job.c revision 1.56       1 /*	$NetBSD: job.c,v 1.56 2002/02/06 16:20:25 pk Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
      5  * Copyright (c) 1988, 1989 by Adam de Boor
      6  * Copyright (c) 1989 by Berkeley Softworks
      7  * All rights reserved.
      8  *
      9  * This code is derived from software contributed to Berkeley by
     10  * Adam de Boor.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *	This product includes software developed by the University of
     23  *	California, Berkeley and its contributors.
     24  * 4. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  */
     40 
     41 #ifdef MAKE_BOOTSTRAP
     42 static char rcsid[] = "$NetBSD: job.c,v 1.56 2002/02/06 16:20:25 pk Exp $";
     43 #else
     44 #include <sys/cdefs.h>
     45 #ifndef lint
     46 #if 0
     47 static char sccsid[] = "@(#)job.c	8.2 (Berkeley) 3/19/94";
     48 #else
     49 __RCSID("$NetBSD: job.c,v 1.56 2002/02/06 16:20:25 pk Exp $");
     50 #endif
     51 #endif /* not lint */
     52 #endif
     53 
     54 /*-
     55  * job.c --
     56  *	handle the creation etc. of our child processes.
     57  *
     58  * Interface:
     59  *	Job_Make  	    	Start the creation of the given target.
     60  *
     61  *	Job_CatchChildren   	Check for and handle the termination of any
     62  *	    	  	    	children. This must be called reasonably
     63  *	    	  	    	frequently to keep the whole make going at
     64  *	    	  	    	a decent clip, since job table entries aren't
     65  *	    	  	    	removed until their process is caught this way.
     66  *	    	  	    	Its single argument is TRUE if the function
     67  *	    	  	    	should block waiting for a child to terminate.
     68  *
     69  *	Job_CatchOutput	    	Print any output our children have produced.
     70  *	    	  	    	Should also be called fairly frequently to
     71  *	    	  	    	keep the user informed of what's going on.
     72  *	    	  	    	If no output is waiting, it will block for
     73  *	    	  	    	a time given by the SEL_* constants, below,
     74  *	    	  	    	or until output is ready.
     75  *
     76  *	Job_Init  	    	Called to intialize this module. in addition,
     77  *	    	  	    	any commands attached to the .BEGIN target
     78  *	    	  	    	are executed before this function returns.
     79  *	    	  	    	Hence, the makefile must have been parsed
     80  *	    	  	    	before this function is called.
     81  *
     82  *	Job_End  	    	Cleanup any memory used.
     83  *
     84  *	Job_Empty 	    	Return TRUE if the job table is completely
     85  *	    	  	    	empty.
     86  *
     87  *	Job_ParseShell	    	Given the line following a .SHELL target, parse
     88  *	    	  	    	the line as a shell specification. Returns
     89  *	    	  	    	FAILURE if the spec was incorrect.
     90  *
     91  *	Job_Finish	    	Perform any final processing which needs doing.
     92  *	    	  	    	This includes the execution of any commands
     93  *	    	  	    	which have been/were attached to the .END
     94  *	    	  	    	target. It should only be called when the
     95  *	    	  	    	job table is empty.
     96  *
     97  *	Job_AbortAll	    	Abort all currently running jobs. It doesn't
     98  *	    	  	    	handle output or do anything for the jobs,
     99  *	    	  	    	just kills them. It should only be called in
    100  *	    	  	    	an emergency, as it were.
    101  *
    102  *	Job_CheckCommands   	Verify that the commands for a target are
    103  *	    	  	    	ok. Provide them if necessary and possible.
    104  *
    105  *	Job_Touch 	    	Update a target without really updating it.
    106  *
    107  *	Job_Wait  	    	Wait for all currently-running jobs to finish.
    108  */
    109 
    110 #include <sys/types.h>
    111 #include <sys/stat.h>
    112 #include <sys/file.h>
    113 #include <sys/time.h>
    114 #include <sys/wait.h>
    115 #include <fcntl.h>
    116 #include <errno.h>
    117 #include <utime.h>
    118 #include <stdio.h>
    119 #include <string.h>
    120 #include <signal.h>
    121 #ifndef RMT_WILL_WATCH
    122 #ifndef USE_SELECT
    123 #include <poll.h>
    124 #endif
    125 #endif
    126 #include "make.h"
    127 #include "hash.h"
    128 #include "dir.h"
    129 #include "job.h"
    130 #include "pathnames.h"
    131 #include "trace.h"
    132 #ifdef REMOTE
    133 #include "rmt.h"
    134 # define STATIC
    135 #else
    136 # define STATIC static
    137 #endif
    138 
    139 /*
    140  * error handling variables
    141  */
    142 static int     	errors = 0;	    /* number of errors reported */
    143 static int    	aborting = 0;	    /* why is the make aborting? */
    144 #define ABORT_ERROR	1   	    /* Because of an error */
    145 #define ABORT_INTERRUPT	2   	    /* Because it was interrupted */
    146 #define ABORT_WAIT	3   	    /* Waiting for jobs to finish */
    147 
    148 /*
    149  * XXX: Avoid SunOS bug... FILENO() is fp->_file, and file
    150  * is a char! So when we go above 127 we turn negative!
    151  */
    152 #define FILENO(a) ((unsigned) fileno(a))
    153 
    154 /*
    155  * post-make command processing. The node postCommands is really just the
    156  * .END target but we keep it around to avoid having to search for it
    157  * all the time.
    158  */
    159 static GNode   	  *postCommands;    /* node containing commands to execute when
    160 				     * everything else is done */
    161 static int     	  numCommands; 	    /* The number of commands actually printed
    162 				     * for a target. Should this number be
    163 				     * 0, no shell will be executed. */
    164 
    165 /*
    166  * Return values from JobStart.
    167  */
    168 #define JOB_RUNNING	0   	/* Job is running */
    169 #define JOB_ERROR 	1   	/* Error in starting the job */
    170 #define JOB_FINISHED	2   	/* The job is already finished */
    171 #define JOB_STOPPED	3   	/* The job is stopped */
    172 
    173 
    174 
    175 /*
    176  * Descriptions for various shells.
