pm_system()

Updated: 12 May 2016

Name

pm_system - run a Netpbm program with program input and output

Synopsis


#include <netpbm/pm_system.h>

pm_system(void                  stdinFeeder(int, void *),
          void *          const feederParm,
          void                  stdoutAccepter(int, void *),
          void *          const accepterParm,
          const char *    const shellCommand);

pm_system_lp(const char *  const progName,
             void                stdinFeeder(int, void *),
             void *        const feederParm,
             void                stdoutAccepter(int, void *),
             void *        const accepterParm,
             ...);

pm_system_vp(const char *  const progName,
             const char ** const argArray,
             void                stdinFeeder(int, void *),
             void *        const feederParm,
             void                stdoutAccepter(int, void *),
             void *        const accepterParm);

pm_system2(void                  stdinFeeder(int, void *),
           void *          const feederParm,
           void                  stdoutAccepter(int, void *),
           void *          const accepterParm,
           const char *    const shellCommand,
           int *           const termStatusP );

pm_system2_lp(const char *  const progName,
             void                stdinFeeder(int, void *),
             void *        const feederParm,
             void                stdoutAccepter(int, void *),
             void *        const accepterParm,
             int *         const terminationStatusP,
             ...);

pm_system2_vp(const char *  const progName,
             const char ** const argArray,
             void                stdinFeeder(int, void *),
             void *        const feederParm,
             void                stdoutAccepter(int, void *),
             void *        const accepterParm,
             int *         const termStatusP);


Example

This simple example converts a PNM image on Standard Input to a JFIF (JPEG) image on Standard Output. In this case, pm_system() is doing no more than system() would do.


    pm_system(NULL, NULL, NULL, NULL, "pnmtojpeg");

This example does the same thing, but moves the data through memory buffers to illustrate use with memory buffers, and we throw in a stage to shrink the image too:

#include <netpbm/pm_system.h>

char              pnmData[100*1024];   /* Input file better be < 100K */
char              jfifData[100*1024];
struct bufferDesc pnmBuffer;
struct bufferDesc jfifBuffer;
unsigned int      jfifSize;

pnmBuffer.size = fread(pnmData, 1, sizeof(pnmData), stdin);
pnmBuffer.buffer = pnmData;
pnmBuffer.bytesTransferredP = NULL;

jfifBuffer.size = sizeof(jfifData);
jfifBuffer.buffer = jfifData;
jfifBuffer.bytesTransferredP = &jfifSize; 

pm_system(&pm_feed_from_memory, &pnmBuffer,
          &pm_accept_to_memory, &jfifBuffer,
          "pamscale .5 | pnmtojpeg");

fwrite(jfifData, 1, jfifSize, stdout);

This example reads an image into libnetpbm PAM structures, then brightens it, then writes it out, to illustrate use of pm_system with PAM structures.

#include <netpbm/pam.h>
#include <netpbm/pm_system.h>

struct pam       inpam;
struct pam       outpam;
tuple **         inTuples;
tuple **         outTuples;
struct pamtuples inPamtuples;
struct pamtuples outPamtuples;

inTuples = pnm_readpam(stdin, &inpam, sizeof(inpam));

outpam = inpam;

inPamtuples.pamP = &inpam;
inPamtuples.tuplesP = &inTuples;
outPamtuples.pamP = &outpam;
outPamtuples.tuplesP = &outTuples;

pm_system(&pm_feed_from_pamtuples, &inPamtuples,
          &pm_accept_to_pamtuples, &outPamtuples,
          "pambrighten -value +100");

outpam.file = stdout;
pnm_writepam(&outpam, outTuples);

DESCRIPTION

These library functions are part of Netpbm.

pm_system() is a lot like the standard C library system() subroutine. It runs a shell and has that shell execute a shell command that you specify. But pm_system() gives you more control over the Standard Input and Standard Output of that shell command than system(). system() passes to the shell command as Standard Input and Output whatever is the Standard Input and Output of the process that calls system(). But with pm_system(), you specify as arguments subroutines to execute to generate the shell command's Standard Input stream and to process the shell command's Standard Output stream.

Your Standard Input feeder subroutine can generate the stream in limitless ways. pm_system() gives it a file descriptor of a pipe to which to write the stream it generates. pm_system() hooks up the other end of that pipe to the shell command's Standard Input.

Likewise, your Standard Output accepter subroutine can do anything it wants with the stream it gets. pm_system() gives it a file descriptor of a pipe from which to read the stream. pm_system() hooks up the other end of that pipe to the shell command's Standard Output.

The argument stdinFeeder is a function pointer that identifies your Standard Input feeder subroutine. pm_system() runs it in a child process and waits for that process to terminate (and accepts its completion status) before returning. feederParm is the argument that pm_system() passes to the subroutine; it is opaque to pm_system().

If you pass stdinFeeder = NULL, pm_system() simply passes your current Standard Input stream to the shell command (as system() would do), and does not create a child process.

