xref: /netbsd-src/usr.bin/make/job.c (revision ae1bfcddc410612bc8c58b807e1830becb69a24c)
1 /*
2  * Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
3  * Copyright (c) 1988, 1989 by Adam de Boor
4  * Copyright (c) 1989 by Berkeley Softworks
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Adam de Boor.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #ifndef lint
40 /* from: static char sccsid[] = "@(#)job.c	5.15 (Berkeley) 3/1/91"; */
41 static char *rcsid = "$Id: job.c,v 1.5 1994/03/05 00:34:48 cgd Exp $";
42 #endif /* not lint */
43 
44 /*-
45  * job.c --
46  *	handle the creation etc. of our child processes.
47  *
48  * Interface:
49  *	Job_Make  	    	Start the creation of the given target.
50  *
51  *	Job_CatchChildren   	Check for and handle the termination of any
52  *	    	  	    	children. This must be called reasonably
53  *	    	  	    	frequently to keep the whole make going at
54  *	    	  	    	a decent clip, since job table entries aren't
55  *	    	  	    	removed until their process is caught this way.
56  *	    	  	    	Its single argument is TRUE if the function
57  *	    	  	    	should block waiting for a child to terminate.
58  *
59  *	Job_CatchOutput	    	Print any output our children have produced.
60  *	    	  	    	Should also be called fairly frequently to
61  *	    	  	    	keep the user informed of what's going on.
62  *	    	  	    	If no output is waiting, it will block for
63  *	    	  	    	a time given by the SEL_* constants, below,
64  *	    	  	    	or until output is ready.
65  *
66  *	Job_Init  	    	Called to intialize this module. in addition,
67  *	    	  	    	any commands attached to the .BEGIN target
68  *	    	  	    	are executed before this function returns.
69  *	    	  	    	Hence, the makefile must have been parsed
70  *	    	  	    	before this function is called.
71  *
72  *	Job_Full  	    	Return TRUE if the job table is filled.
73  *
74  *	Job_Empty 	    	Return TRUE if the job table is completely
75  *	    	  	    	empty.
76  *
77  *	Job_ParseShell	    	Given the line following a .SHELL target, parse
78  *	    	  	    	the line as a shell specification. Returns
79  *	    	  	    	FAILURE if the spec was incorrect.
80  *
81  *	Job_End	  	    	Perform any final processing which needs doing.
82  *	    	  	    	This includes the execution of any commands
83  *	    	  	    	which have been/were attached to the .END
84  *	    	  	    	target. It should only be called when the
85  *	    	  	    	job table is empty.
86  *
87  *	Job_AbortAll	    	Abort all currently running jobs. It doesn't
88  *	    	  	    	handle output or do anything for the jobs,
89  *	    	  	    	just kills them. It should only be called in
90  *	    	  	    	an emergency, as it were.
91  *
92  *	Job_CheckCommands   	Verify that the commands for a target are
93  *	    	  	    	ok. Provide them if necessary and possible.
94  *
95  *	Job_Touch 	    	Update a target without really updating it.
96  *
97  *	Job_Wait  	    	Wait for all currently-running jobs to finish.
98  */
99 
100 #include <sys/types.h>
101 #include <sys/signal.h>
102 #include <sys/stat.h>
103 #include <sys/file.h>
104 #include <sys/time.h>
105 #include <sys/wait.h>
106 #include <fcntl.h>
107 #include <errno.h>
108 #include <stdio.h>
109 #include <string.h>
110 #include <signal.h>
111 #include "make.h"
112 #include "hash.h"
113 #include "dir.h"
114 #include "job.h"
115 #include "pathnames.h"
116 
117 extern int  errno;
118 
119 /*
120  * error handling variables
121  */
122 static int     	errors = 0;	    /* number of errors reported */
123 static int    	aborting = 0;	    /* why is the make aborting? */
124 #define ABORT_ERROR	1   	    /* Because of an error */
125 #define ABORT_INTERRUPT	2   	    /* Because it was interrupted */
126 #define ABORT_WAIT	3   	    /* Waiting for jobs to finish */
127 
128 
129 /*
130  * post-make command processing. The node postCommands is really just the
131  * .END target but we keep it around to avoid having to search for it
132  * all the time.
133  */
134 static GNode   	  *postCommands;    /* node containing commands to execute when
135 				     * everything else is done */
136 static int     	  numCommands; 	    /* The number of commands actually printed
137 				     * for a target. Should this number be
138 				     * 0, no shell will be executed. */
139 
140 
141 /*
142  * Return values from JobStart.
143  */
144 #define JOB_RUNNING	0   	/* Job is running */
145 #define JOB_ERROR 	1   	/* Error in starting the job */
146 #define JOB_FINISHED	2   	/* The job is already finished */
147 #define JOB_STOPPED	3   	/* The job is stopped */
148 
149 /*
150  * tfile is the name of a file into which all shell commands are put. It is
151  * used over by removing it before the child shell is executed. The XXXXX in
152  * the string are replaced by the pid of the make process in a 5-character
153  * field with leading zeroes.
154  */
155 static char     tfile[] = TMPPAT;
156 
157 
158 /*
159  * Descriptions for various shells.
160  */
161 static Shell    shells[] = {
162     /*
163      * CSH description. The csh can do echo control by playing
164      * with the setting of the 'echo' shell variable. Sadly,
165      * however, it is unable to do error control nicely.
166      */
167 {
168     "csh",
169     TRUE, "unset verbose", "set verbose", "unset verbose", 10,
170     FALSE, "echo \"%s\"\n", "csh -c \"%s || exit 0\"",
171     "v", "e",
172 },
173     /*
174      * SH description. Echo control is also possible and, under
175      * sun UNIX anyway, one can even control error checking.
176      */
177 {
178     "sh",
179     TRUE, "set -", "set -v", "set -", 5,
180     FALSE, "echo \"%s\"\n", "sh -c '%s || exit 0'\n",
181     "v", "e",
182 },
183     /*
184      * UNKNOWN.
185      */
186 {
187     (char *)0,
188     FALSE, (char *)0, (char *)0, (char *)0, 0,
189     FALSE, (char *)0, (char *)0,
190     (char *)0, (char *)0,
191 }
192 };
193 static Shell 	*commandShell = &shells[DEFSHELL];/* this is the shell to
194 						   * which we pass all
195 						   * commands in the Makefile.
196 						   * It is set by the
197 						   * Job_ParseShell function */
198 static char   	*shellPath = (char *) NULL,	  /* full pathname of
199 						   * executable image */
200                	*shellName;	      	      	  /* last component of shell */
201 
202 
203 static int  	maxJobs;    	/* The most children we can run at once */
204 static int  	maxLocal;    	/* The most local ones we can have */
205 int     	nJobs;	    	/* The number of children currently running */
206 int  		nLocal;    	/* The number of local children */
207 Lst     	jobs;		/* The structures that describe them */
208 Boolean		jobFull;    	/* Flag to tell when the job table is full. It
209 				 * is set TRUE when (1) the total number of
210 				 * running jobs equals the maximum allowed or
211 				 * (2) a job can only be run locally, but
212 				 * nLocal equals maxLocal */
213 #ifndef RMT_WILL_WATCH
214 static fd_set  	outputs;    	/* Set of descriptors of pipes connected to
215 				 * the output channels of children */
216 #endif
217 
218 GNode   	*lastNode;	/* The node for which output was most recently
219 				 * produced. */
220 char    	*targFmt;   	/* Format string to use to head output from a
221 				 * job when it's not the most-recent job heard
222 				 * from */
223 #define TARG_FMT  "--- %s ---\n" /* Default format */
224 
225 /*
226  * When JobStart attempts to run a job remotely but can't, and isn't allowed
227  * to run the job locally, or when Job_CatchChildren detects a job that has
228  * been migrated home, the job is placed on the stoppedJobs queue to be run
229  * when the next job finishes.
230  */
231 Lst		stoppedJobs;	/* Lst of Job structures describing
232 				 * jobs that were stopped due to concurrency
233 				 * limits or migration home */
234 
235 
236 #if defined(USE_PGRP) && defined(SYSV)
237 #define KILL(pid,sig)	killpg (-(pid),(sig))
238 #else
239 # if defined(USE_PGRP)
240 #define KILL(pid,sig)	killpg ((pid),(sig))
241 # else
242 #define KILL(pid,sig)	kill ((pid),(sig))
243 # endif
244 #endif
245 
246 static int JobCondPassSig __P((Job *, int));
247 static void JobPassSig __P((int));
248 static int JobCmpPid __P((Job *, int));
249 static int JobPrintCommand __P((char *, Job *));
250 static int JobSaveCommand __P((char *, GNode *));
251 static void JobFinish __P((Job *, union wait));
252 static void JobExec __P((Job *, char **));
253 static void JobMakeArgv __P((Job *, char **));
254 static void JobRestart __P((Job *));
255 static int JobStart __P((GNode *, int, Job *));
256 static void JobDoOutput __P((Job *, Boolean));
257 static Shell *JobMatchShell __P((char *));
258 static void JobInterrupt __P((int));
259 
260 /*-
261  *-----------------------------------------------------------------------
262  * JobCondPassSig --
263  *	Pass a signal to a job if the job is remote or if USE_PGRP
264  *	is defined.
265  *
266  * Results:
267  *	=== 0
268  *
269  * Side Effects:
270  *	None, except the job may bite it.
271  *
272  *-----------------------------------------------------------------------
273  */
274 static int
275 JobCondPassSig(job, signo)
276     Job	    	*job;	    /* Job to biff */
277     int	    	signo;	    /* Signal to send it */
278 {
279 #ifdef RMT_WANTS_SIGNALS
280     if (job->flags & JOB_REMOTE) {
281 	(void)Rmt_Signal(job, signo);
282     } else {
283 	KILL(job->pid, signo);
284     }
285 #else
286     /*
287      * Assume that sending the signal to job->pid will signal any remote
288      * job as well.
289      */
290     KILL(job->pid, signo);
291 #endif
292     return(0);
293 }
294 
295 /*-
296  *-----------------------------------------------------------------------
297  * JobPassSig --
298  *	Pass a signal on to all remote jobs and to all local jobs if
299  *	USE_PGRP is defined, then die ourselves.
300  *
301  * Results:
302  *	None.
303  *
304  * Side Effects:
305  *	We die by the same signal.
306  *
307  *-----------------------------------------------------------------------
308  */
309 static void
310 JobPassSig(signo)
311     int	    signo;	/* The signal number we've received */
312 {
313     int	    mask;
314 
315     Lst_ForEach(jobs, JobCondPassSig, (ClientData)signo);
316 
317     /*
318      * Deal with proper cleanup based on the signal received. We only run
319      * the .INTERRUPT target if the signal was in fact an interrupt. The other
320      * three termination signals are more of a "get out *now*" command.
321      */
322     if (signo == SIGINT) {
323 	JobInterrupt(TRUE);
324     } else if ((signo == SIGHUP) || (signo == SIGTERM) || (signo == SIGQUIT)) {
325 	JobInterrupt(FALSE);
326     }
327 
328     /*
329      * Leave gracefully if SIGQUIT, rather than core dumping.
330      */
331     if (signo == SIGQUIT) {
332 	Finish(0);
333     }
334 
335     /*
336      * Send ourselves the signal now we've given the message to everyone else.
337      * Note we block everything else possible while we're getting the signal.
338      * This ensures that all our jobs get continued when we wake up before
339      * we take any other signal.
340      */
341     mask = sigblock(0);
342     (void) sigsetmask(~0 & ~(1 << (signo-1)));
343     signal(signo, SIG_DFL);
344 
345     kill(getpid(), signo);
346 
347     Lst_ForEach(jobs, JobCondPassSig, (ClientData)SIGCONT);
348 
349     sigsetmask(mask);
350     signal(signo, JobPassSig);
351 
352 }
353 
354 /*-
355  *-----------------------------------------------------------------------
356  * JobCmpPid  --
357  *	Compare the pid of the job with the given pid and return 0 if they
358  *	are equal. This function is called from Job_CatchChildren via
359  *	Lst_Find to find the job descriptor of the finished job.