    177  */
    178 static Shell    shells[] = {
    179     /*
    180      * CSH description. The csh can do echo control by playing
    181      * with the setting of the 'echo' shell variable. Sadly,
    182      * however, it is unable to do error control nicely.
    183      */
    184 {
    185     "csh",
    186     TRUE, "unset verbose", "set verbose", "unset verbose", 10,
    187     FALSE, "echo \"%s\"\n", "csh -c \"%s || exit 0\"",
    188     "v", "e",
    189 },
    190     /*
    191      * SH description. Echo control is also possible and, under
    192      * sun UNIX anyway, one can even control error checking.
    193      */
    194 {
    195     "sh",
    196     TRUE, "set -", "set -v", "set -", 5,
    197     TRUE, "set -e", "set +e",
    198 #ifdef OLDBOURNESHELL
    199     FALSE, "echo \"%s\"\n", "sh -c '%s || exit 0'\n",
    200 #endif
    201 #ifdef __NetBSD__
    202     "vq",
    203 #else
    204     "v",
    205 #endif
    206     "e",
    207 },
    208     /*
    209      * UNKNOWN.
    210      */
    211 {
    212     (char *) 0,
    213     FALSE, (char *) 0, (char *) 0, (char *) 0, 0,
    214     FALSE, (char *) 0, (char *) 0,
    215     (char *) 0, (char *) 0,
    216 }
    217 };
    218 static Shell 	*commandShell = &shells[DEFSHELL];/* this is the shell to
    219 						   * which we pass all
    220 						   * commands in the Makefile.
    221 						   * It is set by the
    222 						   * Job_ParseShell function */
    223 static char   	*shellPath = NULL,		  /* full pathname of
    224 						   * executable image */
    225                	*shellName = NULL,	      	  /* last component of shell */
    226 		*shellArgv = NULL;		  /* Custom shell args */
    227 
    228 
    229 static int  	maxJobs;    	/* The most children we can run at once */
    230 static int  	maxLocal;    	/* The most local ones we can have */
    231 STATIC int     	nJobs;	    	/* The number of children currently running */
    232 STATIC int	nLocal;    	/* The number of local children */
    233 STATIC Lst     	jobs;		/* The structures that describe them */
    234 static Boolean	wantToken;	/* we want a token */
    235 
    236 /*
    237  * Set of descriptors of pipes connected to
    238  * the output channels of children
    239  */
    240 #ifndef RMT_WILL_WATCH
    241 #ifdef USE_SELECT
    242 static fd_set  	outputs;
    243 #else
    244 static struct pollfd *fds = NULL;
    245 static Job **jobfds = NULL;
    246 static int nfds = 0;
    247 static int maxfds = 0;
    248 static void watchfd __P((Job *));
    249 static void clearfd __P((Job *));
    250 static int readyfd __P((Job *));
    251 #define JBSTART 256
    252 #define JBFACTOR 2
    253 #endif
    254 #endif
    255 
    256 STATIC GNode   	*lastNode;	/* The node for which output was most recently
    257 				 * produced. */
    258 STATIC char    	*targFmt;   	/* Format string to use to head output from a
    259 				 * job when it's not the most-recent job heard
    260 				 * from */
    261 static Job tokenWaitJob;	/* token wait pseudo-job */
    262 int	job_pipe[2] = { -1, -1 }; /* job server pipes. */
    263 
    264 #ifdef REMOTE
    265 # define TARG_FMT  "--- %s at %s ---\n" /* Default format */
    266 # define MESSAGE(fp, gn) \
    267 	(void) fprintf(fp, targFmt, gn->name, gn->rem.hname)
    268 #else
    269 # define TARG_FMT  "--- %s ---\n" /* Default format */
    270 # define MESSAGE(fp, gn) \
    271 	(void) fprintf(fp, targFmt, gn->name)
    272 #endif
    273 
    274 /*
    275  * When JobStart attempts to run a job remotely but can't, and isn't allowed
    276  * to run the job locally, or when Job_CatchChildren detects a job that has
    277  * been migrated home, the job is placed on the stoppedJobs queue to be run
    278  * when the next job finishes.
    279  */
    280 STATIC Lst	stoppedJobs;	/* Lst of Job structures describing
    281 				 * jobs that were stopped due to concurrency
    282 				 * limits or migration home */
    283 
    284 
    285 #if defined(USE_PGRP) && defined(SYSV)
    286 # define KILL(pid, sig)		kill(-(pid), (sig))
    287 #else
    288 # if defined(USE_PGRP)
    289 #  define KILL(pid, sig)	killpg((pid), (sig))
    290 # else
    291 #  define KILL(pid, sig)	kill((pid), (sig))
    292 # endif
    293 #endif
    294 
    295 /*
    296  * Grmpf... There is no way to set bits of the wait structure
    297  * anymore with the stupid W*() macros. I liked the union wait
    298  * stuff much more. So, we devise our own macros... This is
    299  * really ugly, use dramamine sparingly. You have been warned.
    300  */
    301 #ifndef W_STOPCODE
    302 #define W_STOPCODE(sig) (((sig) << 8) | 0177)
    303 #endif
    304 #ifndef W_EXITCODE
    305 #define W_EXITCODE(ret, sig) ((ret << 8) | (sig))
    306 #endif
    307 
    308 static int JobCondPassSig __P((ClientData, ClientData));
    309 static void JobPassSig __P((int));
    310 static void JobIgnoreSig __P((int));
    311 #ifdef USE_PGRP
    312 static void JobContinueSig __P((int));
    313 #endif
    314 static int JobCmpPid __P((ClientData, ClientData));
    315 static int JobPrintCommand __P((ClientData, ClientData));
    316 static int JobSaveCommand __P((ClientData, ClientData));
    317 static void JobClose __P((Job *));
    318 #ifdef REMOTE
    319 static int JobCmpRmtID __P((Job *, int));
    320 # ifdef RMT_WILL_WATCH
    321 static void JobLocalInput __P((int, Job *));
    322 # endif
    323 #else
    324 static void JobFinish __P((Job *, int *));
    325 static void JobExec __P((Job *, char **));
    326 #endif
    327 static void JobMakeArgv __P((Job *, char **));
    328 static void JobRestart __P((Job *));
    329 static int JobStart __P((GNode *, int, Job *));
    330 static char *JobOutput __P((Job *, char *, char *, int));
    331 static void JobDoOutput __P((Job *, Boolean));
    332 static Shell *JobMatchShell __P((char *));
    333 static void JobInterrupt __P((int, int));
    334 static void JobRestartJobs __P((void));
    335 static void JobTokenAdd __P((void));
    336 
    337 /*-
    338  *-----------------------------------------------------------------------
    339  * JobCondPassSig --
    340  *	Pass a signal to a job if the job is remote or if USE_PGRP
    341  *	is defined.
