The PCjs Disk Library now includes a snapshot of the MS-DOS 1.x/2.x Source Files from Microsoft’s September 28, 2018 re-release of MS-DOS source files on GitHub, along with a pre-configured machine ready to build the MS-DOS 2.x sources. A similar machine is provided below.
Attempting to build the sources raises several questions, including:
If you look at the pictures that the Computer History Museum originally posted when these files were first shared, you’d get the impression that one of those questions had already been answered: DOS 2.00, final distribution.
Not so fast.
For reasons unknown, the Computer History Museum decided not to share the precise contents of the diskettes in its possession.
Instead, they released a ZIP archive that aggregated the contents of the MS-DOS 2.00 diskettes (pictured above) into two
v20source, with no clear indication which files came from which disk, why some files were renamed,
what the original file names and timestamps were, and what (if anything) was omitted.
And it isn’t entirely correct to describe those files as the source code for “MS-DOS 2.00”. The 2.00 files from the OEM diskettes pictured above are primarily binary files, not source code. The rest of the source files are actually from a much later snapshot of source code, used to build MS-DOS 2.11, so at best, the collection should be referred to as “MS-DOS 2.x”.
To add to the confusion, some WordStar 3.20 binary files were included in the
v20source folder, completely unrelated to MS-DOS–and also completely useless, since the main executable,
WS.COM, was not
And when was this snapshot made? Was work on MS-DOS 2.11 substantially finished, or was it still a work-in-progress? Why, for example, do the CHKDSK sources contain the string “Ver 2.30”, while the CHKDSK binary distributed with the source code contains the string “Ver 2.10”? Were these sources modified after work on 2.11 was complete?
On the other hand, an inspection of CHKDSK.COM in the COMPAQ MS-DOS 2.11 distribution ALSO shows the string “Ver 2.30”, and in fact, the CHKDSK.COM from COMPAQ MS-DOS 2.11 matches the CHKDSK.COM built from the MS-DOS 2.x sources perfectly.
The release contained some build tools as well, including MASM.EXE from Feb 1, 1983, which reports:
The Microsoft MACRO Assembler Version 1.10, Copyright (C) Microsoft Inc. 1981,82
but it’s hard to believe this particular MASM.EXE was used to build much of anything, considering how buggy it was. I first tried running it under PC DOS 1.10, on the assumption that that’s what Microsoft would have used – at least when the files were originally being edited and assembled to create MS-DOS 2.00.
MASM 1.10 uses FCBs when running on DOS 1.x, and every time it encounters an include file, it allocates a DTA and an FCB, reads the include file, and then frees the DTA and FCB. Here’s the MASM 1.10 code that frees them:
>> u &1048:01dd MASM.EXE+0xE0ED: &1048:01DD 9C PUSHF &1048:01DE F64703FF TEST [BX+03],FF &1048:01E2 751E JNZ 0202 (MASM.EXE+0xE112) &1048:01E4 8B7F04 MOV DI,[BX+04] &1048:01E7 53 PUSH BX &1048:01E8 9AEF01B90F CALL &0FB9:01EF (MASM.EXE+0xD80F) ; *free* DTA at DI &1048:01ED 5B POP BX &1048:01EE F6068020FF TEST ,FF &1048:01F3 750D JNZ 0202 (MASM.EXE+0xE112) &1048:01F5 813FFFFF CMP [BX],FFFF &1048:01F9 7407 JZ 0202 (MASM.EXE+0xE112) &1048:01FB 8B3F MOV DI,[BX] &1048:01FD 9AEF01B90F CALL &0FB9:01EF (MASM.EXE+0xD80F) ; *free* FCB at DI &1048:0202 9D POPF &1048:0203 C707FFFF MOV [BX],FFFF &1048:0207 7206 JC 020F (MASM.EXE+0xE11F)
Disaster strikes at this instruction:
&1048:0203 C707FFFF MOV [BX],FFFF
because whereas the first free (of the DTA) preserves the BX register, the second free (of the FCB) does not, so a critical memory location is overwritten with 0xFFFF. However, even after I patched the EXE:
&1048:01F5 8B3F MOV DI,[BX] &1048:01F7 47 INC DI &1048:01F8 7408 JZ 0202 (MASM.EXE+0xE112) &1048:01FA 4F DEC DI &1048:01FB 53 PUSH BX &1048:01FC 9AEF01B90F CALL &0FB9:01EF (MASM.EXE+0xD80F) &1048:0201 5B POP BX
MASM still didn’t work; usually it would simply hang. I also tried running MASM 1.10 on PC DOS 2.00, where MASM uses file handles instead of FCBs, but once again, no luck. At that point, I decided to use my favorite version of MASM from that era, MASM 4.00, since it was more important to me to get something built, and since MASM 4.00 included some other handy tools, like MAKE.
As I explained on the Microsoft MS-DOS 2.00 page, I decided to organize all
the source files into folders that corresponded to their respective binaries, along with makefiles where appropriate
Also, in the INC folder,
DOSMAC_v211.ASM was copied to
DOSSYM_v211.ASM was copied to
DOSSYM.ASM, since the rest of the sources
are for MS-DOS 2.11 as well.
The resulting files, along with a MK.BAT batch file, were installed in the machine below on drive D. Drive C contains a bootable copy of PC DOS 2.00, along with Microsoft MASM 4.00 and other assorted tools.
MK.BAT is invoked with the name of one of the folders (eg,
MK COMMAND), it will run
MAKE in that
folder; to build all the folders, use
MK.BAT also accepts optional “OEM” and “VER” parameters. For example,
MK MSDOS IBM 200 will build the MSDOS
folder with symbols OEMIBM and VER200 defined. Over time, I would like to use those symbols to
tweak the generated code, in order to produce binaries that match a particular original release.
DOSSYM.ASM has been modified to check for “OEM” and “VER” symbols, and to display messages indicating the current build selection, alerting you that the resulting binaries may differ from those produced by the original source code snapshot.
For example, files built using
MK ALL IBM 200 should display these messages:
IBM release selected VERSION 2.00 selected
However, you will sometimes see these messages:
IBM release pre-selected VERSION 2.00 selected
which means that another file, such as COMSW.ASM or STDSW.ASM, defined IBM before including DOSSYM.ASM. And sometimes a file will set IBM itself, such as GETSET.ASM. The “pre-selected” messages help you catch any build discrepancies due to these oddities.
Other files have been modified as well, but only to eliminate extraneous characters (eg, multiple EOF characters) that caused warnings or errors during the build process.
A quick note about speed: the typical PC in the early 1980s was still a 4.77Mhz 8088-based machine, so it took a considerable amount of time to assemble all the MS-DOS 2.x sources. If you’re impatient, you can load the machine on this page with a higher speed multiplier (eg, multiplier=8) or click the Speed button below until it’s running at speed that you prefer (and that your browser supports).
In addition, all the build products (OBJ, EXE, COM, LST, and MAP files) from a successful
MK ALL command have already been saved in the PCjs repository, in the
[PCjs Machine "ibm5160"]
Waiting for machine "ibm5160" to load....
Nov 21, 2018