360  *
361  * Results:
362  *	0 if the pid's match
363  *
364  * Side Effects:
365  *	None
366  *-----------------------------------------------------------------------
367  */
368 static int
369 JobCmpPid (job, pid)
370     int             pid;	/* process id desired */
371     Job            *job;	/* job to examine */
372 {
373     return (pid - job->pid);
374 }
375 
376 /*-
377  *-----------------------------------------------------------------------
378  * JobPrintCommand  --
379  *	Put out another command for the given job. If the command starts
380  *	with an @ or a - we process it specially. In the former case,
381  *	so long as the -s and -n flags weren't given to make, we stick
382  *	a shell-specific echoOff command in the script. In the latter,
383  *	we ignore errors for the entire job, unless the shell has error
384  *	control.
385  *	If the command is just "..." we take all future commands for this
386  *	job to be commands to be executed once the entire graph has been
387  *	made and return non-zero to signal that the end of the commands
388  *	was reached. These commands are later attached to the postCommands
389  *	node and executed by Job_End when all things are done.
390  *	This function is called from JobStart via Lst_ForEach.
391  *
392  * Results:
393  *	Always 0, unless the command was "..."
394  *
395  * Side Effects:
396  *	If the command begins with a '-' and the shell has no error control,
397  *	the JOB_IGNERR flag is set in the job descriptor.
398  *	If the command is "..." and we're not ignoring such things,
399  *	tailCmds is set to the successor node of the cmd.
400  *	numCommands is incremented if the command is actually printed.
401  *-----------------------------------------------------------------------
402  */
403 static int
404 JobPrintCommand (cmd, job)
405     char     	  *cmd;	    	    /* command string to print */
406     Job           *job;	    	    /* job for which to print it */
407 {
408     Boolean	  noSpecials;	    /* true if we shouldn't worry about
409 				     * inserting special commands into
410 				     * the input stream. */
411     Boolean       shutUp = FALSE;   /* true if we put a no echo command
412 				     * into the command file */
413     Boolean	  errOff = FALSE;   /* true if we turned error checking
414 				     * off before printing the command
415 				     * and need to turn it back on */
416     char       	  *cmdTemplate;	    /* Template to use when printing the
417 				     * command */
418     char    	  *cmdStart;	    /* Start of expanded command */
419     LstNode 	  cmdNode;  	    /* Node for replacing the command */
420 
421     noSpecials = (noExecute && ! (job->node->type & OP_MAKE));
422 
423     if (strcmp (cmd, "...") == 0) {
424 	job->node->type |= OP_SAVE_CMDS;
425 	if ((job->flags & JOB_IGNDOTS) == 0) {
426 	    job->tailCmds = Lst_Succ (Lst_Member (job->node->commands,
427 						  (ClientData)cmd));
428 	    return (1);
429 	}
430 	return (0);
431     }
432 
433 #define DBPRINTF(fmt, arg) if (DEBUG(JOB)) printf (fmt, arg); fprintf (job->cmdFILE, fmt, arg)
434 
435     numCommands += 1;
436 
437     /*
438      * For debugging, we replace each command with the result of expanding
439      * the variables in the command.
440      */
441     cmdNode = Lst_Member (job->node->commands, (ClientData)cmd);
442     cmdStart = cmd = Var_Subst (NULL, cmd, job->node, FALSE);
443     Lst_Replace (cmdNode, (ClientData)cmdStart);
444 
445     cmdTemplate = "%s\n";
446 
447     /*
448      * Check for leading @' and -'s to control echoing and error checking.
449      */
450     while (*cmd == '@' || *cmd == '-') {
451 	if (*cmd == '@') {
452 	    shutUp = TRUE;
453 	} else {
454 	    errOff = TRUE;
455 	}
456 	cmd++;
457     }
458 
459     while (isspace((unsigned char) *cmd))
460 	cmd++;
461 
462     if (shutUp) {
463 	if (! (job->flags & JOB_SILENT) && !noSpecials &&
464 	    commandShell->hasEchoCtl) {
465 		DBPRINTF ("%s\n", commandShell->echoOff);
466 	} else {
467 	    shutUp = FALSE;
468 	}
469     }
470 
471     if (errOff) {
472 	if ( ! (job->flags & JOB_IGNERR) && !noSpecials) {
473 	    if (commandShell->hasErrCtl) {
474 		/*
475 		 * we don't want the error-control commands showing
476 		 * up either, so we turn off echoing while executing
477 		 * them. We could put another field in the shell
478 		 * structure to tell JobDoOutput to look for this
479 		 * string too, but why make it any more complex than
480 		 * it already is?
481 		 */
482 		if (! (job->flags & JOB_SILENT) && !shutUp &&
483 		    commandShell->hasEchoCtl) {
484 			DBPRINTF ("%s\n", commandShell->echoOff);
485 			DBPRINTF ("%s\n", commandShell->ignErr);
486 			DBPRINTF ("%s\n", commandShell->echoOn);
487 		} else {
488 		    DBPRINTF ("%s\n", commandShell->ignErr);
489 		}
490 	    } else if (commandShell->ignErr &&
491 		       (*commandShell->ignErr != '\0'))
492 	    {
493 		/*
494 		 * The shell has no error control, so we need to be
495 		 * weird to get it to ignore any errors from the command.
496 		 * If echoing is turned on, we turn it off and use the
497 		 * errCheck template to echo the command. Leave echoing
498 		 * off so the user doesn't see the weirdness we go through
499 		 * to ignore errors. Set cmdTemplate to use the weirdness
500 		 * instead of the simple "%s\n" template.
501 		 */
502 		if (! (job->flags & JOB_SILENT) && !shutUp &&
503 		    commandShell->hasEchoCtl) {
504 			DBPRINTF ("%s\n", commandShell->echoOff);
505 			DBPRINTF (commandShell->errCheck, cmd);
506 			shutUp = TRUE;
507 		}
508 		cmdTemplate = commandShell->ignErr;
509 		/*
510 		 * The error ignoration (hee hee) is already taken care
511 		 * of by the ignErr template, so pretend error checking
512 		 * is still on.
513 		 */
514 		errOff = FALSE;
515 	    } else {
516 		errOff = FALSE;
517 	    }
518 	} else {
519 	    errOff = FALSE;
520 	}
521     }
522 
523     DBPRINTF (cmdTemplate, cmd);
524 
525     if (errOff) {
526 	/*
527 	 * If echoing is already off, there's no point in issuing the
528 	 * echoOff command. Otherwise we issue it and pretend it was on
529 	 * for the whole command...
530 	 */
531 	if (!shutUp && !(job->flags & JOB_SILENT) && commandShell->hasEchoCtl){
532 	    DBPRINTF ("%s\n", commandShell->echoOff);
533 	    shutUp = TRUE;
534 	}
535 	DBPRINTF ("%s\n", commandShell->errCheck);
536     }
537     if (shutUp) {
538 	DBPRINTF ("%s\n", commandShell->echoOn);
539     }
540     return (0);
541 }
542 
543 /*-
544  *-----------------------------------------------------------------------
545  * JobSaveCommand --
546  *	Save a command to be executed when everything else is done.
547  *	Callback function for JobFinish...
548  *
549  * Results:
550  *	Always returns 0
551  *
552  * Side Effects:
553  *	The command is tacked onto the end of postCommands's commands list.
554  *
555  *-----------------------------------------------------------------------
556  */
557 static int
558 JobSaveCommand (cmd, gn)
559     char    *cmd;
560     GNode   *gn;
561 {
562     cmd = Var_Subst (NULL, cmd, gn, FALSE);
563     (void)Lst_AtEnd (postCommands->commands, (ClientData)cmd);
564     return (0);
565 }
566 
567 /*-
568  *-----------------------------------------------------------------------
569  * JobFinish  --
570  *	Do final processing for the given job including updating
571  *	parents and starting new jobs as available/necessary. Note
572  *	that we pay no attention to the JOB_IGNERR flag here.
573  *	This is because when we're called because of a noexecute flag
574  *	or something, jstat.w_status is 0 and when called from
575  *	Job_CatchChildren, the status is zeroed if it s/b ignored.
576  *
577  * Results:
578  *	None
579  *
580  * Side Effects:
581  *	Some nodes may be put on the toBeMade queue.
582  *	Final commands for the job are placed on postCommands.
583  *
584  *	If we got an error and are aborting (aborting == ABORT_ERROR) and
585  *	the job list is now empty, we are done for the day.
586  *	If we recognized an error (errors !=0), we set the aborting flag
587  *	to ABORT_ERROR so no more jobs will be started.
588  *-----------------------------------------------------------------------
589  */
590 /*ARGSUSED*/
591 static void
592 JobFinish (job, status)
593     Job           *job;	      	  /* job to finish */
594     union wait	  status;     	  /* sub-why job went away */
595 {
596     Boolean 	  done;
597 
598     if ((WIFEXITED(status) &&
599 	  (((status.w_retcode != 0) && !(job->flags & JOB_IGNERR)))) ||
600 	(WIFSIGNALED(status) && (status.w_termsig != SIGCONT)))
601     {
602 	/*
603 	 * If it exited non-zero and either we're doing things our
604 	 * way or we're not ignoring errors, the job is finished.
605 	 * Similarly, if the shell died because of a signal
606 	 * the job is also finished. In these
607 	 * cases, finish out the job's output before printing the exit
608 	 * status...
609 	 */
610 	if (usePipes) {
611 #ifdef RMT_WILL_WATCH
612 	    Rmt_Ignore(job->inPipe);
613 #else
614 	    FD_CLR(job->inPipe, &outputs);
615 #endif /* RMT_WILL_WATCH */
616 	    if (job->outPipe != job->inPipe) {
617 		(void)close (job->outPipe);
618 	    }
619 	    JobDoOutput (job, TRUE);
620 	    (void)close (job->inPipe);
621 	} else {
622 	    (void)close (job->outFd);
623 	    JobDoOutput (job, TRUE);
624 	}
625 
626 	if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
627 	    fclose(job->cmdFILE);
628 	}
629 	done = TRUE;
630     } else if (WIFEXITED(status) && status.w_retcode != 0) {
631 	/*
632 	 * Deal with ignored errors in -B mode. We need to print a message
633 	 * telling of the ignored error as well as setting status.w_status
634 	 * to 0 so the next command gets run. To do this, we set done to be
635 	 * TRUE if in -B mode and the job exited non-zero. Note we don't
636 	 * want to close down any of the streams until we know we're at the
637 	 * end.
638 	 */
639 	done = TRUE;
640     } else {
641 	/*
642 	 * No need to close things down or anything.
643 	 */
644 	done = FALSE;
645     }
646 
647     if (done ||
648 	WIFSTOPPED(status) ||
649 	(WIFSIGNALED(status) && (status.w_termsig == SIGCONT)) ||
650 	DEBUG(JOB))
651     {
652 	FILE	  *out;
653 
654 	if (!usePipes && (job->flags & JOB_IGNERR)) {
655 	    /*
656 	     * If output is going to a file and this job is ignoring
657 	     * errors, arrange to have the exit status sent to the
658 	     * output file as well.