    342  *
    343  * Results:
    344  *	=== 0
    345  *
    346  * Side Effects:
    347  *	None, except the job may bite it.
    348  *
    349  *-----------------------------------------------------------------------
    350  */
    351 static int
    352 JobCondPassSig(jobp, signop)
    353     ClientData	    	jobp;	    /* Job to biff */
    354     ClientData	    	signop;	    /* Signal to send it */
    355 {
    356     Job	*job = (Job *) jobp;
    357     int	signo = *(int *) signop;
    358 #ifdef RMT_WANTS_SIGNALS
    359     if (job->flags & JOB_REMOTE) {
    360 	(void) Rmt_Signal(job, signo);
    361     } else {
    362 	KILL(job->pid, signo);
    363     }
    364 #else
    365     /*
    366      * Assume that sending the signal to job->pid will signal any remote
    367      * job as well.
    368      */
    369     if (DEBUG(JOB)) {
    370 	(void) fprintf(stdout,
    371 		       "JobCondPassSig passing signal %d to child %d.\n",
    372 		       signo, job->pid);
    373 	(void) fflush(stdout);
    374     }
    375     KILL(job->pid, signo);
    376 #endif
    377     return 0;
    378 }
    379 
    380 /*-
    381  *-----------------------------------------------------------------------
    382  * JobIgnoreSig --
    383  *	No-op signal handler so we wake up from poll.
    384  *
    385  * Results:
    386  *	None.
    387  *
    388  * Side Effects:
    389  *	None.
    390  *
    391  *-----------------------------------------------------------------------
    392  */
    393 static void
    394 JobIgnoreSig(signo)
    395     int	    signo;	/* The signal number we've received */
    396 {
    397 	/*
    398 	 * Do nothing.  The mere fact that we've been called will cause
    399 	 * poll/select in Job_CatchOutput() to return early.
    400 	 */
    401 }
    402 
    403 
    404 #ifdef USE_PGRP
    405 /*-
    406  *-----------------------------------------------------------------------
    407  * JobContinueSig --
    408  *	Resume all stopped jobs.
    409  *
    410  * Results:
    411  *	None.
    412  *
    413  * Side Effects:
    414  *	Jobs start running again.
    415  *
    416  *-----------------------------------------------------------------------
    417  */
    418 static void
    419 JobContinueSig(signo)
    420     int	    signo;	/* The signal number we've received */
    421 {
    422     if (signal(SIGTSTP, SIG_IGN) != SIG_IGN) {
    423 	(void) signal(SIGTSTP, JobPassSig);
    424     }
    425     JobRestartJobs();
    426 }
    427 #endif
    428 
    429 /*-
    430  *-----------------------------------------------------------------------
    431  * JobPassSig --
    432  *	Pass a signal on to all remote jobs and to all local jobs if
    433  *	USE_PGRP is defined, then die ourselves.
    434  *
    435  * Results:
    436  *	None.
    437  *
    438  * Side Effects:
    439  *	We die by the same signal.
    440  *
    441  *-----------------------------------------------------------------------
    442  */
    443 static void
    444 JobPassSig(signo)
    445     int	    signo;	/* The signal number we've received */
    446 {
    447     sigset_t nmask, omask;
    448     struct sigaction act;
    449     int sigcont;
    450 
    451     if (DEBUG(JOB)) {
    452 	(void) fprintf(stdout, "JobPassSig(%d) called.\n", signo);
    453 	(void) fflush(stdout);
    454     }
    455     Lst_ForEach(jobs, JobCondPassSig, (ClientData) &signo);
    456 
    457     /*
    458      * Deal with proper cleanup based on the signal received. We only run
    459      * the .INTERRUPT target if the signal was in fact an interrupt. The other
    460      * three termination signals are more of a "get out *now*" command.
    461      */
    462     if (signo == SIGINT) {
    463 	JobInterrupt(TRUE, signo);
    464     } else if ((signo == SIGHUP) || (signo == SIGTERM) || (signo == SIGQUIT)) {
    465 	JobInterrupt(FALSE, signo);
    466     }
    467 
    468     /*
    469      * Leave gracefully if SIGQUIT, rather than core dumping.
    470      */
    471     if (signo == SIGQUIT) {
    472 	Finish(0);
    473     }
    474 
    475     if (signo == SIGTSTP) {
    476 	Job_CatchChildren(FALSE);
    477     }
    478     /*
    479      * Send ourselves the signal now we've given the message to everyone else.
    480      * Note we block everything else possible while we're getting the signal.
    481      * This ensures that all our jobs get continued when we wake up before
    482      * we take any other signal.
    483      */
    484     sigfillset(&nmask);
    485     sigprocmask(SIG_SETMASK, &nmask, &omask);
    486     act.sa_handler = SIG_DFL;
    487     sigemptyset(&act.sa_mask);
    488     act.sa_flags = 0;
    489     sigaction(signo, &act, NULL);
    490 
    491     if (DEBUG(JOB)) {
    492 	(void) fprintf(stdout,
    493 		       "JobPassSig passing signal %d to self.\n", signo);
    494 	(void) fflush(stdout);
    495     }
    496 
    497     (void) kill(getpid(), signo);
    498     if (signo != SIGTSTP) {
    499 	sigcont = SIGCONT;
    500 	Lst_ForEach(jobs, JobCondPassSig, (ClientData) &sigcont);
    501     }
    502 
    503     (void) sigprocmask(SIG_SETMASK, &omask, NULL);
    504     sigprocmask(SIG_SETMASK, &omask, NULL);
    505     if (signo != SIGCONT && signo != SIGTSTP) {
    506 	act.sa_handler = JobPassSig;
    507 	sigaction(sigcont, &act, NULL);
    508     }
    509 }
    510 
    511 /*-
    512  *-----------------------------------------------------------------------
    513  * JobCmpPid  --
    514  *	Compare the pid of the job with the given pid and return 0 if they
    515  *	are equal. This function is called from Job_CatchChildren via
    516  *	Lst_Find to find the job descriptor of the finished job.
