C语言中奇怪的输出缓冲行为

Question

I am trying out a C shell implementation from an open course, yet there is something intriguing about the behavior of the output buffering.

The code goes like this (note the line where I use pid = waitpid(-1, &r, WNOHANG) ):

int
main(void)
{
  static char buf[100];
  int fd, r;
  pid_t pid = 0;

  // Read and run input commands.
  while(getcmd(buf, sizeof(buf)) >= 0){
    if(buf[0] == 'c' && buf[1] == 'd' && buf[2] == ' '){
      buf[strlen(buf)-1] = 0;  // chop \n
      if(chdir(buf+3) < 0)
        fprintf(stderr, "cannot cd %s\n", buf+3);
      continue;
    }
    if((pid = fork1()) == 0)
      runcmd(parsecmd(buf));
    while ((pid = waitpid(-1, &r, WNOHANG)) >= 0) {
      if (errno == ECHILD) {
          break;
      }
    }
  }
  exit(0);
}

The runcmd function is like this (note that in pipe handling I create 2 child processes and wait for them to terminate):

void
runcmd(struct cmd *cmd)
{
  int p[2], r;
  struct execcmd *ecmd;
  struct pipecmd *pcmd;
  struct redircmd *rcmd;

  if(cmd == 0)
    exit(0);

  switch(cmd->type){
  case ' ':
    ecmd = (struct execcmd*)cmd;
    if(ecmd->argv[0] == 0) {
      exit(0);
    }
    // Your code here ...
    // fprintf(stderr, "starting to run cmd: %s\n", ecmd->argv[0]);
    execvp(ecmd->argv[0], ecmd->argv);
    fprintf(stderr, "exec error !\n");
    exit(-1);

    break;

  case '>':
  case '<':
    rcmd = (struct redircmd*)cmd;
    // fprintf(stderr, "starting to run <> cmd: %s\n", rcmd->file);
    // Your code here ...
    mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH;
    if (rcmd->type == '<') {
      // input
      close(0);
      if (open(rcmd->file, O_RDONLY, mode) != 0) {
        fprintf(stderr, "Opening file error !\n");
        exit(-1);
      }
    } else {
      // output
      close(1);
      if (open(rcmd->file, O_WRONLY|O_CREAT|O_TRUNC, mode) != 1) {
        fprintf(stderr, "Opening file error !\n");
        exit(-1);
      }
    }

    runcmd(rcmd->cmd);
    break;

  case '|':
    pcmd = (struct pipecmd*)cmd;
    // fprintf(stderr, "starting to run pcmd\n");
    // Your code here ...
    pipe(p);

    if (fork1() == 0) {
      // child for read, right side command
      close(0);
      if (dup(p[0]) != 0) {
        fprintf(stderr, "error when dup !\n");
        exit(-1);
      }
      close(p[0]);
      close(p[1]);
      runcmd(pcmd->right);
      fprintf(stderr, "exec error !\n");
    } 
    if (fork1() == 0) {
      // left side command for writing
      close(1);
      if (dup(p[1]) != 1) {
        fprintf(stderr, "dup error !\n");
        exit(-1);
      }
      close(p[0]);
      close(p[1]);
      runcmd(pcmd->left);
      fprintf(stderr, "exec error !\n");
    }
    close(p[0]);
    close(p[1]);
    int stat;
    wait(&stat);
    wait(&stat);

    break;

  default:
    fprintf(stderr, "unknown runcmd\n");
    exit(-1);
  }    
  exit(0);
}

The wierd thing is, when I execute "ls | sort" in the terminal, I constantly get the following output

6.828$ ls | sort
6.828$ a.out
sh.c
t.sh

This indicates that before the next command prompt "6828$" is printed, the output from the child process is still not flushed to terminal.

However, if I don't use pid = waitpid(-1, &r, WNOHANG)) and use pid = waitpid(-1, &r, 0)) (or wait() ), the output would be normal like:

6.828$ ls | sort
a.out
sh.c
t.sh

I have been thinking about the cause of the problem for a long time but did not come up with a possible reason. Can anyone suggest some possible reason?

Thanks a lot!

Answer 1

This code does not have well-defined behaviour:

while ((pid = waitpid(-1, &r, WNOHANG)) >= 0) {
  if (errno == ECHILD) {
      break;
  }
}

The while loop breaks immediately if waitpid returns -1, which is precisely what it returns in the case of an error. So if the body of the loop is entered, waitpid returned some non-negative value: either 0 -- indicating that the child is still executing -- or the pid of a child which had exited. Those are not error conditions, so the value of errno is not meaningful. It might be ECHILD , in which case the loop will incorrectly break.

You must only check the value of errno in cases where the value is meaningful. Or, to be more precise, quoting the Posix standard :

The value of errno shall be defined only after a call to a function for which it is explicitly stated to be set and until it is changed by the next function call or if the application assigns it a value. The value of errno should only be examined when it is indicated to be valid by a function's return value.

But I'm puzzled why you feel it necessary to busy loop using WNOHANG . That's a massive waste of resources, since your parent process will repeatedly execute the system call until the child actually terminates. Since you really intend to wait until the child terminates, it would make much more sense to just call wait or to specify 0 as a flag value to waitpid .

On the other hand, you might want to repeat the wait (or waitpid ) if it returns -1 with errno set to EINTR . And if it returns -1 and errno is neither EINTR nor ECHILD , then some hard error has occurred which you might want to log. But that's not related to your problem, afaics.