659 	     */
660 	    out = fdopen (job->outFd, "w");
661 	} else {
662 	    out = stdout;
663 	}
664 
665 	if (WIFEXITED(status)) {
666 	    if (status.w_retcode != 0) {
667 		if (usePipes && job->node != lastNode) {
668 		    fprintf (out, targFmt, job->node->name);
669 		    lastNode = job->node;
670 		}
671 		fprintf (out, "*** Error code %d%s\n", status.w_retcode,
672 			 (job->flags & JOB_IGNERR) ? " (ignored)" : "");
673 
674 		if (job->flags & JOB_IGNERR) {
675 		    status.w_status = 0;
676 		}
677 	    } else if (DEBUG(JOB)) {
678 		if (usePipes && job->node != lastNode) {
679 		    fprintf (out, targFmt, job->node->name);
680 		    lastNode = job->node;
681 		}
682 		fprintf (out, "*** Completed successfully\n");
683 	    }
684 	} else if (WIFSTOPPED(status)) {
685 	    if (usePipes && job->node != lastNode) {
686 		fprintf (out, targFmt, job->node->name);
687 		lastNode = job->node;
688 	    }
689 	    if (! (job->flags & JOB_REMIGRATE)) {
690 		fprintf (out, "*** Stopped -- signal %d\n", status.w_stopsig);
691 	    }
692 	    job->flags |= JOB_RESUME;
693 	    (void)Lst_AtEnd(stoppedJobs, (ClientData)job);
694 	    fflush(out);
695 	    return;
696 	} else if (status.w_termsig == SIGCONT) {
697 	    /*
698 	     * If the beastie has continued, shift the Job from the stopped
699 	     * list to the running one (or re-stop it if concurrency is
700 	     * exceeded) and go and get another child.
701 	     */
702 	    if (job->flags & (JOB_RESUME|JOB_REMIGRATE|JOB_RESTART)) {
703 		if (usePipes && job->node != lastNode) {
704 		    fprintf (out, targFmt, job->node->name);
705 		    lastNode = job->node;
706 		}
707 		fprintf (out, "*** Continued\n");
708 	    }
709 	    if (! (job->flags & JOB_CONTINUING)) {
710 		JobRestart(job);
711 	    } else {
712 		Lst_AtEnd(jobs, (ClientData)job);
713 		nJobs += 1;
714 		if (! (job->flags & JOB_REMOTE)) {
715 		    nLocal += 1;
716 		}
717 		if (nJobs == maxJobs) {
718 		    jobFull = TRUE;
719 		    if (DEBUG(JOB)) {
720 			printf("Job queue is full.\n");
721 		    }
722 		}
723 	    }
724 	    fflush(out);
725 	    return;
726 	} else {
727 	    if (usePipes && job->node != lastNode) {
728 		fprintf (out, targFmt, job->node->name);
729 		lastNode = job->node;
730 	    }
731 	    fprintf (out, "*** Signal %d\n", status.w_termsig);
732 	}
733 
734 	fflush (out);
735     }
736 
737     /*
738      * Now handle the -B-mode stuff. If the beast still isn't finished,
739      * try and restart the job on the next command. If JobStart says it's
740      * ok, it's ok. If there's an error, this puppy is done.
741      */
742     if ((status.w_status == 0) &&
743 	!Lst_IsAtEnd (job->node->commands))
744     {
745 	switch (JobStart (job->node,
746 			  job->flags & JOB_IGNDOTS,
747 			  job))
748 	{
749 	    case JOB_RUNNING:
750 		done = FALSE;
751 		break;
752 	    case JOB_ERROR:
753 		done = TRUE;
754 		status.w_retcode = 1;
755 		break;
756 	    case JOB_FINISHED:
757 		/*
758 		 * If we got back a JOB_FINISHED code, JobStart has already
759 		 * called Make_Update and freed the job descriptor. We set
760 		 * done to false here to avoid fake cycles and double frees.
761 		 * JobStart needs to do the update so we can proceed up the
762 		 * graph when given the -n flag..
763 		 */
764 		done = FALSE;
765 		break;
766 	}
767     } else {
768 	done = TRUE;
769     }
770 
771 
772     if (done &&
773 	(aborting != ABORT_ERROR) &&
774 	(aborting != ABORT_INTERRUPT) &&
775 	(status.w_status == 0))
776     {
777 	/*
778 	 * As long as we aren't aborting and the job didn't return a non-zero
779 	 * status that we shouldn't ignore, we call Make_Update to update
780 	 * the parents. In addition, any saved commands for the node are placed
781 	 * on the .END target.
782 	 */
783 	if (job->tailCmds != NILLNODE) {
784 	    Lst_ForEachFrom (job->node->commands, job->tailCmds,
785 			     JobSaveCommand,
786 			     (ClientData)job->node);
787 	}
788 	job->node->made = MADE;
789 	Make_Update (job->node);
790 	free((Address)job);
791     } else if (status.w_status) {
792 	errors += 1;
793 	free((Address)job);
794     }
795 
796     while (!errors && !jobFull && !Lst_IsEmpty(stoppedJobs)) {
797 	JobRestart((Job *)Lst_DeQueue(stoppedJobs));
798     }
799 
800     /*
801      * Set aborting if any error.
802      */
803     if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) {
804 	/*
805 	 * If we found any errors in this batch of children and the -k flag
806 	 * wasn't given, we set the aborting flag so no more jobs get
807 	 * started.
808 	 */
809 	aborting = ABORT_ERROR;
810     }
811 
812     if ((aborting == ABORT_ERROR) && Job_Empty()) {
813 	/*
814 	 * If we are aborting and the job table is now empty, we finish.
815 	 */
816 	(void) unlink (tfile);
817 	Finish (errors);
818     }
819 }
820 
821 /*-
822  *-----------------------------------------------------------------------
823  * Job_Touch --
824  *	Touch the given target. Called by JobStart when the -t flag was
825  *	given
826  *
827  * Results:
828  *	None
829  *
830  * Side Effects:
831  *	The data modification of the file is changed. In addition, if the
832  *	file did not exist, it is created.
833  *-----------------------------------------------------------------------
834  */
835 void
836 Job_Touch (gn, silent)
837     GNode         *gn;	      	/* the node of the file to touch */
838     Boolean 	  silent;   	/* TRUE if should not print messages */
839 {
840     int		  streamID;   	/* ID of stream opened to do the touch */
841     struct timeval times[2];	/* Times for utimes() call */
842 
843     if (gn->type & (OP_JOIN|OP_USE|OP_EXEC|OP_OPTIONAL)) {
844 	/*
845 	 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets
846 	 * and, as such, shouldn't really be created.
847 	 */
848 	return;
849     }
850 
851     if (!silent) {
852 	printf ("touch %s\n", gn->name);
853     }
854 
855     if (noExecute) {
856 	return;
857     }
858 
859     if (gn->type & OP_ARCHV) {
860 	Arch_Touch (gn);
861     } else if (gn->type & OP_LIB) {
862 	Arch_TouchLib (gn);
863     } else {
864 	char	*file = gn->path ? gn->path : gn->name;
865 
866 	times[0].tv_sec = times[1].tv_sec = now;
867 	times[0].tv_usec = times[1].tv_usec = 0;
868 	if (utimes(file, times) < 0){
869 	    streamID = open (file, O_RDWR | O_CREAT, 0666);
870 
871 	    if (streamID >= 0) {
872 		char	c;
873 
874 		/*
875 		 * Read and write a byte to the file to change the
876 		 * modification time, then close the file.
877 		 */
878 		if (read(streamID, &c, 1) == 1) {
879 		    lseek(streamID, 0L, L_SET);
880 		    write(streamID, &c, 1);
881 		}
882 
883 		(void)close (streamID);
884 	    } else
885 		printf("*** couldn't touch %s: %s", file, strerror(errno));
886 	}
887     }
888 }
889 
890 /*-
891  *-----------------------------------------------------------------------
892  * Job_CheckCommands --
893  *	Make sure the given node has all the commands it needs.
894  *
895  * Results:
896  *	TRUE if the commands list is/was ok.
897  *
898  * Side Effects:
899  *	The node will have commands from the .DEFAULT rule added to it
900  *	if it needs them.
901  *-----------------------------------------------------------------------
902  */
903 Boolean
904 Job_CheckCommands (gn, abortProc)
905     GNode          *gn;	    	    /* The target whose commands need
906 				     * verifying */
907     void    	  (*abortProc) __P((char *, ...));
908 			/* Function to abort with message */
909 {
910     if (OP_NOP(gn->type) && Lst_IsEmpty (gn->commands) &&
911 	(gn->type & OP_LIB) == 0) {
912 	/*
913 	 * No commands. Look for .DEFAULT rule from which we might infer
914 	 * commands
915 	 */
916 	if ((DEFAULT != NILGNODE) && !Lst_IsEmpty(DEFAULT->commands)) {
917 	    /*
918 	     * Make only looks for a .DEFAULT if the node was never the
919 	     * target of an operator, so that's what we do too. If
920 	     * a .DEFAULT was given, we substitute its commands for gn's
921 	     * commands and set the IMPSRC variable to be the target's name
922 	     * The DEFAULT node acts like a transformation rule, in that
923 	     * gn also inherits any attributes or sources attached to
924 	     * .DEFAULT itself.
925 	     */
926 	    Make_HandleUse(DEFAULT, gn);
927 	    Var_Set (IMPSRC, Var_Value (TARGET, gn), gn);
928 	} else if (Dir_MTime (gn) == 0) {
929 	    /*
930 	     * The node wasn't the target of an operator we have no .DEFAULT
931 	     * rule to go on and the target doesn't already exist. There's
932 	     * nothing more we can do for this branch. If the -k flag wasn't
933 	     * given, we stop in our tracks, otherwise we just don't update
934 	     * this node's parents so they never get examined.
935 	     */
936 	    if (gn->type & OP_OPTIONAL) {
937 		printf ("make: don't know how to make %s (ignored)\n",
938 			gn->name);
939 	    } else if (keepgoing) {
940 		printf ("make: don't know how to make %s (continuing)\n",
941 			gn->name);
942 		return (FALSE);
943 	    } else {
944 		(*abortProc) ("make: don't know how to make %s. Stop",
945 			     gn->name);
946 		return(FALSE);
947 	    }
948 	}
949     }
950     return (TRUE);
951 }
952 #ifdef RMT_WILL_WATCH
953 /*-
954  *-----------------------------------------------------------------------
955  * JobLocalInput --
956  *	Handle a pipe becoming readable. Callback function for Rmt_Watch
957  *
958  * Results:
959  *	None
960  *
961  * Side Effects:
962  *	JobDoOutput is called.
963  *
964  *-----------------------------------------------------------------------
965  */
966 /*ARGSUSED*/
967 static void
968 JobLocalInput(stream, job)
969     int	    stream; 	/* Stream that's ready (ignored) */
970     Job	    *job;   	/* Job to which the stream belongs */
971 {
972     JobDoOutput(job, FALSE);
973 }
974 #endif /* RMT_WILL_WATCH */
975 
976 /*-
977  *-----------------------------------------------------------------------
978  * JobExec --
979  *	Execute the shell for the given job. Called from JobStart and
980  *	JobRestart.
981  *
982  * Results:
983  *	None.
984  *
985  * Side Effects:
986  *	A shell is executed, outputs is altered and the Job structure added
987  *	to the job table.
988  *
989  *-----------------------------------------------------------------------
990  */
991 static void
992 JobExec(job, argv)
993     Job	    	  *job; 	/* Job to execute */
994     char    	  **argv;
995 {
996     int	    	  cpid;	    	/* ID of new child */
997 
998     if (DEBUG(JOB)) {
999 	int 	  i;
1000 
1001 	printf("Running %s %sly\n", job->node->name,
1002 	       job->flags&JOB_REMOTE?"remote":"local");
1003 	printf("\tCommand: ");
1004 	for (i = 0; argv[i] != (char *)NULL; i++) {
1005 	    printf("%s ", argv[i]);
1006 	}
1007 	printf("\n");
1008     }
1009 
1010     /*
1011      * Some jobs produce no output and it's disconcerting to have
1012      * no feedback of their running (since they produce no output, the
1013      * banner with their name in it never appears). This is an attempt to
1014      * provide that feedback, even if nothing follows it.
1015      */
1016     if ((lastNode != job->node) && (job->flags & JOB_FIRST) &&
1017 	!(job->flags & JOB_SILENT))
1018     {
1019 	printf(targFmt, job->node->name);
1020 	lastNode = job->node;
1021     }
1022 
1023 #ifdef RMT_NO_EXEC
1024     if (job->flags & JOB_REMOTE) {
1025 	goto jobExecFinish;
1026     }
1027 #endif /* RMT_NO_EXEC */
1028 
1029     if ((cpid =  vfork()) == -1) {
1030 	Punt ("Cannot fork");
1031     } else if (cpid == 0) {
1032 
1033 	/*
1034 	 * Must duplicate the input stream down to the child's input and
1035 	 * reset it to the beginning (again). Since the stream was marked
1036 	 * close-on-exec, we must clear that bit in the new input.