    517  *
    518  * Results:
    519  *	0 if the pid's match
    520  *
    521  * Side Effects:
    522  *	None
    523  *-----------------------------------------------------------------------
    524  */
    525 static int
    526 JobCmpPid(job, pid)
    527     ClientData        job;	/* job to examine */
    528     ClientData        pid;	/* process id desired */
    529 {
    530     return *(int *) pid - ((Job *) job)->pid;
    531 }
    532 
    533 #ifdef REMOTE
    534 /*-
    535  *-----------------------------------------------------------------------
    536  * JobCmpRmtID  --
    537  *	Compare the rmtID of the job with the given rmtID and return 0 if they
    538  *	are equal.
    539  *
    540  * Results:
    541  *	0 if the rmtID's match
    542  *
    543  * Side Effects:
    544  *	None.
    545  *-----------------------------------------------------------------------
    546  */
    547 static int
    548 JobCmpRmtID(job, rmtID)
    549     ClientData      job;	/* job to examine */
    550     ClientData      rmtID;	/* remote id desired */
    551 {
    552     return(*(int *) rmtID - *(int *) job->rmtID);
    553 }
    554 #endif
    555 
    556 /*-
    557  *-----------------------------------------------------------------------
    558  * JobPrintCommand  --
    559  *	Put out another command for the given job. If the command starts
    560  *	with an @ or a - we process it specially. In the former case,
    561  *	so long as the -s and -n flags weren't given to make, we stick
    562  *	a shell-specific echoOff command in the script. In the latter,
    563  *	we ignore errors for the entire job, unless the shell has error
    564  *	control.
    565  *	If the command is just "..." we take all future commands for this
    566  *	job to be commands to be executed once the entire graph has been
    567  *	made and return non-zero to signal that the end of the commands
    568  *	was reached. These commands are later attached to the postCommands
    569  *	node and executed by Job_End when all things are done.
    570  *	This function is called from JobStart via Lst_ForEach.
    571  *
    572  * Results:
    573  *	Always 0, unless the command was "..."
    574  *
    575  * Side Effects:
    576  *	If the command begins with a '-' and the shell has no error control,
    577  *	the JOB_IGNERR flag is set in the job descriptor.
    578  *	If the command is "..." and we're not ignoring such things,
    579  *	tailCmds is set to the successor node of the cmd.
    580  *	numCommands is incremented if the command is actually printed.
    581  *-----------------------------------------------------------------------
    582  */
    583 static int
    584 JobPrintCommand(cmdp, jobp)
    585     ClientData    cmdp;	    	    /* command string to print */
    586     ClientData    jobp;	    	    /* job for which to print it */
    587 {
    588     Boolean	  noSpecials;	    /* true if we shouldn't worry about
    589 				     * inserting special commands into
    590 				     * the input stream. */
    591     Boolean       shutUp = FALSE;   /* true if we put a no echo command
    592 				     * into the command file */
    593     Boolean	  errOff = FALSE;   /* true if we turned error checking
    594 				     * off before printing the command
    595 				     * and need to turn it back on */
    596     char       	  *cmdTemplate;	    /* Template to use when printing the
    597 				     * command */
    598     char    	  *cmdStart;	    /* Start of expanded command */
    599     char     	  *cmd = (char *) cmdp;
    600     Job           *job = (Job *) jobp;
    601     char	*cp;
    602 
    603     noSpecials = NoExecute(job->node);
    604 
    605     if (strcmp(cmd, "...") == 0) {
    606 	job->node->type |= OP_SAVE_CMDS;
    607 	if ((job->flags & JOB_IGNDOTS) == 0) {
    608 	    job->tailCmds = Lst_Succ(Lst_Member(job->node->commands,
    609 						(ClientData)cmd));
    610 	    return 1;
    611 	}
    612 	return 0;
    613     }
    614 
    615 #define DBPRINTF(fmt, arg) if (DEBUG(JOB)) {	\
    616 	(void) fprintf(stdout, fmt, arg); 	\
    617 	(void) fflush(stdout); 			\
    618     }						\
    619    (void) fprintf(job->cmdFILE, fmt, arg);	\
    620    (void) fflush(job->cmdFILE);
    621 
    622     numCommands += 1;
    623 
    624     cmdStart = cmd = Var_Subst(NULL, cmd, job->node, FALSE);
    625 
    626     cmdTemplate = "%s\n";
    627 
    628     /*
    629      * Check for leading @' and -'s to control echoing and error checking.
    630      */
    631     while (*cmd == '@' || *cmd == '-') {
    632 	if (*cmd == '@') {
    633 	    shutUp = TRUE;
    634 	} else {
    635 	    errOff = TRUE;
    636 	}
    637 	cmd++;
    638     }
    639 
    640     while (isspace((unsigned char) *cmd))
    641 	cmd++;
    642 
    643     if (shutUp) {
    644 	if (!(job->flags & JOB_SILENT) && !noSpecials &&
    645 	    commandShell->hasEchoCtl) {
    646 		DBPRINTF("%s\n", commandShell->echoOff);
    647 	} else {
    648 	    shutUp = FALSE;
    649 	}
    650     }
    651 
    652     if (errOff) {
    653 	if ( !(job->flags & JOB_IGNERR) && !noSpecials) {
    654 	    if (commandShell->hasErrCtl) {
    655 		/*
    656 		 * we don't want the error-control commands showing
    657 		 * up either, so we turn off echoing while executing
    658 		 * them. We could put another field in the shell
    659 		 * structure to tell JobDoOutput to look for this
    660 		 * string too, but why make it any more complex than
    661 		 * it already is?
