environ.txt

国外网站上的一些精典的C程序

Q. Is it possible to set an environment variable from within a program? I've
   tried setnev() and it doesn't work.


A. Generally, the answer is no. First of all, you must understand how
   environment variables work. The command shell (COMMAND.COM, 4DOS, or any
   of the Unix shells) is what maintains the environment variables. They are
   stored, usually concatenated, in a block of memory owned by the command
   shell. When your program runs, it inherits a copy of the environment
   block. This is a crucial concept! The command shell only copies its
   "master environment" and passes the copy to the program. When the program
   terminates, its copy, which was in memory it owned, is discarded along
   with any changes which were made to it. This is why putenv() seems not to
   work, since it only modifies the copy and the changes are lost upon
   program termination. Ultimately, the question you must answer is which
   copy of the environment you wish to modify? If you only wish to modify
   your local environment, use putenv(). If you want to modify something
   else, you have more questions to answer. Remember that several copies of
   the command shell may be running at a time, so which one's environment
   block do you want to modify? If you choose to modify your parent process'
   environment block, that too will be lost when your parent process
   terminates. If you want to modify the original master, then your changes
   may still be lost when you terminate since your parent process may not be
   the "master" command shell (see below for an illustration).

   It's a can of worms that the makers of most operating systems (OS's),
   including DOS, Unix, OS/2, Win95, et al, have wisely chosen to leave
   sealed!



Q. What good is putenv() if its changes are lost when the program ends?


A. Its primary use is to set up environment variables prior to spawning
   another subordinate process (program). In the PC world, you'd use it prior
   to issuing a spawn call. In the Unix world, use it prior to using fork()
   and exec().



Q. I know I've seen DOS programs which do modify the master environment, so I
   know it can be done!


A. Yes it can, but not legally or with guaranteed results. Most OS's do have
   undocumented ways of doing it. DOS is probably the safest to try since
   it's not multiuser or multitasking, and so the only person you can screw
   up is yourself! There are 4 ways to modify a non-local environment under
   DOS. Only one is marginally legal and the others rely on undocumented DOS
   features.



Q. Why was only the DOS batch file and "Stuff-key-buffer method"
   (SETENVAR.C) included in the original SNIPPETS?


A. The reason that I only included the "batch&stuff" method in my SNIPPETS
   collection is simply that it's the *only* method you can rely on if your
   program is going to be distributed. Quite simply, there is *NO* safe,
   documented way under DOS to set an environment variable in the master
   environment block - period! By back-tracking PSPs or MCBs, you can try
   to locate the master environment and change it. You can also try to use
   Int 2Eh, the command processor's "back door". But all of these methods
   suffer from several shortcomings:

1)  Someone using the program might be using 4DOS, COMMAND PLUS, or some other
    COMMAND.COM replacement. These don't always do things the same way as
    COMMAND.COM and the differences can cause you to crash, roll, & burn! For
    example, several COMMAND.COM replacements allow the master environment
    block to be located in extended, expanded, or high memory. In such a case,
    backtracking PSPs or MCBs is less than useless, they're guaranteed to
    yield undefined errors.

2)  Int 2Eh seems to be the most universally supported, but cannot be used in
    a program invoked from a batch file. The book, "Undocumented DOS" details
    some procedures for making an Int 2Eh call safer but, again, these
    techniques rely on implementation features of COMMAND.COM which might not
    be available in alternate command processors.

3)  Even if everything else is safe, you still need a way of error trapping in
    case your new environment variable might overwrite the end of the
    available master environment block. This error trapping in inherent in
    COMMAND.COM and alternate command processors (one reason why using the
    Int 2Eh back door is potentially the safest way to try), but if you try to
    modify things manually, you're on your own. If you do overwrite the end of
    the master environment block, you'll have automatically corrupted your MCB
    chain and possibly set yourself up for some *really* nasty surprises!

4)  Finally, there's the very fundamental question of which environment block
    really is the master? Say you're in your comm program and hit the "shell
    to DOS" key. A secondary copy of the command processor, be it COMMAND.COM
    or whatever, is spawned and you're off and running. If you now run your
    program from this secondary DOS shell, is its environment block the master
    or is it the one from which you ran your comm program? Worse yet,
    depending on how you set up CONFIG.SYS, the secondary shell may have a
    considerably smaller environment block than the original. Despite having
    set the "/E:" switch, your secondary shell will likely only have an
    environment block whose size is equal to the current block size rounded
    up to the next paragraph boundary. If you trace PSPs, you'll find the
    secondary shell which you stand a good chance of over-running due to the
    difference in the block size. If you trace MCBs, you'll find the real
    master block, but then your changes will have disappeared when you return
    to your comm program, defeating the purpose of your program in the first
    place. 

   The inability to alter a parent program's environment block isn't a DOS
   exclusive, BTW - it's an inheritance from Unix where the same limitation
   applies.

   Finally, SNIPPETS now includes several of these alternate unsafe ways of
   setting the master environment. INT2E.ASM & CCOMCALL.C together provide
   access to the DOS command processor back door, GLBL_ENV.C provides means
   for TC/TC++/BC++ and MSC/QC programmers to modify the master environment
   by backtracking PSP pointers, and MCB_ENV.C serves the same purpose only
   using the MCB tracking method.