1037 	 */
1038 	(void) dup2(fileno(job->cmdFILE), 0);
1039 	fcntl(0, F_SETFD, 0);
1040 	lseek(0, 0, L_SET);
1041 
1042 	if (usePipes) {
1043 	    /*
1044 	     * Set up the child's output to be routed through the pipe
1045 	     * we've created for it.
1046 	     */
1047 	    (void) dup2 (job->outPipe, 1);
1048 	} else {
1049 	    /*
1050 	     * We're capturing output in a file, so we duplicate the
1051 	     * descriptor to the temporary file into the standard
1052 	     * output.
1053 	     */
1054 	    (void) dup2 (job->outFd, 1);
1055 	}
1056 	/*
1057 	 * The output channels are marked close on exec. This bit was
1058 	 * duplicated by the dup2 (on some systems), so we have to clear
1059 	 * it before routing the shell's error output to the same place as
1060 	 * its standard output.
1061 	 */
1062 	fcntl(1, F_SETFD, 0);
1063 	(void) dup2 (1, 2);
1064 
1065 #ifdef USE_PGRP
1066 	/*
1067 	 * We want to switch the child into a different process family so
1068 	 * we can kill it and all its descendants in one fell swoop,
1069 	 * by killing its process family, but not commit suicide.
1070 	 */
1071 
1072 	(void) setpgrp(0, getpid());
1073 #endif USE_PGRP
1074 
1075 	(void) execv (shellPath, argv);
1076 	(void) write (2, "Could not execute shell\n",
1077 		 sizeof ("Could not execute shell"));
1078 	_exit (1);
1079     } else {
1080 	job->pid = cpid;
1081 
1082 	if (usePipes && (job->flags & JOB_FIRST) ) {
1083 	    /*
1084 	     * The first time a job is run for a node, we set the current
1085 	     * position in the buffer to the beginning and mark another
1086 	     * stream to watch in the outputs mask
1087 	     */
1088 	    job->curPos = 0;
1089 
1090 #ifdef RMT_WILL_WATCH
1091 	    Rmt_Watch(job->inPipe, JobLocalInput, job);
1092 #else
1093 	    FD_SET(job->inPipe, &outputs);
1094 #endif /* RMT_WILL_WATCH */
1095 	}
1096 
1097 	if (job->flags & JOB_REMOTE) {
1098 	    job->rmtID = 0;
1099 	} else {
1100 	    nLocal += 1;
1101 	    /*
1102 	     * XXX: Used to not happen if CUSTOMS. Why?
1103 	     */
1104 	    if (job->cmdFILE != stdout) {
1105 		fclose(job->cmdFILE);
1106 		job->cmdFILE = NULL;
1107 	    }
1108 	}
1109     }
1110 
1111 #ifdef RMT_NO_EXEC
1112 jobExecFinish:
1113 #endif
1114     /*
1115      * Now the job is actually running, add it to the table.
1116      */
1117     nJobs += 1;
1118     (void)Lst_AtEnd (jobs, (ClientData)job);
1119     if (nJobs == maxJobs) {
1120 	jobFull = TRUE;
1121     }
1122 }
1123 
1124 /*-
1125  *-----------------------------------------------------------------------
1126  * JobMakeArgv --
1127  *	Create the argv needed to execute the shell for a given job.
1128  *
1129  *
1130  * Results:
1131  *
1132  * Side Effects:
1133  *
1134  *-----------------------------------------------------------------------
1135  */
1136 static void
1137 JobMakeArgv(job, argv)
1138     Job	    	  *job;
1139     char	  **argv;
1140 {
1141     int	    	  argc;
1142     static char	  args[10]; 	/* For merged arguments */
1143 
1144     argv[0] = shellName;
1145     argc = 1;
1146 
1147     if ((commandShell->exit && (*commandShell->exit != '-')) ||
1148 	(commandShell->echo && (*commandShell->echo != '-')))
1149     {
1150 	/*
1151 	 * At least one of the flags doesn't have a minus before it, so
1152 	 * merge them together. Have to do this because the *(&(@*#*&#$#
1153 	 * Bourne shell thinks its second argument is a file to source.
1154 	 * Grrrr. Note the ten-character limitation on the combined arguments.
1155 	 */
1156 	(void)sprintf(args, "-%s%s",
1157 		      ((job->flags & JOB_IGNERR) ? "" :
1158 		       (commandShell->exit ? commandShell->exit : "")),
1159 		      ((job->flags & JOB_SILENT) ? "" :
1160 		       (commandShell->echo ? commandShell->echo : "")));
1161 
1162 	if (args[1]) {
1163 	    argv[argc] = args;
1164 	    argc++;
1165 	}
1166     } else {
1167 	if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
1168 	    argv[argc] = commandShell->exit;
1169 	    argc++;
1170 	}
1171 	if (!(job->flags & JOB_SILENT) && commandShell->echo) {
1172 	    argv[argc] = commandShell->echo;
1173 	    argc++;
1174 	}
1175     }
1176     argv[argc] = (char *)NULL;
1177 }
1178 
1179 /*-
1180  *-----------------------------------------------------------------------
1181  * JobRestart --
1182  *	Restart a job that stopped for some reason.
1183  *
1184  * Results:
1185  *	None.
1186  *
1187  * Side Effects:
1188  *	jobFull will be set if the job couldn't be run.
1189  *
1190  *-----------------------------------------------------------------------
1191  */
1192 static void
1193 JobRestart(job)
1194     Job 	  *job;    	/* Job to restart */
1195 {
1196     if (job->flags & JOB_REMIGRATE) {
1197 	if (DEBUG(JOB)) {
1198 	    printf("Remigrating %x\n", job->pid);
1199 	}
1200 	if (nLocal != maxLocal) {
1201 		/*
1202 		 * Job cannot be remigrated, but there's room on the local
1203 		 * machine, so resume the job and note that another
1204 		 * local job has started.
1205 		 */
1206 		if (DEBUG(JOB)) {
1207 		    printf("resuming on local machine\n");
1208 	        }
1209 		KILL(job->pid, SIGCONT);
1210 		nLocal +=1;
1211 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1212 	} else {
1213 		/*
1214 		 * Job cannot be restarted. Mark the table as full and
1215 		 * place the job back on the list of stopped jobs.
1216 		 */
1217 		if (DEBUG(JOB)) {
1218 		    printf("holding\n");
1219 		}
1220 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
1221 		jobFull = TRUE;
1222 		if (DEBUG(JOB)) {
1223 		    printf("Job queue is full.\n");
1224 		}
1225 		return;
1226 	}
1227 
1228 	(void)Lst_AtEnd(jobs, (ClientData)job);
1229 	nJobs += 1;
1230 	if (nJobs == maxJobs) {
1231 	    jobFull = TRUE;
1232 	    if (DEBUG(JOB)) {
1233 		printf("Job queue is full.\n");
1234 	    }
1235 	}
1236     } else if (job->flags & JOB_RESTART) {
1237 	/*
1238 	 * Set up the control arguments to the shell. This is based on the
1239 	 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
1240 	 * the 'exit' flag of the commandShell is used to cause it to exit
1241 	 * upon receiving an error. If the JOB_SILENT flag is clear, the
1242 	 * 'echo' flag of the commandShell is used to get it to start echoing
1243 	 * as soon as it starts processing commands.
1244 	 */
1245 	char	  *argv[4];
1246 
1247 	JobMakeArgv(job, argv);
1248 
1249 	if (DEBUG(JOB)) {
1250 	    printf("Restarting %s...", job->node->name);
1251 	}
1252 	if (((nLocal >= maxLocal) && ! (job->flags & JOB_SPECIAL))) {
1253 		/*
1254 		 * Can't be exported and not allowed to run locally -- put it
1255 		 * back on the hold queue and mark the table full
1256 		 */
1257 		if (DEBUG(JOB)) {
1258 		    printf("holding\n");
1259 		}
1260 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
1261 		jobFull = TRUE;
1262 		if (DEBUG(JOB)) {
1263 		    printf("Job queue is full.\n");
1264 		}
1265 		return;
1266 	} else {
1267 		/*
1268 		 * Job may be run locally.
1269 		 */
1270 		if (DEBUG(JOB)) {
1271 		    printf("running locally\n");
1272 		}
1273 		job->flags &= ~JOB_REMOTE;
1274 	}
1275 	JobExec(job, argv);
1276     } else {
1277 	/*
1278 	 * The job has stopped and needs to be restarted. Why it stopped,
1279 	 * we don't know...
1280 	 */
1281 	if (DEBUG(JOB)) {
1282 	    printf("Resuming %s...", job->node->name);
1283 	}
1284 	if (((job->flags & JOB_REMOTE) ||
1285 	     (nLocal < maxLocal) ||
1286 	     (((job->flags & JOB_SPECIAL)) &&
1287 	      (maxLocal == 0))) &&
1288 	    (nJobs != maxJobs))
1289 	{
1290 	    /*
1291 	     * If the job is remote, it's ok to resume it as long as the
1292 	     * maximum concurrency won't be exceeded. If it's local and
1293 	     * we haven't reached the local concurrency limit already (or the
1294 	     * job must be run locally and maxLocal is 0), it's also ok to
1295 	     * resume it.
1296 	     */
1297 	    Boolean error;
1298 	    extern int errno;
1299 	    union wait status;
1300 
1301 #ifdef RMT_WANTS_SIGNALS
1302 	    if (job->flags & JOB_REMOTE) {
1303 		error = !Rmt_Signal(job, SIGCONT);
1304 	    } else
1305 #endif	/* RMT_WANTS_SIGNALS */
1306 		error = (KILL(job->pid, SIGCONT) != 0);
1307 
1308 	    if (!error) {
1309 		/*
1310 		 * Make sure the user knows we've continued the beast and
1311 		 * actually put the thing in the job table.
1312 		 */
1313 		job->flags |= JOB_CONTINUING;
1314 		status.w_termsig = SIGCONT;
1315 		JobFinish(job, status);
1316 
1317 		job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
1318 		if (DEBUG(JOB)) {
1319 		    printf("done\n");
1320 		}
1321 	    } else {
1322 		Error("couldn't resume %s: %s",
1323 		    job->node->name, strerror(errno));
1324 		status.w_status = 0;
1325 		status.w_retcode = 1;
1326 		JobFinish(job, status);
1327 	    }
1328 	} else {
1329 	    /*
1330 	     * Job cannot be restarted. Mark the table as full and
1331 	     * place the job back on the list of stopped jobs.
1332 	     */
1333 	    if (DEBUG(JOB)) {
1334 		printf("table full\n");
1335 	    }
1336 	    (void)Lst_AtFront(stoppedJobs, (ClientData)job);
1337 	    jobFull = TRUE;
1338 	    if (DEBUG(JOB)) {
1339 		printf("Job queue is full.\n");
1340 	    }
1341 	}
1342     }
1343 }
1344 
1345 /*-
1346  *-----------------------------------------------------------------------
1347  * JobStart  --
1348  *	Start a target-creation process going for the target described
1349  *	by the graph node gn.
1350  *
1351  * Results:
1352  *	JOB_ERROR if there was an error in the commands, JOB_FINISHED
1353  *	if there isn't actually anything left to do for the job and
1354  *	JOB_RUNNING if the job has been started.
1355  *
1356  * Side Effects:
1357  *	A new Job node is created and added to the list of running
1358  *	jobs. PMake is forked and a child shell created.
1359  *-----------------------------------------------------------------------
1360  */
1361 static int
1362 JobStart (gn, flags, previous)
1363     GNode         *gn;	      /* target to create */
1364     short	  flags;      /* flags for the job to override normal ones.