    662 		 */
    663 		if (!(job->flags & JOB_SILENT) && !shutUp &&
    664 		    commandShell->hasEchoCtl) {
    665 			DBPRINTF("%s\n", commandShell->echoOff);
    666 			DBPRINTF("%s\n", commandShell->ignErr);
    667 			DBPRINTF("%s\n", commandShell->echoOn);
    668 		} else {
    669 		    DBPRINTF("%s\n", commandShell->ignErr);
    670 		}
    671 	    } else if (commandShell->ignErr &&
    672 		      (*commandShell->ignErr != '\0'))
    673 	    {
    674 		/*
    675 		 * The shell has no error control, so we need to be
    676 		 * weird to get it to ignore any errors from the command.
    677 		 * If echoing is turned on, we turn it off and use the
    678 		 * errCheck template to echo the command. Leave echoing
    679 		 * off so the user doesn't see the weirdness we go through
    680 		 * to ignore errors. Set cmdTemplate to use the weirdness
    681 		 * instead of the simple "%s\n" template.
    682 		 */
    683 		if (!(job->flags & JOB_SILENT) && !shutUp &&
    684 		    commandShell->hasEchoCtl) {
    685 			DBPRINTF("%s\n", commandShell->echoOff);
    686 			DBPRINTF(commandShell->errCheck, cmd);
    687 			shutUp = TRUE;
    688 		}
    689 		cmdTemplate = commandShell->ignErr;
    690 		/*
    691 		 * The error ignoration (hee hee) is already taken care
    692 		 * of by the ignErr template, so pretend error checking
    693 		 * is still on.
    694 		 */
    695 		errOff = FALSE;
    696 	    } else {
    697 		errOff = FALSE;
    698 	    }
    699 	} else {
    700 	    errOff = FALSE;
    701 	}
    702     }
    703 
    704     if (DEBUG(SHELL) && strcmp(shellName, "sh") == 0 &&
    705 	(job->flags & JOB_TRACED) == 0) {
    706 	    DBPRINTF("set -%s\n", "x");
    707 	    job->flags |= JOB_TRACED;
    708     }
    709 
    710     if ((cp = Check_Cwd_Cmd(cmd)) != NULL) {
    711 	    DBPRINTF("test -d %s && ", cp);
    712 	    DBPRINTF("cd %s; ", cp);
    713     }
    714     DBPRINTF(cmdTemplate, cmd);
    715     free(cmdStart);
    716 
    717     if (errOff) {
    718 	/*
    719 	 * If echoing is already off, there's no point in issuing the
    720 	 * echoOff command. Otherwise we issue it and pretend it was on
    721 	 * for the whole command...
    722 	 */
    723 	if (!shutUp && !(job->flags & JOB_SILENT) && commandShell->hasEchoCtl){
    724 	    DBPRINTF("%s\n", commandShell->echoOff);
    725 	    shutUp = TRUE;
    726 	}
    727 	DBPRINTF("%s\n", commandShell->errCheck);
    728     }
    729     if (shutUp) {
    730 	DBPRINTF("%s\n", commandShell->echoOn);
    731     }
    732     return 0;
    733 }
    734 
    735 /*-
    736  *-----------------------------------------------------------------------
    737  * JobSaveCommand --
    738  *	Save a command to be executed when everything else is done.
    739  *	Callback function for JobFinish...
    740  *
    741  * Results:
    742  *	Always returns 0
    743  *
    744  * Side Effects:
    745  *	The command is tacked onto the end of postCommands's commands list.
    746  *
    747  *-----------------------------------------------------------------------
    748  */
    749 static int
    750 JobSaveCommand(cmd, gn)
    751     ClientData   cmd;
    752     ClientData   gn;
    753 {
    754     cmd = (ClientData) Var_Subst(NULL, (char *) cmd, (GNode *) gn, FALSE);
    755     (void) Lst_AtEnd(postCommands->commands, cmd);
    756     return(0);
    757 }
    758 
    759 
    760 /*-
    761  *-----------------------------------------------------------------------
    762  * JobClose --
    763  *	Called to close both input and output pipes when a job is finished.
    764  *
    765  * Results:
    766  *	Nada
    767  *
    768  * Side Effects:
    769  *	The file descriptors associated with the job are closed.
    770  *
    771  *-----------------------------------------------------------------------
    772  */
    773 static void
    774 JobClose(job)
    775     Job *job;
    776 {
    777     if (usePipes && (job->flags & JOB_FIRST)) {
    778 #ifdef RMT_WILL_WATCH
    779 	Rmt_Ignore(job->inPipe);
    780 #else
    781 #ifdef USE_SELECT
    782 	FD_CLR(job->inPipe, &outputs);
    783 #else
    784 	clearfd(job);
    785 #endif
    786 #endif
    787 	if (job->outPipe != job->inPipe) {
    788 	   (void) close(job->outPipe);
    789 	}
    790 	JobDoOutput(job, TRUE);
    791 	(void) close(job->inPipe);
    792     } else {
    793 	(void) close(job->outFd);
    794 	JobDoOutput(job, TRUE);
    795     }
    796 }
    797 
    798 /*-
    799  *-----------------------------------------------------------------------
    800  * JobFinish  --
    801  *	Do final processing for the given job including updating
    802  *	parents and starting new jobs as available/necessary. Note
    803  *	that we pay no attention to the JOB_IGNERR flag here.
    804  *	This is because when we're called because of a noexecute flag
    805  *	or something, jstat.w_status is 0 and when called from
    806  *	Job_CatchChildren, the status is zeroed if it s/b ignored.
    807  *
    808  * Results:
    809  *	None
    810  *
    811  * Side Effects:
    812  *	Some nodes may be put on the toBeMade queue.
    813  *	Final commands for the job are placed on postCommands.
    814  *
    815  *	If we got an error and are aborting (aborting == ABORT_ERROR) and
    816  *	the job list is now empty, we are done for the day.
    817  *	If we recognized an error (errors !=0), we set the aborting flag
    818  *	to ABORT_ERROR so no more jobs will be started.
