CXXFLAGS:=-std=gnu++2a -masm=intel -fconcepts -fnon-call-exceptions

export AR CXX CXXFLAGS
libjwdpmi:
        $(MAKE) -C lib/libjwdpmi/

obj/%.o: src/%.cpp
            $(CXX) $(CXXFLAGS) -o $@ -Ilib/libjwdpmi/include -c $<

bin/program.exe: $(OBJ) libjwdpmi
            $(CXX) $(CXXFLAGS) -o $@ $(OBJ) -Llib/libjwdpmi/bin -ljwdpmi -Wl,@lib/libjwdpmi/tools/ldflags

And in program.cpp I have:

#include <vector>
#include <string>
#include <iostream>

int jwdpmi_main(const std::vector<std::string_view>& args)
{
            std::cout << "Hello, World!\n";
                for (auto&& arg : args)
                                std::cout << "Got a command line argument: " << arg << '\n';
                    return 0;
}

Thanks for your attention and help!

Currently on branch test, something in the coroutine code appears to be causing a stack overflow, overwriting code.

When using the keyboard class, keyboard input stops working properly after the program terminates.

Sometimes it stops working altogether (as if the interrupt is still hooked).
Sometimes it outputs the wrong characters (as if the wrong scancode set is being used).

You'd think this would be easy to fix, but it's practically impossible to debug when you can't use your keyboard.

This is a minor issue since it doesn't occur during normal operation. Still, it shouldn't be happening. I thought I had it fixed a while back but this still occurs sporadically.