1365 			       * e.g. JOB_SPECIAL or JOB_IGNDOTS */
1366     Job 	  *previous;  /* The previous Job structure for this node,
1367 			       * if any. */
1368 {
1369     register Job  *job;       /* new job descriptor */
1370     char	  *argv[4];   /* Argument vector to shell */
1371     static int    jobno = 0;  /* job number of catching output in a file */
1372     Boolean	  cmdsOK;     /* true if the nodes commands were all right */
1373     Boolean 	  local;      /* Set true if the job was run locally */
1374     Boolean 	  noExec;     /* Set true if we decide not to run the job */
1375 
1376     if (previous != (Job *)NULL) {
1377 	previous->flags &= ~ (JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
1378 	job = previous;
1379     } else {
1380 	job = (Job *) emalloc (sizeof (Job));
1381 	if (job == (Job *)NULL) {
1382 	    Punt("JobStart out of memory");
1383 	}
1384 	flags |= JOB_FIRST;
1385     }
1386 
1387     job->node = gn;
1388     job->tailCmds = NILLNODE;
1389 
1390     /*
1391      * Set the initial value of the flags for this job based on the global
1392      * ones and the node's attributes... Any flags supplied by the caller
1393      * are also added to the field.
1394      */
1395     job->flags = 0;
1396     if (Targ_Ignore (gn)) {
1397 	job->flags |= JOB_IGNERR;
1398     }
1399     if (Targ_Silent (gn)) {
1400 	job->flags |= JOB_SILENT;
1401     }
1402     job->flags |= flags;
1403 
1404     /*
1405      * Check the commands now so any attributes from .DEFAULT have a chance
1406      * to migrate to the node
1407      */
1408     if (job->flags & JOB_FIRST) {
1409 	cmdsOK = Job_CheckCommands(gn, Error);
1410     } else {
1411 	cmdsOK = TRUE;
1412     }
1413 
1414     /*
1415      * If the -n flag wasn't given, we open up OUR (not the child's)
1416      * temporary file to stuff commands in it. The thing is rd/wr so we don't
1417      * need to reopen it to feed it to the shell. If the -n flag *was* given,
1418      * we just set the file to be stdout. Cute, huh?
1419      */
1420     if ((gn->type & OP_MAKE) || (!noExecute && !touchFlag)) {
1421 	/*
1422 	 * We're serious here, but if the commands were bogus, we're
1423 	 * also dead...
1424 	 */
1425 	if (!cmdsOK) {
1426 	    DieHorribly();
1427 	}
1428 
1429 	job->cmdFILE = fopen (tfile, "w+");
1430 	if (job->cmdFILE == (FILE *) NULL) {
1431 	    Punt ("Could not open %s", tfile);
1432 	}
1433 	fcntl(fileno(job->cmdFILE), F_SETFD, 1);
1434 	/*
1435 	 * Send the commands to the command file, flush all its buffers then
1436 	 * rewind and remove the thing.
1437 	 */
1438 	noExec = FALSE;
1439 
1440 	/*
1441 	 * used to be backwards; replace when start doing multiple commands
1442 	 * per shell.
1443 	 */
1444 	if (compatMake) {
1445 	    /*
1446 	     * Be compatible: If this is the first time for this node,
1447 	     * verify its commands are ok and open the commands list for
1448 	     * sequential access by later invocations of JobStart.
1449 	     * Once that is done, we take the next command off the list
1450 	     * and print it to the command file. If the command was an
1451 	     * ellipsis, note that there's nothing more to execute.
1452 	     */
1453 	    if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
1454 		cmdsOK = FALSE;
1455 	    } else {
1456 		LstNode	ln = Lst_Next (gn->commands);
1457 
1458 		if ((ln == NILLNODE) ||
1459 		    JobPrintCommand ((char *)Lst_Datum (ln), job))
1460 		{
1461 		    noExec = TRUE;
1462 		    Lst_Close (gn->commands);
1463 		}
1464 		if (noExec && !(job->flags & JOB_FIRST)) {
1465 		    /*
1466 		     * If we're not going to execute anything, the job
1467 		     * is done and we need to close down the various
1468 		     * file descriptors we've opened for output, then
1469 		     * call JobDoOutput to catch the final characters or
1470 		     * send the file to the screen... Note that the i/o streams
1471 		     * are only open if this isn't the first job.
1472 		     * Note also that this could not be done in
1473 		     * Job_CatchChildren b/c it wasn't clear if there were
1474 		     * more commands to execute or not...
1475 		     */
1476 		    if (usePipes) {
1477 #ifdef RMT_WILL_WATCH
1478 			Rmt_Ignore(job->inPipe);
1479 #else
1480 			FD_CLR(job->inPipe, &outputs);
1481 #endif
1482 			if (job->outPipe != job->inPipe) {
1483 			    (void)close (job->outPipe);
1484 			}
1485 			JobDoOutput (job, TRUE);
1486 			(void)close (job->inPipe);
1487 		    } else {
1488 			(void)close (job->outFd);
1489 			JobDoOutput (job, TRUE);
1490 		    }
1491 		}
1492 	    }
1493 	} else {
1494 	    /*
1495 	     * We can do all the commands at once. hooray for sanity
1496 	     */
1497 	    numCommands = 0;
1498 	    Lst_ForEach (gn->commands, JobPrintCommand, (ClientData)job);
1499 
1500 	    /*
1501 	     * If we didn't print out any commands to the shell script,
1502 	     * there's not much point in executing the shell, is there?
1503 	     */
1504 	    if (numCommands == 0) {
1505 		noExec = TRUE;
1506 	    }
1507 	}
1508     } else if (noExecute) {
1509 	/*
1510 	 * Not executing anything -- just print all the commands to stdout
1511 	 * in one fell swoop. This will still set up job->tailCmds correctly.
1512 	 */
1513 	if (lastNode != gn) {
1514 	    printf (targFmt, gn->name);
1515 	    lastNode = gn;
1516 	}
1517 	job->cmdFILE = stdout;
1518 	/*
1519 	 * Only print the commands if they're ok, but don't die if they're
1520 	 * not -- just let the user know they're bad and keep going. It
1521 	 * doesn't do any harm in this case and may do some good.
1522 	 */
1523 	if (cmdsOK) {
1524 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
1525 	}
1526 	/*
1527 	 * Don't execute the shell, thank you.
1528 	 */
1529 	noExec = TRUE;
1530     } else {
1531 	/*
1532 	 * Just touch the target and note that no shell should be executed.
1533 	 * Set cmdFILE to stdout to make life easier. Check the commands, too,
1534 	 * but don't die if they're no good -- it does no harm to keep working
1535 	 * up the graph.
1536 	 */
1537 	job->cmdFILE = stdout;
1538     	Job_Touch (gn, job->flags&JOB_SILENT);
1539 	noExec = TRUE;
1540     }
1541 
1542     /*
1543      * If we're not supposed to execute a shell, don't.
1544      */
1545     if (noExec) {
1546 	/*
1547 	 * Unlink and close the command file if we opened one
1548 	 */
1549 	if (job->cmdFILE != stdout) {
1550 	    (void) unlink (tfile);
1551 	    fclose(job->cmdFILE);
1552 	} else {
1553 	    fflush (stdout);
1554 	}
1555 
1556 	/*
1557 	 * We only want to work our way up the graph if we aren't here because
1558 	 * the commands for the job were no good.
1559 	 */
1560 	if (cmdsOK) {
1561 	    if (aborting == 0) {
1562 		if (job->tailCmds != NILLNODE) {
1563 		    Lst_ForEachFrom(job->node->commands, job->tailCmds,
1564 				    JobSaveCommand,
1565 				    (ClientData)job->node);
1566 		}
1567 		Make_Update(job->node);
1568 	    }
1569 	    free((Address)job);
1570 	    return(JOB_FINISHED);
1571 	} else {
1572 	    free((Address)job);
1573 	    return(JOB_ERROR);
1574 	}
1575     } else {
1576 	fflush (job->cmdFILE);
1577 	(void) unlink (tfile);
1578     }
1579 
1580     /*
1581      * Set up the control arguments to the shell. This is based on the flags
1582      * set earlier for this job.
1583      */
1584     JobMakeArgv(job, argv);
1585 
1586     /*
1587      * If we're using pipes to catch output, create the pipe by which we'll
1588      * get the shell's output. If we're using files, print out that we're
1589      * starting a job and then set up its temporary-file name. This is just
1590      * tfile with two extra digits tacked on -- jobno.
1591      */
1592     if (job->flags & JOB_FIRST) {
1593 	if (usePipes) {
1594 	    int fd[2];
1595 	    (void) pipe(fd);
1596 	    job->inPipe = fd[0];
1597 	    job->outPipe = fd[1];
1598 	    (void)fcntl (job->inPipe, F_SETFD, 1);
1599 	    (void)fcntl (job->outPipe, F_SETFD, 1);
1600 	} else {
1601 	    printf ("Remaking `%s'\n", gn->name);
1602 	    fflush (stdout);
1603 	    sprintf (job->outFile, "%s%02d", tfile, jobno);
1604 	    jobno = (jobno + 1) % 100;
1605 	    job->outFd = open(job->outFile,O_WRONLY|O_CREAT|O_APPEND,0600);
1606 	    (void)fcntl (job->outFd, F_SETFD, 1);
1607 	}
1608     }
1609 
1610     local = TRUE;
1611 
1612     if (local && (((nLocal >= maxLocal) &&
1613 	 !(job->flags & JOB_SPECIAL) &&
1614 	 (maxLocal != 0))))
1615     {
1616 	/*
1617 	 * The job can only be run locally, but we've hit the limit of
1618 	 * local concurrency, so put the job on hold until some other job
1619 	 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
1620 	 * may be run locally even when the local limit has been reached
1621 	 * (e.g. when maxLocal == 0), though they will be exported if at
1622 	 * all possible.
1623 	 */
1624 	jobFull = TRUE;
1625 
1626 	if (DEBUG(JOB)) {
1627 	    printf("Can only run job locally.\n");
1628 	}
1629 	job->flags |= JOB_RESTART;
1630 	(void)Lst_AtEnd(stoppedJobs, (ClientData)job);
1631     } else {
1632 	if ((nLocal >= maxLocal) && local) {
1633 	    /*
1634 	     * If we're running this job locally as a special case (see above),
1635 	     * at least say the table is full.
1636 	     */
1637 	    jobFull = TRUE;
1638 	    if (DEBUG(JOB)) {
1639 		printf("Local job queue is full.\n");
1640 	    }
1641 	}
1642 	JobExec(job, argv);
1643     }
1644     return(JOB_RUNNING);
1645 }
1646 
1647 /*-
1648  *-----------------------------------------------------------------------
1649  * JobDoOutput  --
1650  *	This function is called at different times depending on
1651  *	whether the user has specified that output is to be collected
1652  *	via pipes or temporary files. In the former case, we are called
1653  *	whenever there is something to read on the pipe. We collect more
1654  *	output from the given job and store it in the job's outBuf. If
1655  *	this makes up a line, we print it tagged by the job's identifier,
1656  *	as necessary.
1657  *	If output has been collected in a temporary file, we open the
1658  *	file and read it line by line, transfering it to our own
1659  *	output channel until the file is empty. At which point we
1660  *	remove the temporary file.
1661  *	In both cases, however, we keep our figurative eye out for the
1662  *	'noPrint' line for the shell from which the output came. If
1663  *	we recognize a line, we don't print it. If the command is not
1664  *	alone on the line (the character after it is not \0 or \n), we
1665  *	do print whatever follows it.
1666  *
1667  * Results:
1668  *	None
1669  *
1670  * Side Effects:
1671  *	curPos may be shifted as may the contents of outBuf.
