GEMDOS Extended Argument (ARGV) specification
GEMDOS EXTENDED ARGUMENT (ARGV) SPECIFICATION
Introduction
The Pexec() function of GEMDOS allows a program to pass to a child process a command line up to 125 characters long, with arguments separated by spaces. No provision is made in GEMDOS for the child to know its own name. This makes it difficult for C programs to correctly fill in argv[0], the standard place where a C program finds the command which invoked it. Because the command line arguments are separated by spaces, it is difficult to pass an argument with an embedded space. This document will specify a method of passing arguments which allows arbitrary argument length, embedded spaces, and support for argv[0].
Standard Argument Passing
The Pexec Cookbook specifies how to use Pexec() to launch a child process, passing a command tail (argument string) and an environment. Before getting into the extended argument scheme, let's review how arguments are normally passed to a child.
A parent process builds a command line into an argument string - a null terminated string whose first byte contains the length of the rest of the string - and its address is passed as one of the arguments to Pexec(). GEMDOS copies this argument string to the basepage which it creates for the child. Thus the parent is responsible for gathering all the child's arguments into one string. This is normally handled by a library exec() function. The child is responsible for parsing the string of space-separated arguments back into an array of strings. This parsing is normally handled by the child's startup code.
Evolution
Several methods of bypassing the limits imposed by Pexec() have been used by GEMDOS programs. Some allow a user to specify a file on the command line which contains the rest of the arguments. Others get a pointer to the arguments, or the arguments themselves, from the environment string. Most MS-DOS programs use a command file for the extra arguments. This can be inconvenient for a user, cluttering the file system with command files, and making the operation of batch files and makefiles more confusing.
Several "standards" have arisen on the ST which use the environment to pass arguments. While more convenient than command files, these standards have other problems. Some rely on sharing memory between parent and child processes. Some take advantage of undocumented features of the operating system to get argv[0]. Others give the child process no way to validate that the arguments it finds are intended for it.
Rationale
In order to pass more than the standard 125 characters worth of arguments to a child, or to let the child find its name, the parent must place the extra information in a place where the child can access it safely and legally. The most convenient place is in the child's environment string. An environment string is a series of null-terminated strings of the format "VARIABLE=value" (e.g. PATH=c:\bin,c:\etc, or ShellP=YES). The last null-terminated string in the environment is followed by a zero byte, thus two consecutive nulls indicates the end of the environment. The environment is allocated for the child by GEMDOS, it is owned by the child, and its contents can be specified by the parent.
The child must have some way of knowing that the arguments which it finds in its environment are intended for it. The child may have been invoked by a parent which does not conform to this specification. Such a parent would leave _its_ arguments in the environment, and could pass that environment on to the child. The child would mistakenly interpret its parent's arguments as its own.
Placing arguments in the environment passed to the child gets around all of the command line limits of the standard Pexec() command tail. Because there is no limit on the length of the environment, arbitrary length arguments are supported. Arguments placed in the environment are null terminated, so they may contain spaces. A parent can also place the name of the command with which it invokes the child in the child's environment, providing support for argv[0]. Validation of the extended arguments can be placed in the standard Pexec() command line, by assigning a special meaning to an invalid length byte.
The GEMDOS Extended Argument Specification
This specification uses the convention that the presence of an environment variable named ARGV (all upper case) indicates that extended arguments are being passed to the child in its environment. This means that ARGV is a "boolean" environment variable. For the purpose of this specification, its value is not significant, but its presence indicates that the strings following it are the arguments for the child. Implementations of this specification are free to give the ARGV environment variable any value. The ARGV environment variable must be the last one in the environment passed to the child, so that the child can truncate its environment at that point, and treat everything before the ARGV as environment, and everything after it as arguments.
The first argument to the child (argv[0]) is the first string in the environment after the ARGV variable. This argument is the "pathname" parameter passed by the parent to Pexec(). The remaining arguments are those that the child would normally find in the command tail in its basepage. Even if all of the arguments would normally fit in a child's command tail, the parent should set up the arguments in the environment to take advantage of the benefits of this extended argument scheme.