The argument stdoutAccepter is a function pointer that identifies your Standard Output accepter subroutine. pm_system() calls it in the current process. accepterParm is an argument analogous to feederParm.

If you pass stdoutAccepter = NULL, pm_system() simply passes your current Standard Output stream to the shell command (as system() would do.

The argument shellCommand is a null-terminated string containing the shell command that the shell is to execute. It can be any command that means something to the shell and can take a pipe for Standard Input and Output. Example:


    pambrighten -vale +100 | pamdepth 255 | pamscale .5

pm_system() creates a child process to run the shell and waits for that process to terminate (and accepts its completion status) before returning.

If the shell fails, i.e. does not exit voluntarily with zero exit status, pm_system calls pm_error(), which normally issues an error message to Standard Error and exits the program. Use pm_system2() if you don't want that.

Note that the "termination status" of a Unix process is a value which is a combination of 1) whether the process exited voluntarily or was killed by the operating system; 2) in the case of termination by the OS, what class of signal did it; and 3) in the case of voluntary exit, what "exit status" the program declared.

Interface Header File

These interfaces are declared by <netpbm/pm_system.h<.

pm_system2()

This is the same as pm_system() except that rather than respond to the shell process' termination status, it just returns it to you (via your termStatusP argument).

pm_system_lp()

pm_system_lp() is like pm_system() except that instead of running a shell, which in turn typically runs another program, you run a program of your choice directly.

Argument progName identifies the program to run, the same way as with execlp() or a shell command: if it contains a slash (/), it is the full name of the file that contains the program. If not, it is a name to be looked up in the system's program search path (determined by the PATH environment variable).

You identify the arguments to the program the same way as for execlp(): with the variable arguments at the end of the pm_system_lp() argument list. Each is a NUL-terminated string. The last argument must be NULL to tell pm_system_lp() where the arguments end.

Note that the first argument ("arg0") to a program is conventionally the first word of the command used to run the program, as if it were being run for a shell command. In other words, typically the name of the program.

Example:


    pm_system_lp("pnmtojpeg", NULL, NULL, NULL, NULL,
                 "pnmtojpeg", "mypicture.jpg", "-quality=50", NULL);

pm_system_lp() is much safer than pm_system() when your program computes the arguments or gets them from a user. If you build a shell command using such arguments, unless you're really careful, you may end up building a shell command that does something very different from what you intended, because the argument could contain characters that mean something to the shell such as "|".

pm_system_lp() can also be considerably faster that pm_system(), since it skips the whole running of the shell.

If the process fails, i.e. produces nonzero termination status, pm_system_lp calls pm_error(), which normally issues an error message to Standard Error and exits the program. Use pm_system2_lp() if you don't want that.

pm_system2_lp()

This is the same as pm_system_lp() except that rather than respond to the process' termination status, it just returns it to you (via your termStatusP argument).

pm_system_vp()

pm_system_vp() is like pm_system_lp() except that instead of supplying the program arguments as variable arguments, you supply them as an array, as with execvp(). A NULL element in the array identifies the end of the arguments.

Example:


    const char * argArray[3];

    argArray[0] = "pnmtojpeg";
    argArray[1] = "-quality=50";
    argArray[2] = NULL;

    pm_system_vp("pnmtojpeg", argArray, NULL, NULL, NULL, NULL);

If the process fails, i.e. produces nonzero termination status, pm_system_vp calls pm_error(), which normally issues an error message to Standard Error and exits the program. Use pm_system2_vp() if you don't want that.

pm_system2_vp()

This is the same as pm_system_vp() except that rather than respond to the process' termination status, it just returns it to you (via your termStatusP argument).

Applications

The point of pm_system() and friends is to allow you to write a C program that uses other programs internally, as a shell script would. This is particularly desirable with Netpbm, because Netpbm consists of a lot of programs that perform basic graphic manipulations and you'd like to be able to build a program that does a more sophisticated graphic manipulation by calling the more basic Netpbm programs. These building block programs typically take input from Standard Input and write output to Standard Output.

The obvious alternative is to use a higher level language -- Bourne Shell or Perl, for example. But often you want your program to do manipulations of your graphical data that are easier and more efficient in C. Or you want to use the Netpbm subroutine library in your program. The Netpbm subroutine library is a C-linkage library; the subroutines in it are not usable from a Bourne Shell or Perl program.

A typical use of pm_system() is to place the contents of some graphical image file in memory, run a Netpbm program against it, and have what would ordinarily go into an output file in memory too, for further processing. To do that, you can use the memory buffer Standard Input feeder and Standard Output accepter described below.

If your program uses the Netpbm subroutine library to read, write, and manipulate images, you may have an image in an array of PAM tuples. If you want to manipulate that image with a Netpbm program (perhaps remap the colors using pnmremap), you can use the pamtuple Standard Input feeder and Standard Output acceptor described below.