1672  *-----------------------------------------------------------------------
1673  */
1674 static void
1675 JobDoOutput (job, finish)
1676     register Job   *job;	  /* the job whose output needs printing */
1677     Boolean	   finish;	  /* TRUE if this is the last time we'll be
1678 				   * called for this job */
1679 {
1680     Boolean       gotNL = FALSE;  /* true if got a newline */
1681     register int  nr;	      	  /* number of bytes read */
1682     register int  i;	      	  /* auxiliary index into outBuf */
1683     register int  max;	      	  /* limit for i (end of current data) */
1684     int		  nRead;      	  /* (Temporary) number of bytes read */
1685 
1686     FILE      	  *oFILE;	  /* Stream pointer to shell's output file */
1687     char          inLine[132];
1688 
1689 
1690     if (usePipes) {
1691 	/*
1692 	 * Read as many bytes as will fit in the buffer.
1693 	 */
1694 end_loop:
1695 
1696 	nRead = read (job->inPipe, &job->outBuf[job->curPos],
1697 			 JOB_BUFSIZE - job->curPos);
1698 	if (nRead < 0) {
1699 	    if (DEBUG(JOB)) {
1700 		perror("JobDoOutput(piperead)");
1701 	    }
1702 	    nr = 0;
1703 	} else {
1704 	    nr = nRead;
1705 	}
1706 
1707 	/*
1708 	 * If we hit the end-of-file (the job is dead), we must flush its
1709 	 * remaining output, so pretend we read a newline if there's any
1710 	 * output remaining in the buffer.
1711 	 * Also clear the 'finish' flag so we stop looping.
1712 	 */
1713 	if ((nr == 0) && (job->curPos != 0)) {
1714 	    job->outBuf[job->curPos] = '\n';
1715 	    nr = 1;
1716 	    finish = FALSE;
1717 	} else if (nr == 0) {
1718 	    finish = FALSE;
1719 	}
1720 
1721 	/*
1722 	 * Look for the last newline in the bytes we just got. If there is
1723 	 * one, break out of the loop with 'i' as its index and gotNL set
1724 	 * TRUE.
1725 	 */
1726 	max = job->curPos + nr;
1727 	for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
1728 	    if (job->outBuf[i] == '\n') {
1729 		gotNL = TRUE;
1730 		break;
1731 	    } else if (job->outBuf[i] == '\0') {
1732 		/*
1733 		 * Why?
1734 		 */
1735 		job->outBuf[i] = ' ';
1736 	    }
1737 	}
1738 
1739 	if (!gotNL) {
1740 	    job->curPos += nr;
1741 	    if (job->curPos == JOB_BUFSIZE) {
1742 		/*
1743 		 * If we've run out of buffer space, we have no choice
1744 		 * but to print the stuff. sigh.
1745 		 */
1746 		gotNL = TRUE;
1747 		i = job->curPos;
1748 	    }
1749 	}
1750 	if (gotNL) {
1751 	    /*
1752 	     * Need to send the output to the screen. Null terminate it
1753 	     * first, overwriting the newline character if there was one.
1754 	     * So long as the line isn't one we should filter (according
1755 	     * to the shell description), we print the line, preceeded
1756 	     * by a target banner if this target isn't the same as the
1757 	     * one for which we last printed something.
1758 	     * The rest of the data in the buffer are then shifted down
1759 	     * to the start of the buffer and curPos is set accordingly.
1760 	     */
1761 	    job->outBuf[i] = '\0';
1762 	    if (i >= job->curPos) {
1763 		register char	*cp, *ecp;
1764 
1765 		cp = job->outBuf;
1766 		if (commandShell->noPrint) {
1767 		    ecp = Str_FindSubstring(job->outBuf,
1768 					    commandShell->noPrint);
1769 		    while (ecp != (char *)NULL) {
1770 			if (cp != ecp) {
1771 			    *ecp = '\0';
1772 			    if (job->node != lastNode) {
1773 				printf (targFmt, job->node->name);
1774 				lastNode = job->node;
1775 			    }
1776 			    /*
1777 			     * The only way there wouldn't be a newline after
1778 			     * this line is if it were the last in the buffer.
1779 			     * however, since the non-printable comes after it,
1780 			     * there must be a newline, so we don't print one.
1781 			     */
1782 			    printf ("%s", cp);
1783 			}
1784 			cp = ecp + commandShell->noPLen;
1785 			if (cp != &job->outBuf[i]) {
1786 			    /*
1787 			     * Still more to print, look again after skipping
1788 			     * the whitespace following the non-printable
1789 			     * command....
1790 			     */
1791 			    cp++;
1792 			    while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
1793 				cp++;
1794 			    }
1795 			    ecp = Str_FindSubstring (cp,
1796 						     commandShell->noPrint);
1797 			} else {
1798 			    break;
1799 			}
1800 		    }
1801 		}
1802 
1803 		/*
1804 		 * There's still more in that thar buffer. This time, though,
1805 		 * we know there's no newline at the end, so we add one of
1806 		 * our own free will.
1807 		 */
1808 		if (*cp != '\0') {
1809 		    if (job->node != lastNode) {
1810 			printf (targFmt, job->node->name);
1811 			lastNode = job->node;
1812 		    }
1813 		    printf ("%s\n", cp);
1814 		}
1815 
1816 		fflush (stdout);
1817 	    }
1818 	    if (i < max - 1) {
1819 		/* shift the remaining characters down */
1820 		memcpy ( job->outBuf, &job->outBuf[i + 1], max - (i + 1));
1821 		job->curPos = max - (i + 1);
1822 
1823 	    } else {
1824 		/*
1825 		 * We have written everything out, so we just start over
1826 		 * from the start of the buffer. No copying. No nothing.
1827 		 */
1828 		job->curPos = 0;
1829 	    }
1830 	}
1831 	if (finish) {
1832 	    /*
1833 	     * If the finish flag is true, we must loop until we hit
1834 	     * end-of-file on the pipe. This is guaranteed to happen eventually
1835 	     * since the other end of the pipe is now closed (we closed it
1836 	     * explicitly and the child has exited). When we do get an EOF,
1837 	     * finish will be set FALSE and we'll fall through and out.
1838 	     */
1839 	    goto end_loop;
1840 	}
1841     } else {
1842 	/*
1843 	 * We've been called to retrieve the output of the job from the
1844 	 * temporary file where it's been squirreled away. This consists of
1845 	 * opening the file, reading the output line by line, being sure not
1846 	 * to print the noPrint line for the shell we used, then close and
1847 	 * remove the temporary file. Very simple.
1848 	 *
1849 	 * Change to read in blocks and do FindSubString type things as for
1850 	 * pipes? That would allow for "@echo -n..."
1851 	 */
1852 	oFILE = fopen (job->outFile, "r");
1853 	if (oFILE != (FILE *) NULL) {
1854 	    printf ("Results of making %s:\n", job->node->name);
1855 	    while (fgets (inLine, sizeof(inLine), oFILE) != NULL) {
1856 		register char	*cp, *ecp, *endp;
1857 
1858 		cp = inLine;
1859 		endp = inLine + strlen(inLine);
1860 		if (endp[-1] == '\n') {
1861 		    *--endp = '\0';
1862 		}
1863 		if (commandShell->noPrint) {
1864 		    ecp = Str_FindSubstring(cp, commandShell->noPrint);
1865 		    while (ecp != (char *)NULL) {
1866 			if (cp != ecp) {
1867 			    *ecp = '\0';
1868 			    /*
1869 			     * The only way there wouldn't be a newline after
1870 			     * this line is if it were the last in the buffer.
1871 			     * however, since the non-printable comes after it,
1872 			     * there must be a newline, so we don't print one.
1873 			     */
1874 			    printf ("%s", cp);
1875 			}
1876 			cp = ecp + commandShell->noPLen;
1877 			if (cp != endp) {
1878 			    /*
1879 			     * Still more to print, look again after skipping
1880 			     * the whitespace following the non-printable
1881 			     * command....
1882 			     */
1883 			    cp++;
1884 			    while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
1885 				cp++;
1886 			    }
1887 			    ecp = Str_FindSubstring(cp, commandShell->noPrint);
1888 			} else {
1889 			    break;
1890 			}
1891 		    }
1892 		}
1893 
1894 		/*
1895 		 * There's still more in that thar buffer. This time, though,
1896 		 * we know there's no newline at the end, so we add one of
1897 		 * our own free will.
1898 		 */
1899 		if (*cp != '\0') {
1900 		    printf ("%s\n", cp);
1901 		}
1902 	    }
1903 	    fclose (oFILE);
1904 	    (void) unlink (job->outFile);
1905 	}
1906     }
1907     fflush(stdout);
1908 }
1909 
1910 /*-
1911  *-----------------------------------------------------------------------
1912  * Job_CatchChildren --
1913  *	Handle the exit of a child. Called from Make_Make.
1914  *
1915  * Results:
1916  *	none.
1917  *
1918  * Side Effects:
1919  *	The job descriptor is removed from the list of children.
1920  *
1921  * Notes:
1922  *	We do waits, blocking or not, according to the wisdom of our
1923  *	caller, until there are no more children to report. For each
1924  *	job, call JobFinish to finish things off. This will take care of
1925  *	putting jobs on the stoppedJobs queue.
1926  *
1927  *-----------------------------------------------------------------------
1928  */
1929 void
1930 Job_CatchChildren (block)
1931     Boolean	  block;    	/* TRUE if should block on the wait. */
1932 {
1933     int    	  pid;	    	/* pid of dead child */
1934     register Job  *job;	    	/* job descriptor for dead child */
1935     LstNode       jnode;    	/* list element for finding job */
1936     union wait	  status;   	/* Exit/termination status */
1937 
1938     /*
1939      * Don't even bother if we know there's no one around.
1940      */
1941     if (nLocal == 0) {
1942 	return;
1943     }
1944 
1945     while ((pid = wait3((int *)&status, (block?0:WNOHANG)|WUNTRACED,
1946 			(struct rusage *)0)) > 0)
1947     {
1948 	if (DEBUG(JOB))
1949 	    printf("Process %d exited or stopped.\n", pid);
1950 
1951 
1952 	jnode = Lst_Find (jobs, (ClientData)pid, JobCmpPid);
1953 
1954 	if (jnode == NILLNODE) {
1955 	    if (WIFSIGNALED(status) && (status.w_termsig == SIGCONT)) {
1956 		jnode = Lst_Find(stoppedJobs, (ClientData)pid, JobCmpPid);
1957 		if (jnode == NILLNODE) {
1958 		    Error("Resumed child (%d) not in table", pid);
1959 		    continue;
1960 		}
1961 		job = (Job *)Lst_Datum(jnode);
1962 		(void)Lst_Remove(stoppedJobs, jnode);
1963 	    } else {
1964 		Error ("Child (%d) not in table?", pid);
1965 		continue;
1966 	    }
1967 	} else {
1968 	    job = (Job *) Lst_Datum (jnode);
1969 	    (void)Lst_Remove (jobs, jnode);
1970 	    nJobs -= 1;
1971 	    if (jobFull && DEBUG(JOB)) {
1972 		printf("Job queue is no longer full.\n");
1973 	    }
1974 	    jobFull = FALSE;
1975 	    nLocal -= 1;
1976 	}
1977 
1978 	JobFinish (job, status);
1979     }
1980 }
1981 
1982 /*-
1983  *-----------------------------------------------------------------------
1984  * Job_CatchOutput --
1985  *	Catch the output from our children, if we're using
1986  *	pipes do so. Otherwise just block time until we get a
1987  *	signal (most likely a SIGCHLD) since there's no point in
1988  *	just spinning when there's nothing to do and the reaping
1989  *	of a child can wait for a while.
1990  *
1991  * Results:
1992  *	None
1993  *
1994  * Side Effects:
1995  *	Output is read from pipes if we're piping.
1996  * -----------------------------------------------------------------------
1997  */
1998 void
1999 Job_CatchOutput ()
2000 {
2001     int           	  nfds;
2002     struct timeval	  timeout;
2003     fd_set           	  readfds;
2004     register LstNode	  ln;
2005     register Job   	  *job;
2006 #ifdef RMT_WILL_WATCH
2007     int	    	  	  pnJobs;   	/* Previous nJobs */
2008 #endif
2009 
2010     fflush(stdout);
2011 #ifdef RMT_WILL_WATCH
2012     pnJobs = nJobs;
2013 
2014     /*
2015      * It is possible for us to be called with nJobs equal to 0. This happens
2016      * if all the jobs finish and a job that is stopped cannot be run
2017      * locally (eg if maxLocal is 0) and cannot be exported. The job will
2018      * be placed back on the stoppedJobs queue, Job_Empty() will return false,
2019      * Make_Run will call us again when there's nothing for which to wait.