As many arguments as will fit in the command tail will be passed there as well as in the environment, to support non-conforming programs. As a flag that arguments are also in the environment, the length byte of the command tail will be 127 (hex 7f). Non-conforming programs should not have a problem with this length byte, because it is longer than the maximum 125 bytes allowed by Pexec().
As an aside, the Pexec Cookbook erroneously implies that a command line can be 126 or 127 characters long. In fact, GEMDOS only copies to the child's basepage up to 125 bytes, or until it encounters a null, from the argument string passed to Pexec(). It ignores the length byte, placing a null at the same place it found one or at the 126th byte if no null is found. This has several implications: the length byte is not validated by GEMDOS (necessitating validation in the child's startup code, but also making this extended argument spec possible), and the null terminator _can_ be located after the end of the real command tail (the Desktop places a CR character after the command tail and before the null). The ARGSTART.S startup code listing below demonstrates how to correctly validate and parse a GEMDOS command tail.
A child which finds an ARGV environment variable can use the command tail length byte value of 127 to validate that the arguments following the variable are valid, and not just left over from a non-conforming parent which left its own ARGV arguments in the environment.
Because the strings in the environment following an ARGV variable are not environment variables, a child should truncate its own environment at the ARGV variable by changing the 'A' to a null.
Implementation: Parental Responsibilities
To pass arguments in the environment, a parent must create an environment string for the child. This can be achieved by first allocating as much space as is used in the parent's own environment, plus enough room for the ARGV variable and the arguments to the child, and then copying the parent's environment to the newly allocated area. Next, the ARGV variable must be appended, since it must be the last variable in the child's environment string. Following the ARGV variable is the null-terminated pathname of the child as passed to Pexec(), then the null-terminated arguments to the child, followed by a final null byte indicating the end of the environment.
After setting up the arguments in the environment, the parent must place as many arguments as it can fit in the command tail it passes to Pexec(). This way, a child which does not conform to this specification can still get arguments from the command tail in its basepage. When placing arguments in the environment, the parent must set the first (length) byte of the command tail to 127 (hex 7f), validating the arguments in the environment.
Here is an example execv() library routine in C. It uses three local utility routines, e_strlen(), e_strcpy(), and str0cpy() for getting environment size and copying strings into the environment created for the child.
/* EXECV.C - example execv() library routine
* ================================================================
* 890910 kbad
*/
long Malloc( long nbytes );
long Pexec( short mode, char *filename, char *tail, char *env );
long Mfree( void *address );
/* Return the total length of the characters and null terminators in
* an array of strings.
* `strings' is an array of pointers to strings, with a null pointer
* as the last element.
*/
static long
e_strlen( char *strings[] )
{
char *pstring;
long length = 0;
while( *strings != 0 ) { /* Until reaching null pointer, */
pstring = *strings++; /* get a string pointer, */
do { /* find the length of this string, */
++length; /* using do-while to count the */
} while( *pstring++ != 0 ); /* null terminator. */
}
return length; /* Return total length of all strings */
}
/* Copy a string, including the null terminator, and return a pointer
* to the end of the destination string.
*/
static char *
str0cpy( char *dest, char *source )
{
do { /* use do-while to include null terminator */
*dest++ = *source;
} while( *source++ != 0 );
return dest;
}
/* Copy an array of strings into an environment string, and return a pointer
* to the end of the environment string.
* `strings' is an array of pointers to strings with a null pointer
* as the last element.
* `envstring' points to the environment string.
*/
static char *
e_strcpy( char *envstring, char *strings[] )
{
while( *strings != 0 ) {
envstring = str0cpy( envstring, *strings );
++strings;
}
return envstring; /* Return end of environment string */
}
/* Run a program, passing it arguments according to the
* GEMDOS Extended Argument Spec.
*
* `childname' is the relative path\filename of the child to execute.
* `args' is an array of pointers to strings to be used as arguments
* to the child. The last array element must be a null pointer.
* `environ' is a global array of pointers to strings
* which make up the caller's environment.