    819  *-----------------------------------------------------------------------
    820  */
    821 /*ARGSUSED*/
    822 static void
    823 JobFinish(job, status)
    824     Job         *job;	      	  /* job to finish */
    825     int	  	*status;     	  /* sub-why job went away */
    826 {
    827     Boolean 	 done;
    828 
    829     if ((WIFEXITED(*status) &&
    830 	 (((WEXITSTATUS(*status) != 0) && !(job->flags & JOB_IGNERR)))) ||
    831 	WIFSIGNALED(*status))
    832     {
    833 	/*
    834 	 * If it exited non-zero and either we're doing things our
    835 	 * way or we're not ignoring errors, the job is finished.
    836 	 * Similarly, if the shell died because of a signal
    837 	 * the job is also finished. In these
    838 	 * cases, finish out the job's output before printing the exit
    839 	 * status...
    840 	 */
    841 #ifdef REMOTE
    842 	KILL(job->pid, SIGCONT);
    843 #endif
    844 	JobClose(job);
    845 	if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
    846 	   (void) fclose(job->cmdFILE);
    847 	   job->cmdFILE = NULL;
    848 	}
    849 	done = TRUE;
    850 #ifdef REMOTE
    851 	if (job->flags & JOB_REMOTE)
    852 	    Rmt_Done(job->rmtID, job->node);
    853 #endif
    854     } else if (WIFEXITED(*status)) {
    855 	/*
    856 	 * Deal with ignored errors in -B mode. We need to print a message
    857 	 * telling of the ignored error as well as setting status.w_status
    858 	 * to 0 so the next command gets run. To do this, we set done to be
    859 	 * TRUE if in -B mode and the job exited non-zero.
    860 	 */
    861 	done = WEXITSTATUS(*status) != 0;
    862 	/*
    863 	 * Old comment said: "Note we don't
    864 	 * want to close down any of the streams until we know we're at the
    865 	 * end."
    866 	 * But we do. Otherwise when are we going to print the rest of the
    867 	 * stuff?
    868 	 */
    869 	JobClose(job);
    870 #ifdef REMOTE
    871 	if (job->flags & JOB_REMOTE)
    872 	    Rmt_Done(job->rmtID, job->node);
    873 #endif /* REMOTE */
    874     } else {
    875 	/*
    876 	 * No need to close things down or anything.
    877 	 */
    878 	done = FALSE;
    879     }
    880 
    881     if (done ||
    882 	WIFSTOPPED(*status) ||
    883 	(WIFSIGNALED(*status) && (WTERMSIG(*status) == SIGCONT)))
    884     {
    885 	FILE	  *out;
    886 
    887 	if (compatMake && !usePipes && (job->flags & JOB_IGNERR)) {
    888 	    /*
    889 	     * If output is going to a file and this job is ignoring
    890 	     * errors, arrange to have the exit status sent to the
    891 	     * output file as well.
    892 	     */
    893 	    out = fdopen(job->outFd, "w");
    894 	    if (out == NULL)
    895 		Punt("Cannot fdopen");
    896 	} else {
    897 	    out = stdout;
    898 	}
    899 
    900 	if (WIFEXITED(*status)) {
    901 	    if (DEBUG(JOB)) {
    902 		(void) fprintf(stdout, "Process %d exited.\n", job->pid);
    903 		(void) fflush(stdout);
    904 	    }
    905 	    if (WEXITSTATUS(*status) != 0) {
    906 		if (usePipes && job->node != lastNode) {
    907 		    MESSAGE(out, job->node);
    908 		    lastNode = job->node;
    909 		}
    910 		(void) fprintf(out, "*** Error code %d%s\n",
    911 			       WEXITSTATUS(*status),
    912 			       (job->flags & JOB_IGNERR) ? "(ignored)" : "");
    913 
    914 		if (job->flags & JOB_IGNERR) {
    915 		    *status = 0;
    916 		}
    917 	    } else if (DEBUG(JOB)) {
    918 		if (usePipes && job->node != lastNode) {
    919 		    MESSAGE(out, job->node);
    920 		    lastNode = job->node;
    921 		}
    922 		(void) fprintf(out, "*** Completed successfully\n");
    923 	    }
    924 	} else if (WIFSTOPPED(*status) && WSTOPSIG(*status) != SIGCONT) {
    925 	    if (DEBUG(JOB)) {
    926 		(void) fprintf(stdout, "Process %d stopped.\n", job->pid);
    927 		(void) fflush(stdout);
    928 	    }
    929 	    if (usePipes && job->node != lastNode) {
    930 		MESSAGE(out, job->node);
    931 		lastNode = job->node;
    932 	    }
    933 	    if (!(job->flags & JOB_REMIGRATE)) {
    934 		switch (WSTOPSIG(*status)) {
    935 		case SIGTSTP:
    936 		    (void) fprintf(out, "*** Suspended\n");
    937 		    break;
    938 		case SIGSTOP:
    939 		    (void) fprintf(out, "*** Stopped\n");
    940 		    break;
    941 		default:
    942 		    (void) fprintf(out, "*** Stopped -- signal %d\n",
    943 			WSTOPSIG(*status));
    944 		}
    945 	    }
    946 	    job->flags |= JOB_RESUME;
    947 	    (void)Lst_AtEnd(stoppedJobs, (ClientData)job);
    948 #ifdef REMOTE
    949 	    if (job->flags & JOB_REMIGRATE)
    950 		JobRestart(job);
    951 #endif
    952 	    (void) fflush(out);
    953 	    return;
    954 	} else if (WIFSTOPPED(*status) &&  WSTOPSIG(*status) == SIGCONT) {
    955 	    /*
    956 	     * If the beastie has continued, shift the Job from the stopped
    957 	     * list to the running one (or re-stop it if concurrency is
    958 	     * exceeded) and go and get another child.
    959 	     */
    960 	    if (job->flags & (JOB_RESUME|JOB_REMIGRATE|JOB_RESTART)) {
    961 		if (usePipes && job->node != lastNode) {
    962 		    MESSAGE(out, job->node);
    963 		    lastNode = job->node;
    964 		}
    965 		(void) fprintf(out, "*** Continued\n");
    966 	    }
    967 	    if (!(job->flags & JOB_CONTINUING)) {
    968 		if (DEBUG(JOB)) {
    969 		    (void) fprintf(stdout,
    970 				   "Warning: process %d was not continuing.\n",
    971 				   job->pid);
    972 		    (void) fflush(stdout);
    973 		}
    974 #ifdef notdef
    975 		/*
    976 		 * We don't really want to restart a job from scratch just
    977 		 * because it continued, especially not without killing the
    978 		 * continuing process!  That's why this is ifdef'ed out.