2020      * nJobs never changes, so we loop forever. Hence the check. It could
2021      * be argued that we should sleep for a bit so as not to swamp the
2022      * exportation system with requests. Perhaps we should.
2023      *
2024      * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
2025      * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
2026      * It may use the variable nLocal to determine if it needs to call
2027      * Job_CatchChildren (if nLocal is 0, there's nothing for which to
2028      * wait...)
2029      */
2030     while (nJobs != 0 && pnJobs == nJobs) {
2031 	Rmt_Wait();
2032     }
2033 #else
2034     if (usePipes) {
2035 	readfds = outputs;
2036 	timeout.tv_sec = SEL_SEC;
2037 	timeout.tv_usec = SEL_USEC;
2038 
2039 	if ((nfds = select (FD_SETSIZE, &readfds, (fd_set *) 0, (fd_set *) 0, &timeout)) < 0)
2040 	{
2041 	    return;
2042 	} else {
2043 	    if (Lst_Open (jobs) == FAILURE) {
2044 		Punt ("Cannot open job table");
2045 	    }
2046 	    while (nfds && (ln = Lst_Next (jobs)) != NILLNODE) {
2047 		job = (Job *) Lst_Datum (ln);
2048 		if (FD_ISSET(job->inPipe, &readfds)) {
2049 		    JobDoOutput (job, FALSE);
2050 		    nfds -= 1;
2051 		}
2052 	    }
2053 	    Lst_Close (jobs);
2054 	}
2055     }
2056 #endif /* RMT_WILL_WATCH */
2057 }
2058 
2059 /*-
2060  *-----------------------------------------------------------------------
2061  * Job_Make --
2062  *	Start the creation of a target. Basically a front-end for
2063  *	JobStart used by the Make module.
2064  *
2065  * Results:
2066  *	None.
2067  *
2068  * Side Effects:
2069  *	Another job is started.
2070  *
2071  *-----------------------------------------------------------------------
2072  */
2073 void
2074 Job_Make (gn)
2075     GNode   *gn;
2076 {
2077     (void)JobStart (gn, 0, (Job *)NULL);
2078 }
2079 
2080 /*-
2081  *-----------------------------------------------------------------------
2082  * Job_Init --
2083  *	Initialize the process module
2084  *
2085  * Results:
2086  *	none
2087  *
2088  * Side Effects:
2089  *	lists and counters are initialized
2090  *-----------------------------------------------------------------------
2091  */
2092 void
2093 Job_Init (maxproc, maxlocal)
2094     int           maxproc;  /* the greatest number of jobs which may be
2095 			     * running at one time */
2096     int	    	  maxlocal; /* the greatest number of local jobs which may
2097 			     * be running at once. */
2098 {
2099     GNode         *begin;     /* node for commands to do at the very start */
2100 
2101     sprintf (tfile, "/tmp/make%05d", getpid());
2102 
2103     jobs =  	  Lst_Init (FALSE);
2104     stoppedJobs = Lst_Init(FALSE);
2105     maxJobs = 	  maxproc;
2106     maxLocal = 	  maxlocal;
2107     nJobs = 	  0;
2108     nLocal = 	  0;
2109     jobFull = 	  FALSE;
2110 
2111     aborting = 	  0;
2112     errors = 	  0;
2113 
2114     lastNode =	  NILGNODE;
2115 
2116     if (maxJobs == 1) {
2117 	/*
2118 	 * If only one job can run at a time, there's no need for a banner,
2119 	 * no is there?
2120 	 */
2121 	targFmt = "";
2122     } else {
2123 	targFmt = TARG_FMT;
2124     }
2125 
2126     if (shellPath == (char *) NULL) {
2127 	/*
2128 	 * The user didn't specify a shell to use, so we are using the
2129 	 * default one... Both the absolute path and the last component
2130 	 * must be set. The last component is taken from the 'name' field
2131 	 * of the default shell description pointed-to by commandShell.
2132 	 * All default shells are located in _PATH_DEFSHELLDIR.
2133 	 */
2134 	shellName = commandShell->name;
2135 	shellPath = str_concat (_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
2136     }
2137 
2138     if (commandShell->exit == (char *)NULL) {
2139 	commandShell->exit = "";
2140     }
2141     if (commandShell->echo == (char *)NULL) {
2142 	commandShell->echo = "";
2143     }
2144 
2145     /*
2146      * Catch the four signals that POSIX specifies if they aren't ignored.
2147      * JobPassSig will take care of calling JobInterrupt if appropriate.
2148      */
2149     if (signal (SIGINT, SIG_IGN) != SIG_IGN) {
2150 	signal (SIGINT, JobPassSig);
2151     }
2152     if (signal (SIGHUP, SIG_IGN) != SIG_IGN) {
2153 	signal (SIGHUP, JobPassSig);
2154     }
2155     if (signal (SIGQUIT, SIG_IGN) != SIG_IGN) {
2156 	signal (SIGQUIT, JobPassSig);
2157     }
2158     if (signal (SIGTERM, SIG_IGN) != SIG_IGN) {
2159 	signal (SIGTERM, JobPassSig);
2160     }
2161     /*
2162      * There are additional signals that need to be caught and passed if
2163      * either the export system wants to be told directly of signals or if
2164      * we're giving each job its own process group (since then it won't get
2165      * signals from the terminal driver as we own the terminal)
2166      */
2167 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2168     if (signal (SIGTSTP, SIG_IGN) != SIG_IGN) {
2169 	signal (SIGTSTP, JobPassSig);
2170     }
2171     if (signal (SIGTTOU, SIG_IGN) != SIG_IGN) {
2172 	signal (SIGTTOU, JobPassSig);
2173     }
2174     if (signal (SIGTTIN, SIG_IGN) != SIG_IGN) {
2175 	signal (SIGTTIN, JobPassSig);
2176     }
2177     if (signal (SIGWINCH, SIG_IGN) != SIG_IGN) {
2178 	signal (SIGWINCH, JobPassSig);
2179     }
2180 #endif
2181 
2182     begin = Targ_FindNode (".BEGIN", TARG_NOCREATE);
2183 
2184     if (begin != NILGNODE) {
2185 	JobStart (begin, JOB_SPECIAL, (Job *)0);
2186 	while (nJobs) {
2187 	    Job_CatchOutput();
2188 #ifndef RMT_WILL_WATCH
2189 	    Job_CatchChildren (!usePipes);
2190 #endif /* RMT_WILL_WATCH */
2191 	}
2192     }
2193     postCommands = Targ_FindNode (".END", TARG_CREATE);
2194 }
2195 
2196 /*-
2197  *-----------------------------------------------------------------------
2198  * Job_Full --
2199  *	See if the job table is full. It is considered full if it is OR
2200  *	if we are in the process of aborting OR if we have
2201  *	reached/exceeded our local quota. This prevents any more jobs
2202  *	from starting up.
2203  *
2204  * Results:
2205  *	TRUE if the job table is full, FALSE otherwise
2206  * Side Effects:
2207  *	None.
2208  *-----------------------------------------------------------------------
2209  */
2210 Boolean
2211 Job_Full ()
2212 {
2213     return (aborting || jobFull);
2214 }
2215 
2216 /*-
2217  *-----------------------------------------------------------------------
2218  * Job_Empty --
2219  *	See if the job table is empty.  Because the local concurrency may
2220  *	be set to 0, it is possible for the job table to become empty,
2221  *	while the list of stoppedJobs remains non-empty. In such a case,
2222  *	we want to restart as many jobs as we can.
2223  *
2224  * Results:
2225  *	TRUE if it is. FALSE if it ain't.
2226  *
2227  * Side Effects:
2228  *	None.
2229  *
2230  * -----------------------------------------------------------------------
2231  */
2232 Boolean
2233 Job_Empty ()
2234 {
2235     if (nJobs == 0) {
2236 	if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
2237 	    /*
2238 	     * The job table is obviously not full if it has no jobs in
2239 	     * it...Try and restart the stopped jobs.
2240 	     */
2241 	    jobFull = FALSE;
2242 	    while (!jobFull && !Lst_IsEmpty(stoppedJobs)) {
2243 		JobRestart((Job *)Lst_DeQueue(stoppedJobs));
2244 	    }
2245 	    return(FALSE);
2246 	} else {
2247 	    return(TRUE);
2248 	}
2249     } else {
2250 	return(FALSE);
2251     }
2252 }
2253 
2254 /*-
2255  *-----------------------------------------------------------------------
2256  * JobMatchShell --
2257  *	Find a matching shell in 'shells' given its final component.
2258  *
2259  * Results:
2260  *	A pointer to the Shell structure.
2261  *
2262  * Side Effects:
2263  *	None.
2264  *
2265  *-----------------------------------------------------------------------
2266  */
2267 static Shell *
2268 JobMatchShell (name)
2269     char	  *name;      /* Final component of shell path */
2270 {
2271     register Shell *sh;	      /* Pointer into shells table */
2272     Shell	   *match;    /* Longest-matching shell */
2273     register char *cp1,
2274 		  *cp2;
2275     char	  *eoname;
2276 
2277     eoname = name + strlen (name);
2278 
2279     match = (Shell *) NULL;
2280 
2281     for (sh = shells; sh->name != NULL; sh++) {
2282 	for (cp1 = eoname - strlen (sh->name), cp2 = sh->name;
2283 	     *cp1 != '\0' && *cp1 == *cp2;
2284 	     cp1++, cp2++) {
2285 		 continue;
2286 	}
2287 	if (*cp1 != *cp2) {
2288 	    continue;
2289 	} else if (match == (Shell *) NULL ||
2290 		   strlen (match->name) < strlen (sh->name)) {
2291 		       match = sh;
2292 	}
2293     }
2294     return (match == (Shell *) NULL ? sh : match);
2295 }
2296 
2297 /*-
2298  *-----------------------------------------------------------------------
2299  * Job_ParseShell --
2300  *	Parse a shell specification and set up commandShell, shellPath
2301  *	and shellName appropriately.
2302  *
2303  * Results:
2304  *	FAILURE if the specification was incorrect.
2305  *
2306  * Side Effects:
2307  *	commandShell points to a Shell structure (either predefined or
2308  *	created from the shell spec), shellPath is the full path of the
2309  *	shell described by commandShell, while shellName is just the
2310  *	final component of shellPath.
2311  *
2312  * Notes:
2313  *	A shell specification consists of a .SHELL target, with dependency
2314  *	operator, followed by a series of blank-separated words. Double
2315  *	quotes can be used to use blanks in words. A backslash escapes
2316  *	anything (most notably a double-quote and a space) and
2317  *	provides the functionality it does in C. Each word consists of
2318  *	keyword and value separated by an equal sign. There should be no
2319  *	unnecessary spaces in the word. The keywords are as follows:
2320  *	    name  	    Name of shell.
2321  *	    path  	    Location of shell. Overrides "name" if given
2322  *	    quiet 	    Command to turn off echoing.
2323  *	    echo  	    Command to turn echoing on
2324  *	    filter	    Result of turning off echoing that shouldn't be
2325  *	    	  	    printed.
2326  *	    echoFlag	    Flag to turn echoing on at the start
2327  *	    errFlag	    Flag to turn error checking on at the start
2328  *	    hasErrCtl	    True if shell has error checking control
2329  *	    check 	    Command to turn on error checking if hasErrCtl
2330  *	    	  	    is TRUE or template of command to echo a command
2331  *	    	  	    for which error checking is off if hasErrCtl is
2332  *	    	  	    FALSE.