*/
long
execv( char *childname, char *args[] )
{
long envsize, ret;
char *parg, *penvargs, *childenv, *pchildenv;
short lentail;
char argch, tail[128], *ptail;
static char argvar[] = "ARGV=";
extern char *environ[];
/*
* Find out how much memory we'll need for the child's environment
*/
envsize = e_strlen( environ ); /* length of environment */
envsize += e_strlen( args ); /* plus command tail args */
/* plus length of argv[0] */
parg = childname;
do { /* use do-while to include null terminator */
++envsize;
} while( *parg++ != 0 );
/* plus length of ARGV environment variable and final null */
envsize += 7;
envsize += envsize & 1; /* even # of bytes */
/*
* Allocate and fill in the child's environment
*/
ret = Malloc( envsize );
if( ret < 0 )
return ret; /* Malloc error */
childenv = (char *)ret;
pchildenv = e_strcpy( childenv, environ ); /* copy caller environment */
pchildenv = str0cpy( pchildenv, argvar ); /* append ARGV variable */
pchildenv = str0cpy( pchildenv, childname ); /* append argv[0] */
penvargs = pchildenv; /* save start of args */
pchildenv = e_strcpy( pchildenv, args ); /* append args */
*pchildenv = 0; /* terminate environment */
/* put as much in the command tail as will fit */
lentail = 0;
ptail = &tail[1];
while( (lentail < 126) && (penvargs < pchildenv) ) {
argch = *penvargs++;
if( argch == 0 ) {
*ptail++ = ' ';
} else {
*ptail++ = argch;
}
}
/* terminate command tail and validate ARGV */
*ptail = 0;
tail[0] = 127;
/*
* Execute child, returning the return code from Pexec()
*/
ret = Pexec( 0, childname, tail, childenv );
Mfree( childenv );
return ret;
}
/* End of execv() example code */Implementation: Prenatal Responsibilities
A program's startup code must handle getting extended arguments out of the environment. The startup code should get the basepage pointer off the stack, then get the environment pointer from the basepage, and search the environment for "ARGV=". If "ARGV=" is found, the command line length byte in the basepage is checked. If the command line length byte is 127, then the arguments in the environment are valid. The first argument begins after the first null following the "ARGV=". It is important not to assume that the null follows immediately after the "ARGV=", because some implementations may assign a value to the ARGV environment variable. After setting up an array of pointers to the arguments, the startup code should set the 'A' of the "ARGV" variable to null, thus separating the environment from the argument strings (remember: a double null terminates the environment).
Here is some example C startup code which shows how a child could look for arguments in its environment:
* ARGSTART.S - example C startup code
* using GEMDOS Extended Argument Specification
* ================================================================
* 890910 kbad
.globl _main ; external, C entry point
.globl _argv0 ; external, name used for argv[0] if no ARGV
.globl _stksize ; external, size of application stack
.globl _basepage ; allocated here, -> program's basepage
.globl _environ ; allocated here, -> envp[]
.globl _argvecs ; allocated here, -> argv[]
.globl _stklimit ; allocated here, -> lower limit of stack
.BSS
_basepage: ds.l 1
_environ: ds.l 1
_argvecs: ds.l 1
_stklimit: ds.l 1
.TEXT
_start:
move.l 4(sp),a5 ; get basepage
move.l a5,_basepage ; save it
move.l 24(a5),a0 ; bss base
add.l 28(a5),a0 ; plus bss size = envp[] base
move.l a0,_environ ; save start of envp[]
move.l a0,a1 ; start of env/arg vectors
move.l 44(a5),a2 ; basepage environment pointer
tst.b (a2) ; empty environment?
beq.s nargv ; yes, no envp[]
lea.l (sp),a4 ; use dummy return pc on stack for ARGV test
* --- fill in the envp[] array
nxenv: move.l a2,(a1)+ ; envp[n]
move.l a2,a3
nxen1: tst.b (a2)+
bne.s nxen1 ; get the end of this variable
tst.b (a2) ; end of env?
beq.s xenv
* --- check for ARGV
move.b (a3)+,-(a4) ; get 1st 4 bytes of this var
move.b (a3)+,-(a4)
move.b (a3)+,-(a4)
move.b (a3)+,-(a4)
cmp.l #'VGRA',(a4)+ ; is it ARGV?
bne.s nxenv
cmp.b #'=',(a3) ; is it ARGV=?
bne.s nxenv
clr.b -4(a3) ; ARGV marks the end of our environment
cmp.b #127,$80(a5) ; command line validation?
bne.s nargv ; nope... and we're done with the env.