Broken Pipe Behavior

When you set up a shell command to take input from a pipe, as you do with pm_system(), you need to understand how pipes work with respect to the programs at either end of the pipe agreeing to how much data is to be transferred. Here are some notes on that.

It is normal to read a pipe before the process on the other end has written the data you hope to read, and it is normal to write to a pipe before the process on the other end has tried to read your data. Writes to a pipe can be buffered until the reading end requests the data. A process reading or writing a pipe can block until the other end is ready. Or a read or write can complete with an indication that the other end is not ready at the moment and therefore no data, or less data than was requested, was transferred.

The pipe is normally controlled by the writing end. When you read from a pipe, you keep reading until the program on the other end of the pipe closes it, and then you get an end-of-file indication. You then normally close the reading end of the pipe, since it is no longer useful.

When you close the reading end of a pipe before getting the end-of-file indication and the writer subsequently tries to write to the pipe, that is an error condition for the writer. In a typical default Unix environment, that error causes the writer to receive a SIGPIPE signal and that signal causes the writer process to terminate abnormally. But if, alternatively, the writer has ordered that SIGPIPE be blocked, ignored, or handled, the signal does not cause the death of the writer. Instead, the write operation simply completes with an error indication.

Standard Feeders And Acceptors

You can supply anything you like as a Standard Input feeder or Standard Output acceptor, but the Netpbm subroutine library comes with a few that perform commonly needed functions.

Memory Buffer

These routines are for when you just want to treat an area of memory as a file. If the shell command would ordinarily read a 513 byte regular file from its Standard Input, you want it to take 513 bytes from a certain address in your process' memory. Whatever bytes the shell command wants to write to its output file you want it to store at another address in your process' memory.

The Standard Input feeder for this is called pm_feed_from_memory. The Standard Output accepter is pm_accept_to_memory.

For both of these, the argument is a pointer to a struct bufferDesc, which is defined as follows:

struct bufferDesc {
    unsigned int    size;
    unsigned char * buffer;
    unsigned int *  bytesTransferredP;
};
size is the size of the memory buffer and buffer is its location in memory (address). The Standard Input feeder will attempt to feed the entire buffer to the shell command's Standard Input; the Standard Output accepter will not accept any more data from the shell command's Standard Output than will fit in the buffer. Both return the actual amount of data read or written, in bytes, at the location identified by bytesTransferredP. Unless bytesTransferredP is NULL.

Because a process typically terminates abnormally when it is not able to write everything to a pipe that it wanted to, bytesTransferredP is not usually useful in the Standard Input feeder case.

File Stream

These routines are for using an actual file as input or output. I.e. Standard Input comes from a file and Standard Output goes to a file. You open the file and create a libc file stream (type FILE) from it and pass that stream object to the standard input feeder or standard output accepter.

When you do this (using output as an example), your process writes to Standard Output, which is a pipe, the standard output accepter then reads from that pipe into a buffer, and then the standard output accepter writes from that buffer to your file. You could alternatively just arrange for the Standard Input or Output of your process to be the file and skip a copy, but it might be more complex coding.

Note that if the calling program's Standard Output is already set up as the file to which you want your process' output to go, you don't need this. All you have to do is decline to specify a Standard Output accepter (use NULL in place of the Standard Output accepter pointer) and your process' output will go there.

The Standard Input feeder for this is called pm_feed_from_filestream. The Standard Output accepter is pm_accept_to_filestream.

For both of these, the argument is a pointer to a Standard C library FILE object.

Example:

    
      int termStatus;
      FILE * myFileP;
      myFileP = fopen("tmp/myfile", "r");
      pm_system2(pm_feed_from_filestream, myFileP,
                 NULL, NULL,
                 "grep myword", &termstatus);
    
  

These routines were new in Netpbm 11.05 (December 2023).

Pamtuple

These routines are for when you have images in memory in the data structures used by the PAM family of subroutines in the Netpbm library -- i.e. struct PAM and an array of struct tuple. With these routines, you can run a Netpbm program against such an image just as you would against the same image in a regular file.

The Standard Input feeder for this is called pm_feed_from_pamtuples. The Standard Output accepter is pm_accept_to_pamtuples.

For both of these, the argument is the address of a struct pamtuples, which is defined as follows:

struct pamtuples {
    struct pam * pamP;
    tuple ***    tuplesP;
};

For the Standard Input feeder, you supply a struct pam, valid up through the tuple_type member (except it doesn't matter what the file member is) and array of tuples.

For the Standard Output Accepter, you supply only space in memory for the struct pam and the address of the tuple array. The routine fills in the struct pam up through the tuple_type member (except leaves the file member undefined) and allocates space for the tuple array with malloc(). You are responsible for freeing that memory.

HISTORY

pm_system() was introduced in Netpbm 10.13 (January 2003).

pm_system_lp() and pm_system_vp() were introduced in Netpbm 10.40 (September 2007).

pm_system2(), pm_system2_lp(), and pm_system2_vp() were introduce in Netpbm 10.75 (June 2016).