2333  *	    ignore	    Command to turn off error checking if hasErrCtl
2334  *	    	  	    is TRUE or template of command to execute a
2335  *	    	  	    command so as to ignore any errors it returns if
2336  *	    	  	    hasErrCtl is FALSE.
2337  *
2338  *-----------------------------------------------------------------------
2339  */
2340 ReturnStatus
2341 Job_ParseShell (line)
2342     char	  *line;  /* The shell spec */
2343 {
2344     char    	  **words;
2345     int	    	  wordCount;
2346     register char **argv;
2347     register int  argc;
2348     char    	  *path;
2349     Shell   	  newShell;
2350     Boolean 	  fullSpec = FALSE;
2351 
2352     while (isspace (*line)) {
2353 	line++;
2354     }
2355     words = brk_string (line, &wordCount);
2356 
2357     memset ((Address)&newShell, 0, sizeof(newShell));
2358 
2359     /*
2360      * Parse the specification by keyword
2361      */
2362     for (path = (char *)NULL, argc = wordCount - 1, argv = words + 1;
2363 	 argc != 0;
2364 	 argc--, argv++) {
2365 	     if (strncmp (*argv, "path=", 5) == 0) {
2366 		 path = &argv[0][5];
2367 	     } else if (strncmp (*argv, "name=", 5) == 0) {
2368 		 newShell.name = &argv[0][5];
2369 	     } else {
2370 		 if (strncmp (*argv, "quiet=", 6) == 0) {
2371 		     newShell.echoOff = &argv[0][6];
2372 		 } else if (strncmp (*argv, "echo=", 5) == 0) {
2373 		     newShell.echoOn = &argv[0][5];
2374 		 } else if (strncmp (*argv, "filter=", 7) == 0) {
2375 		     newShell.noPrint = &argv[0][7];
2376 		     newShell.noPLen = strlen(newShell.noPrint);
2377 		 } else if (strncmp (*argv, "echoFlag=", 9) == 0) {
2378 		     newShell.echo = &argv[0][9];
2379 		 } else if (strncmp (*argv, "errFlag=", 8) == 0) {
2380 		     newShell.exit = &argv[0][8];
2381 		 } else if (strncmp (*argv, "hasErrCtl=", 10) == 0) {
2382 		     char c = argv[0][10];
2383 		     newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
2384 					    (c != 'T') && (c != 't'));
2385 		 } else if (strncmp (*argv, "check=", 6) == 0) {
2386 		     newShell.errCheck = &argv[0][6];
2387 		 } else if (strncmp (*argv, "ignore=", 7) == 0) {
2388 		     newShell.ignErr = &argv[0][7];
2389 		 } else {
2390 		     Parse_Error (PARSE_FATAL, "Unknown keyword \"%s\"",
2391 				  *argv);
2392 		     return (FAILURE);
2393 		 }
2394 		 fullSpec = TRUE;
2395 	     }
2396     }
2397 
2398     if (path == (char *)NULL) {
2399 	/*
2400 	 * If no path was given, the user wants one of the pre-defined shells,
2401 	 * yes? So we find the one s/he wants with the help of JobMatchShell
2402 	 * and set things up the right way. shellPath will be set up by
2403 	 * Job_Init.
2404 	 */
2405 	if (newShell.name == (char *)NULL) {
2406 	    Parse_Error (PARSE_FATAL, "Neither path nor name specified");
2407 	    return (FAILURE);
2408 	} else {
2409 	    commandShell = JobMatchShell (newShell.name);
2410 	    shellName = newShell.name;
2411 	}
2412     } else {
2413 	/*
2414 	 * The user provided a path. If s/he gave nothing else (fullSpec is
2415 	 * FALSE), try and find a matching shell in the ones we know of.
2416 	 * Else we just take the specification at its word and copy it
2417 	 * to a new location. In either case, we need to record the
2418 	 * path the user gave for the shell.
2419 	 */
2420 	shellPath = path;
2421 	path = strrchr (path, '/');
2422 	if (path == (char *)NULL) {
2423 	    path = shellPath;
2424 	} else {
2425 	    path += 1;
2426 	}
2427 	if (newShell.name != (char *)NULL) {
2428 	    shellName = newShell.name;
2429 	} else {
2430 	    shellName = path;
2431 	}
2432 	if (!fullSpec) {
2433 	    commandShell = JobMatchShell (shellName);
2434 	} else {
2435 	    commandShell = (Shell *) emalloc(sizeof(Shell));
2436 	    *commandShell = newShell;
2437 	}
2438     }
2439 
2440     if (commandShell->echoOn && commandShell->echoOff) {
2441 	commandShell->hasEchoCtl = TRUE;
2442     }
2443 
2444     if (!commandShell->hasErrCtl) {
2445 	if (commandShell->errCheck == (char *)NULL) {
2446 	    commandShell->errCheck = "";
2447 	}
2448 	if (commandShell->ignErr == (char *)NULL) {
2449 	    commandShell->ignErr = "%s\n";
2450 	}
2451     }
2452 
2453     /*
2454      * Do not free up the words themselves, since they might be in use by the
2455      * shell specification...
2456      */
2457     free (words);
2458     return SUCCESS;
2459 }
2460 
2461 /*-
2462  *-----------------------------------------------------------------------
2463  * JobInterrupt --
2464  *	Handle the receipt of an interrupt.
2465  *
2466  * Results:
2467  *	None
2468  *
2469  * Side Effects:
2470  *	All children are killed. Another job will be started if the
2471  *	.INTERRUPT target was given.
2472  *-----------------------------------------------------------------------
2473  */
2474 static void
2475 JobInterrupt (runINTERRUPT)
2476     int	    runINTERRUPT;   	/* Non-zero if commands for the .INTERRUPT
2477 				 * target should be executed */
2478 {
2479     LstNode 	  ln;		/* element in job table */
2480     Job           *job;	    	/* job descriptor in that element */
2481     GNode         *interrupt;	/* the node describing the .INTERRUPT target */
2482 
2483     aborting = ABORT_INTERRUPT;
2484 
2485     (void)Lst_Open (jobs);
2486     while ((ln = Lst_Next (jobs)) != NILLNODE) {
2487 	job = (Job *) Lst_Datum (ln);
2488 
2489 	if (!Targ_Precious (job->node)) {
2490 	    char  	*file = (job->node->path == (char *)NULL ?
2491 				 job->node->name :
2492 				 job->node->path);
2493 	    struct stat st;
2494 	    if (lstat(file, &st) != -1 && !S_ISDIR(st.st_mode) &&
2495 		unlink(file) != -1) {
2496 		Error ("*** %s removed", file);
2497 	    }
2498 	}
2499 #ifdef RMT_WANTS_SIGNALS
2500 	if (job->flags & JOB_REMOTE) {
2501 	    /*
2502 	     * If job is remote, let the Rmt module do the killing.
2503 	     */
2504 	    if (!Rmt_Signal(job, SIGINT)) {
2505 		/*
2506 		 * If couldn't kill the thing, finish it out now with an
2507 		 * error code, since no exit report will come in likely.
2508 		 */
2509 		union wait status;
2510 
2511 		status.w_status = 0;
2512 		status.w_retcode = 1;
2513 		JobFinish(job, status);
2514 	    }
2515 	} else if (job->pid) {
2516 	    KILL(job->pid, SIGINT);
2517 	}
2518 #else
2519 	if (job->pid) {
2520 	    KILL(job->pid, SIGINT);
2521 	}
2522 #endif /* RMT_WANTS_SIGNALS */
2523     }
2524     Lst_Close (jobs);
2525 
2526     if (runINTERRUPT && !touchFlag) {
2527 	interrupt = Targ_FindNode (".INTERRUPT", TARG_NOCREATE);
2528 	if (interrupt != NILGNODE) {
2529 	    ignoreErrors = FALSE;
2530 
2531 	    JobStart (interrupt, JOB_IGNDOTS, (Job *)0);
2532 	    while (nJobs) {
2533 		Job_CatchOutput();
2534 #ifndef RMT_WILL_WATCH
2535 		Job_CatchChildren (!usePipes);
2536 #endif /* RMT_WILL_WATCH */
2537 	    }
2538 	}
2539     }
2540     (void) unlink (tfile);
2541     exit (0);
2542 }
2543 
2544 /*
2545  *-----------------------------------------------------------------------
2546  * Job_End --
2547  *	Do final processing such as the running of the commands
2548  *	attached to the .END target.
2549  *
2550  * Results:
2551  *	Number of errors reported.
2552  *
2553  * Side Effects:
2554  *	The process' temporary file (tfile) is removed if it still
2555  *	existed.
2556  *-----------------------------------------------------------------------
2557  */
2558 int
2559 Job_End ()
2560 {
2561     if (postCommands != NILGNODE && !Lst_IsEmpty (postCommands->commands)) {
2562 	if (errors) {
2563 	    Error ("Errors reported so .END ignored");
2564 	} else {
2565 	    JobStart (postCommands, JOB_SPECIAL | JOB_IGNDOTS,
2566 		       (Job *)0);
2567 
2568 	    while (nJobs) {
2569 		Job_CatchOutput();
2570 #ifndef RMT_WILL_WATCH
2571 		Job_CatchChildren (!usePipes);
2572 #endif /* RMT_WILL_WATCH */
2573 	    }
2574 	}
2575     }
2576     (void) unlink (tfile);
2577     return(errors);
2578 }
2579 
2580 /*-
2581  *-----------------------------------------------------------------------
2582  * Job_Wait --
2583  *	Waits for all running jobs to finish and returns. Sets 'aborting'
2584  *	to ABORT_WAIT to prevent other jobs from starting.
2585  *
2586  * Results:
2587  *	None.
2588  *
2589  * Side Effects:
2590  *	Currently running jobs finish.
2591  *
2592  *-----------------------------------------------------------------------
2593  */
2594 void
2595 Job_Wait()
2596 {
2597     aborting = ABORT_WAIT;
2598     while (nJobs != 0) {
2599 	Job_CatchOutput();
2600 #ifndef RMT_WILL_WATCH
2601 	Job_CatchChildren(!usePipes);
2602 #endif /* RMT_WILL_WATCH */
2603     }
2604     aborting = 0;
2605 }
2606 
2607 /*-
2608  *-----------------------------------------------------------------------
2609  * Job_AbortAll --
2610  *	Abort all currently running jobs without handling output or anything.
2611  *	This function is to be called only in the event of a major
2612  *	error. Most definitely NOT to be called from JobInterrupt.
2613  *
2614  * Results:
2615  *	None
2616  *
2617  * Side Effects:
2618  *	All children are killed, not just the firstborn
2619  *-----------------------------------------------------------------------
2620  */
2621 void
2622 Job_AbortAll ()
2623 {
2624     LstNode           	ln;		/* element in job table */
2625     Job            	*job;	/* the job descriptor in that element */
2626     int     	  	foo;
2627 
2628     aborting = ABORT_ERROR;
2629 
2630     if (nJobs) {
2631 
2632 	(void)Lst_Open (jobs);
2633 	while ((ln = Lst_Next (jobs)) != NILLNODE) {
2634 	    job = (Job *) Lst_Datum (ln);
2635 
2636 	    /*
2637 	     * kill the child process with increasingly drastic signals to make
2638 	     * darn sure it's dead.
2639 	     */
2640 #ifdef RMT_WANTS_SIGNALS
2641 	    if (job->flags & JOB_REMOTE) {
2642 		Rmt_Signal(job, SIGINT);
2643 		Rmt_Signal(job, SIGKILL);
2644 	    } else {
2645 		KILL(job->pid, SIGINT);
2646 		KILL(job->pid, SIGKILL);
2647 	    }
2648 #else
2649 	    KILL(job->pid, SIGINT);
2650 	    KILL(job->pid, SIGKILL);
2651 #endif /* RMT_WANTS_SIGNALS */
2652 	}
2653     }
2654 
2655     /*
2656      * Catch as many children as want to report in at first, then give up
2657      */
2658     while (wait3(&foo, WNOHANG, (struct rusage *)0) > 0)
2659 	continue;
2660     (void) unlink (tfile);
2661 }
2662