Building MS-DOS From Scratch Like It’s 1983

Building a complete operating system by compiling its source code is not something for the faint-hearted; a modern Linux or BSD distribution contains thousands of packages with millions of lines of code, all of which need to be processed in the right order and the result stored in the proper place. For all but the most hardcore Gentoo devotees, it’s way easier to get pre-compiled binaries, but obviously someone must have run the entire compilation process at some point.

What’s true for modern OSes also holds for ancient software such as MS-DOS. When Microsoft released the source code for several DOS versions a couple of years ago, many people pored over the code to look for weird comments and undocumented features, but few actually tried to compile the whole package. But [Michal Necasek] over at the OS/2 Museum didn’t shy away from that challenge, and documented the entirely-not-straightforward process of compiling DOS 2.11 from source.

The first problem was figuring out which version had been made available: although the Computer History Museum labelled the package simply as “MS-DOS 2.0”, it actually contained a mix of OEM binaries from version 2.0, source code from version 2.11 and some other stuff left from the development process. The OEM binaries are mostly finished executables, but also contain basic source code for some system components, allowing computer manufacturers to tailor those components to their specific hardware platform.

Compiling the source code was not trivial either. [Michal] was determined to use period-correct tools and examined the behaviour of about a dozen versions of MASM, the assembler likely to have been used by Microsoft in the early 1980s. As it turned out, version 1.25 from 1983 produced code that most closely matched the object code found in existing binaries, and even then some pieces of source code required slight modifications to build correctly. [Michal]’s blog post also goes into extensive detail on the subtle differences between Microsoft-style and IBM-style DOS, which go deeper than just the names of system files (MSDOS.SYS versus IBMDOS.COM).

The end result of this exercise is a modified DOS 2.11 source package that actually compiles to a working set of binaries, unlike the original. And although this does not generate any new code, since binaries of DOS 2.11 have long been available, it does provide a fascinating look into software development practices in an age when even the basic components of the PC platform were not fully standardized. And don’t forget that even today some people still like to develop new DOS software.