    979 		 * FD - 9/17/90
    980 		 */
    981 		JobRestart(job);
    982 #endif
    983 	    }
    984 	    job->flags &= ~JOB_CONTINUING;
    985  	    Lst_AtEnd(jobs, (ClientData)job);
    986 	    nJobs += 1;
    987 	    if (!(job->flags & JOB_REMOTE)) {
    988 		if (DEBUG(JOB)) {
    989 		    (void) fprintf(stdout,
    990 				   "Process %d is continuing locally.\n",
    991 				   job->pid);
    992 		    (void) fflush(stdout);
    993   		}
    994 		nLocal += 1;
    995 	    }
    996 	    (void) fflush(out);
    997   	    return;
    998 	} else {
    999 	    if (usePipes && job->node != lastNode) {
   1000 		MESSAGE(out, job->node);
   1001 		lastNode = job->node;
   1002 	    }
   1003 	    (void) fprintf(out, "*** Signal %d\n", WTERMSIG(*status));
   1004 	}
   1005 
   1006 	(void) fflush(out);
   1007     }
   1008 
   1009     /*
   1010      * Now handle the -B-mode stuff. If the beast still isn't finished,
   1011      * try and restart the job on the next command. If JobStart says it's
   1012      * ok, it's ok. If there's an error, this puppy is done.
   1013      */
   1014     if (compatMake && (WIFEXITED(*status) &&
   1015 	!Lst_IsAtEnd(job->node->commands))) {
   1016 	switch (JobStart(job->node, job->flags & JOB_IGNDOTS, job)) {
   1017 	case JOB_RUNNING:
   1018 	    done = FALSE;
   1019 	    break;
   1020 	case JOB_ERROR:
   1021 	    done = TRUE;
   1022 	    *status = W_EXITCODE(1, 0);
   1023 	    break;
   1024 	case JOB_FINISHED:
   1025 	    /*
   1026 	     * If we got back a JOB_FINISHED code, JobStart has already
   1027 	     * called Make_Update and freed the job descriptor. We set
   1028 	     * done to false here to avoid fake cycles and double frees.
   1029 	     * JobStart needs to do the update so we can proceed up the
   1030 	     * graph when given the -n flag..
   1031 	     */
   1032 	    done = FALSE;
   1033 	    break;
   1034 	}
   1035     } else {
   1036 	done = TRUE;
   1037     }
   1038 
   1039     if (done) {
   1040 	Trace_Log(JOBEND, job);
   1041 	if (!compatMake && !(job->flags & JOB_SPECIAL)) {
   1042 	    if ((*status != 0) ||
   1043 	        (aborting == ABORT_ERROR) ||
   1044 	        (aborting == ABORT_INTERRUPT))
   1045 		Job_TokenReturn();
   1046 	}
   1047 
   1048     }
   1049 
   1050     if (done &&
   1051 	(aborting != ABORT_ERROR) &&
   1052 	(aborting != ABORT_INTERRUPT) &&
   1053 	(*status == 0))
   1054     {
   1055 	/*
   1056 	 * As long as we aren't aborting and the job didn't return a non-zero
   1057 	 * status that we shouldn't ignore, we call Make_Update to update
   1058 	 * the parents. In addition, any saved commands for the node are placed
   1059 	 * on the .END target.
   1060 	 */
   1061 	if (job->tailCmds != NILLNODE) {
   1062 	    Lst_ForEachFrom(job->node->commands, job->tailCmds,
   1063 			     JobSaveCommand,
   1064 			    (ClientData)job->node);
   1065 	}
   1066 	job->node->made = MADE;
   1067 	if (!(job->flags & JOB_SPECIAL))
   1068 	    Job_TokenReturn();
   1069 	Make_Update(job->node);
   1070 	free((Address)job);
   1071     } else if (*status != 0) {
   1072 	errors += 1;
   1073 	free((Address)job);
   1074     }
   1075     JobRestartJobs();
   1076 
   1077     /*
   1078      * Set aborting if any error.
   1079      */
   1080     if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) {
   1081 	/*
   1082 	 * If we found any errors in this batch of children and the -k flag
   1083 	 * wasn't given, we set the aborting flag so no more jobs get
   1084 	 * started.
   1085 	 */
   1086 	aborting = ABORT_ERROR;
   1087     }
   1088 
   1089     if ((aborting == ABORT_ERROR) && Job_Empty()) {
   1090 	/*
   1091 	 * If we are aborting and the job table is now empty, we finish.
   1092 	 */
   1093 	Finish(errors);
   1094     }
   1095 }
   1096 
   1097 /*-
   1098  *-----------------------------------------------------------------------
   1099  * Job_Touch --
   1100  *	Touch the given target. Called by JobStart when the -t flag was
   1101  *	given
   1102  *
   1103  * Results:
   1104  *	None
   1105  *
   1106  * Side Effects:
   1107  *	The data modification of the file is changed. In addition, if the
   1108  *	file did not exist, it is created.
   1109  *-----------------------------------------------------------------------
   1110  */
   1111 void
   1112 Job_Touch(gn, silent)
   1113     GNode         *gn;	      	/* the node of the file to touch */
   1114     Boolean 	  silent;   	/* TRUE if should not print messages */
   1115 {
   1116     int		  streamID;   	/* ID of stream opened to do the touch */
   1117     struct utimbuf times;	/* Times for utime() call */
   1118 
   1119     if (gn->type & (OP_JOIN|OP_USE|OP_USEBEFORE|OP_EXEC|OP_OPTIONAL|OP_PHONY)) {
   1120 	/*
   1121 	 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets
   1122 	 * and, as such, shouldn't really be created.
