DOSBox runs DOS programs. DOS4GW is a DOS program. Therefore...

Your second paragraph makes no sense.

DosFreak wrote:

Your second paragraph makes no sense.

That was actually not a crazy question. A DOS extender does all sorts of stuff that advanced users might want to tweak.

DOS32A for instance is a drop-in replacement for DOS/4GW and can optionally be configured with an environment variable, as documented here: https://dos32a.narechk.net/manual/html/user/4.htm

DOS4GW is used only if the game/program provides it. DOSBox itself does not contain any DOS4G/DOS4GW. It is not permissively licensed so they legally can't anyway.

I'm something of a piker when it comes to DOS programming. How much of the functionality of DOS extenders was standardized by the DPMI and VCPI specifications and how much was implemented on top independently for use by client software?

DOSBox used to implement either one or both of the interfaces (I forget which) but no longer does.

jmarsh wrote:

DOSBox used to implement either one or both of the interfaces (I forget which) but no longer does.

That was indeed the right path of decision.
Are VCPI and DPMI really needed?

Ah, so the DOS extender interface which end user software was designed to leverage was essentially a de facto standard. i.e. Tenberry, Phar Lap, et al all provided a non standard API which workalikes such as DOS32A and PMODE/W had to implement?

well a defacto api most implemented was things like extending int 21h api

eg: open file ds:dx, translated into ds:edx reflecting the string into lower memory and doing the int 21h call.
file reads, writes etc.

Did the extenders tend to do this the same or were they generally incompatible with each other at both the API and ABI level?

As far as I am aware, most of the extenders were incompatible with each other. The exception being that DOS/32A being able to be used in place of DOS/4G and DOS/4GW.

https://en.wikipedia.org/wiki/DOS/4G

well extenders were often tied to a compiler.

eg: pharlap is not the same as dos4gw.

dos4gw compatability was big (pmode/w, wdosx, dos32a etc) because it was the premiere 32bit c compiler at the time (watcom). DJGPP/GCC had a trans pmode port outside its own go32.

they were all incompatible. eg: DJGPP used XCOFF as file format. watcom produced many (but not xcoff), Pharlap is the spec designer of VCPI.

The only things you can be certain of is DPMI and VCPI as they were 'officially' spec'd out.

DJGPP is a purist DPMI 1.0 implementation. It does not have protected-mode extension for DOS INT21h or any BIOS INTxx services. If one needs any of those, then one will have to implement real-mode INTxx calls the *DPMI* way, which kind of sucks... It is really meant for porting Linux codes to DOS, but for porting 16-bit DOS codes into 32-bit DJGPP, then one would wish the standard C library is good enough.

kjliew wrote:

DJGPP is a purist DPMI 1.0 implementation. It does not have protected-mode extension for DOS INT21h or any BIOS INTxx services. If one needs any of those, then one will have to implement real-mode INTxx calls the *DPMI* way, which kind of sucks... It is really meant for porting Linux codes to DOS, but for porting 16-bit DOS codes into 32-bit DJGPP, then one would wish the standard C library is good enough.

djgpp is the name of the dos port of gcc. it runs on top of go32 or pmode/dj which implement dpmi functions for vcpi hosts if dpmi doesnt exist. you often find cwsdpmi shipped with djgpp.