Programs of the .COM persuasion are stored in disk files that hold an absolute image of the machine instructions to be executed. Because the files contain no relocation information, they are more compact, and are loaded for execution slightly faster, than equivalent .EXE files. Note that MS-DOS does not attempt to ascertain whether a .COM file actually contains executable code (there is no signature or checksum, as in the case of a .EXE file); it simply brings any file with the .COM extension into memory and jumps to it.
Because .COM programs are loaded immediately above the program segment prefix and do not have a header that can specify another entry point, they must always have an origin of 0100H, which is the length of the PSP. Location 0100H must contain an executable instruction. The maximum length of a .COM program is 65,536 bytes, minus the length of the PSP (256 bytes) and a mandatory word of stack (2 bytes).
When control is transferred to the .COM program from MS-DOS, all of the segment registers point to the PSP (Figure 3-2). The stack pointer register contains 0FFFEH if memory allows; otherwise, it is set as high as possible in memory minus 2 bytes. (MS-DOS pushes a zero word on the stack before entry.)
Figure 3-2. A memory image of a typical .COM-type program after loading. The contents of the .COM file are brought into memory just above the program segment prefix. Program, code, and data are mixed together in the same segment, and all segment registers contain the same value.
Please refer to the printed book for this figure.
Although the size of an executable .COM file can't exceed 64 KB, the current versions of MS-DOS allocate all of the transient program area to .COM programs when they are loaded. Because many such programs date from the early days of MS-DOS and are not necessarily "well-behaved" in their approach to memory management, the operating system simply makes the worst-case assumption and gives .COM programs everything that is available. If a .COM program wants to use the EXEC function to invoke another process, it must first shrink down its memory allocation to the minimum memory it needs in order to continue, taking care to protect its stack. (This is discussed in more detail in Chapter 12.)
When a .COM program finishes executing, it can return control to MS-DOS by several means. The preferred method is Int 21H Function 4CH, which allows the program to pass a return code back to the program, shell, or batch file that invoked it. However, if the program is running under MS-DOS version 1, it must exit by means of Int 20H, Int 21H Function 0, or a NEAR RETURN. (Because a word of zero was pushed onto the stack at entry, a NEAR RETURN causes a transfer to PSP:0000, which contains an Int 20H instruction.)
A .COM-type application can be linked together from many separate object modules. All of the modules must use the same code-segment name and class name, and the module with the entry point at offset 0100H within the segment must be linked first. In addition, all of the procedures within a .COM program should have the NEAR attribute, because all executable code resides in one segment.
When linking a .COM program, the linker will display the message
Warning: no stack segment
This message can be ignored. The linker output is a .EXE file, which must be converted into a .COM file with the MS-DOS EXE2BIN utility before execution. You can then delete the .EXE file. (An example of this process is provided in Chapter 4.)