* --- got an ARGV=, create argv[] array
clr.l (a1)+ ; terminate envp[]
move.l a1,_argvecs ; save base of argv[]
nxarg: move.l a2,(a1)+ ; argv[n]
nxar1: tst.b (a2)+
bne.s nxar1
tst.b (a2)
bne.s nxarg
* --- end of environment
xenv: move.l _argvecs,d0 ; if we got an argv[]
bne.s argok ; don't parse command tail
* --- No ARGV, parse the command tail
* NOTE: This code parses the command tail IN PLACE. This can cause problems
* because the default DTA set up by GEMDOS for a program is located
* in the command tail part of the basepage. You should use Fsetdta()
* to set up your own DTA before performing any operations which could
* use the DTA if you want to preserve the arguments in the command tail.
nargv: clr.l (a1)+ ; terminate envp[]
move.l a1,_argvecs ; base of argv[]
move.l #_argv0,(a1)+ ; default name for argv[0]
lea 128(a5),a2 ; command tail
move.b (a2)+,d2 ; length byte
ext d2
moveq #125,d1 ; validate length
cmp d1,d2
bcs.s valen
move d1,d2 ; if invalid length, copy all of tail
valen: clr.b 0(a2,d2) ; null tail because desktop inserts <cr>
moveq #' ',d1 ; space terminator
get1: move.b (a2)+,d2 ; null byte?
beq.s argok ; if so, we're done
cmp.b d1,d2 ; strip leading spaces
beq.s get1
subq #1,a2 ; unstrip start char
move.l a2,(a1)+ ; and store that arg
get2: move.b (a2)+,d2 ; next char
beq.s argok ; if null, we're done
cmp.b d1,d2 ; if not space...
bne.s get2 ; keep looking
clr.b -1(a2) ; terminate argv[argc] in the command tail
bra.s get1 ; get next arg
argok: clr.l (a1)+ ; terminate argv[]
* --- allocate stack
move.l a1,_stklimit ; end of env/arg vectors is stack limit
add.l _stksize,a1 ; allocate _stksize bytes of stack
move.l a1,sp ; set initial stack pointer
* --- release unused memory
sub.l a5,a1 ; size to keep
move.l a1,-(sp)
move.l a5,-(sp) ; base of block to shrink
pea $4a0000 ; Mshrink fn code + junk word of 0
trap #1
lea 12(sp),sp ; pop args
*
* Everything beyond here depends on implementation.
* At this point, _environ points to envp[], _argvecs points to argv[],
* and _stklimit points to the end of the argv array. Thus argc can
* be calculated as ((_stklimit-_argvecs)/4)-1.
* _main could be invoked as follows:
*
move.l a5,-(sp) ; basepage
move.l _environ,-(sp) ; envp[]
move.l _argvecs,-(sp) ; argv[]
move.l _stklimit,d0 ; 4 bytes past end of argv[]
sub.l (sp),d0 ; (argc+1) * sizeof( char * )
asr.l #2,d0 ; argc+1
subq #1,d0 ; argc
move d0,-(sp)
jsr _main ; call mainline
lea 14(sp),sp ; pop argsA Final Note
This specification was formulated with careful deliberation, and with input from several companies and developers who have created development tools for GEMDOS. The Mark Williams extended argument passing scheme was the main influence for this specification, because it has been in use, and supported by Mark Williams and other companies for several years. This specification is very similar to the Mark Williams scheme, with the following important exceptions:
- Under the specification, the arguments after the ARGV environment variable may be validated by checking the command tail length byte. The Mark Williams execve() library function uses the command tail length byte as a telltale, but it is not checked by the crts0 startup code. This validation is important for the reasons mentioned in the Rationale section above.
- The specification allows the ARGV environment variable to take on any value. Mark Williams uses the value of ARGV as an iovector, which is described in the Mark Williams documentation. The iovector should no longer be needed, as its primary purpose was to simplify the MWC implementation of the C library function isatty().
- Some versions of the MWC startup code do not require the ARGV= to have an `='. Because ARGV is an actual environment variable in the specification, the equals character is required.
--
||| Ken Badertscher (ames!atari!kbad)
||| Atari R&D System Software Engine
/ | \ #include <disclaimer>
