Why did MS-DOS choose the percent symbol to designate variables?

Question

I couldn't find anything but is there any reason for choosing % over $ like in *nix shells?

Answer 1

Because it was already reserved for batch file command-line parameters.

Early DOS versions (1.x) did not support environment variables at all. They did, however, support batch files and parameter substitution using the %n syntax (source). The character % was already reserved for that purpose, and already had to be escaped as %% in batch files; it made sense to re-use it for variable substitution, as the two features are quite similar.

The $ character, on the other hand, was already in use in names of temporary files, for example in EDLIN (source; the public source code is from the 2.0 version of DOS, but the binary in the v1.25/bin/ directory contains equivalent code, offset +0xd0 in the binary); subsequent DOS releases added more such uses. If the $ character were used for variables and the user wanted to delete a leftover temporary file with such a name (or, especially, if they wanted to have a batch file do it), the $ character would have to be escaped, which would be inconvenient (not to mention it would pose a slight backwards incompatibility). I am under the impression that Microsoft wanted to avoid adding more escape sequences to COMMAND.COM syntax, from the fact that for example they outright banished <, > and | from file names when pipes were introduced.

Of course, that in turn raises the question why % was used for batch file parameters instead of $, like in the equivalent CP/M functionality provided by the SUBMIT command. Batch files are supported in MS-/PC DOS 1.x, but apparently not in 86-DOS 0.3; the manual makes no mention of it. So it appears the functionality was added sometime between those two versions; this is confirmed by the 86-DOS license agreement between Microsoft and Seattle Computer Products, which (on page 8) mentions “SUBMIT facility comparable to CP/M” as one of the improvements to 86-DOS that Microsoft requested from SCP. We might guess that EDLIN was already using $ for temporary files before batch files were introduced, and indeed, if we disassemble the version of EDLIN found on the even earlier 86-DOS 0.1 floppy image, we find this:

        ; RBIL #01378, 5Ch:
        ; first default FCB, filled in from first commandline argument
        mov         si, 0x5c       ; +0x002d  be5c00
        mov         di, 0x5fb      ; +0x0030  bffb05
        mov         cx, 9          ; +0x0033  b90900
        rep movsb                  ; +0x0036  f3a4
        mov         si, 0x528      ; +0x0038  be2805
        movsw                      ; +0x003b  a5
        movsb                      ; +0x003c  a4
    ; INT 21 - DOS 1+ - DELETE FILE USING FCB
    mov         ah, 0x13       ; +0x003d  b413
    mov         dx, 0x5fb      ; +0x003f  bafb05
    int         0x21           ; +0x0042  cd21

    mov         al, '$'        ; +0x0044  b024
    mov         di, 0x5fb+9    ; +0x0046  bf0406
    stosb                      ; +0x0049  aa
    stosb                      ; +0x004a  aa
    stosb                      ; +0x004b  aa

    ; INT 21 - DOS 1+ - DELETE FILE USING FCB
    mov         ah, 0x13       ; +0x004c  b413
    int         0x21           ; +0x004e  cd21

    ; INT 21 - DOS 1+ - CREATE OR TRUNCATE FILE USING FCB
    mov         ah, 0x16       ; +0x0050  b416
    int         0x21           ; +0x0052  cd21

So presumably % was chosen for that reason. But to be sure you’d probably have to ask Tim Paterson; otherwise your guess is as good as mine.

Answer 2

By using a \ as path separator, they needed a different marker for control character encoding, and did choose $ which in turn meant that the variable marker need to be another one, which turned out as %.

So by choosing % instead they could have kept $ for variables, but then again is the parser structured a bit different as it allows concatenated strings, thus needing a finishing marker at the end of named variables. So they didn't and the rest is history :))

---

I'm not complete so sure right now about the full implications, but $ is used at least in three different ways on the MS-DOS command line and in batch.

• Expand a batch variable to a full path name with %~$PATH:<varnumber>
• Defining special content for the PROMPT command like date and time as in prompt $d $t$g.
• Code the escape character in parameters as $e like echo $e[0m for screen reset.

Especially the does give the lead here, as marking special characters in Unix is done via a backslash \, which in DOS marks a directory level. MS had to use $ instead, so \e became $e.

Further, but way later (DOS 5)

• DOSKEY used $ to indicate special functions in macro generation.

---

Surprisingly, its predecessor CP/M did use $ as variable indicator as well as marker for special meanings in some commands (like privileges in STAT).

In CP/M batch files are run with the SUBMIT command, which will read a file (standard ending .SUB) and feed each line to the OS, after replacing variables marked with $. DOS has this feature build in and treats .BAT files like programs.

Answer 3

I can't know for sure but I suspect one of the reasons was that $ was used as the string terminating character in the standard DOS "print string" API (int 21h ah=9h), a convention apparently inherited from CP/M.