   1123 	 */
   1124 	return;
   1125     }
   1126 
   1127     if (!silent || NoExecute(gn)) {
   1128 	(void) fprintf(stdout, "touch %s\n", gn->name);
   1129 	(void) fflush(stdout);
   1130     }
   1131 
   1132     if (NoExecute(gn)) {
   1133 	return;
   1134     }
   1135 
   1136     if (gn->type & OP_ARCHV) {
   1137 	Arch_Touch(gn);
   1138     } else if (gn->type & OP_LIB) {
   1139 	Arch_TouchLib(gn);
   1140     } else {
   1141 	char	*file = gn->path ? gn->path : gn->name;
   1142 
   1143 	times.actime = times.modtime = now;
   1144 	if (utime(file, ×) < 0){
   1145 	    streamID = open(file, O_RDWR | O_CREAT, 0666);
   1146 
   1147 	    if (streamID >= 0) {
   1148 		char	c;
   1149 
   1150 		/*
   1151 		 * Read and write a byte to the file to change the
   1152 		 * modification time, then close the file.
   1153 		 */
   1154 		if (read(streamID, &c, 1) == 1) {
   1155 		    (void) lseek(streamID, (off_t)0, SEEK_SET);
   1156 		    (void) write(streamID, &c, 1);
   1157 		}
   1158 
   1159 		(void) close(streamID);
   1160 	    } else {
   1161 		(void) fprintf(stdout, "*** couldn't touch %s: %s",
   1162 			       file, strerror(errno));
   1163 		(void) fflush(stdout);
   1164 	    }
   1165 	}
   1166     }
   1167 }
   1168 
   1169 /*-
   1170  *-----------------------------------------------------------------------
   1171  * Job_CheckCommands --
   1172  *	Make sure the given node has all the commands it needs.
   1173  *
   1174  * Results:
   1175  *	TRUE if the commands list is/was ok.
   1176  *
   1177  * Side Effects:
   1178  *	The node will have commands from the .DEFAULT rule added to it
   1179  *	if it needs them.
   1180  *-----------------------------------------------------------------------
   1181  */
   1182 Boolean
   1183 Job_CheckCommands(gn, abortProc)
   1184     GNode          *gn;	    	    /* The target whose commands need
   1185 				     * verifying */
   1186     void    	 (*abortProc) __P((char *, ...));
   1187 			/* Function to abort with message */
   1188 {
   1189     if (OP_NOP(gn->type) && Lst_IsEmpty(gn->commands) &&
   1190 	(gn->type & OP_LIB) == 0) {
   1191 	/*
   1192 	 * No commands. Look for .DEFAULT rule from which we might infer
   1193 	 * commands
   1194 	 */
   1195 	if ((DEFAULT != NILGNODE) && !Lst_IsEmpty(DEFAULT->commands)) {
   1196 	    char *p1;
   1197 	    /*
   1198 	     * Make only looks for a .DEFAULT if the node was never the
   1199 	     * target of an operator, so that's what we do too. If
   1200 	     * a .DEFAULT was given, we substitute its commands for gn's
   1201 	     * commands and set the IMPSRC variable to be the target's name
   1202 	     * The DEFAULT node acts like a transformation rule, in that
   1203 	     * gn also inherits any attributes or sources attached to
   1204 	     * .DEFAULT itself.
   1205 	     */
   1206 	    Make_HandleUse(DEFAULT, gn);
   1207 	    Var_Set(IMPSRC, Var_Value(TARGET, gn, &p1), gn, 0);
   1208 	    if (p1)
   1209 		free(p1);
   1210 	} else if (Dir_MTime(gn) == 0) {
   1211 	    /*
   1212 	     * The node wasn't the target of an operator we have no .DEFAULT
   1213 	     * rule to go on and the target doesn't already exist. There's
   1214 	     * nothing more we can do for this branch. If the -k flag wasn't
   1215 	     * given, we stop in our tracks, otherwise we just don't update
   1216 	     * this node's parents so they never get examined.
   1217 	     */
   1218 	    static const char msg[] = ": don't know how to make";
   1219 
   1220 	    if (gn->type & OP_OPTIONAL) {
   1221 		(void) fprintf(stdout, "%s%s %s(ignored)\n", progname,
   1222 		    msg, gn->name);
   1223 		(void) fflush(stdout);
   1224 	    } else if (keepgoing) {
   1225 		(void) fprintf(stdout, "%s%s %s(continuing)\n", progname,
   1226 		    msg, gn->name);
   1227 		(void) fflush(stdout);
   1228   		return FALSE;
   1229 	    } else {
   1230 		(*abortProc)("%s%s %s. Stop", progname, msg, gn->name);
   1231 		return FALSE;
   1232 	    }
   1233 	}
   1234     }
   1235     return TRUE;
   1236 }
   1237 #ifdef RMT_WILL_WATCH
   1238 /*-
   1239  *-----------------------------------------------------------------------
   1240  * JobLocalInput --
   1241  *	Handle a pipe becoming readable. Callback function for Rmt_Watch
   1242  *
   1243  * Results:
   1244  *	None
   1245  *
   1246  * Side Effects:
   1247  *	JobDoOutput is called.
   1248  *
   1249  *-----------------------------------------------------------------------
   1250  */
   1251 /*ARGSUSED*/
   1252 static void
   1253 JobLocalInput(stream, job)
   1254     int	    stream; 	/* Stream that's ready (ignored) */
   1255     Job	    *job;   	/* Job to which the stream belongs */
   1256 {
   1257     JobDoOutput(job, FALSE);
   1258 }
   1259 #endif /* RMT_WILL_WATCH */
   1260 
   1261 /*-
   1262  *-----------------------------------------------------------------------
   1263  * JobExec --
   1264  *	Execute the shell for the given job. Called from JobStart and
   1265  *	JobRestart.
   1266  *
   1267  * Results:
   1268  *	None.
   1269  *
   1270  * Side Effects:
   1271  *	A shell is executed, outputs is altered and the Job structure added
   1272  *	to the job table.
   1273  *
   1274  *-----------------------------------------------------------------------
   1275  */
   1276 static void
   1277 JobExec(job, argv)
   1278     Job	    	  *job; 	/* Job to execute */
   1279     char    	  **argv;
   1280 {
   1281     int	    	  cpid;	    	/* ID of new child */
   1282 
   1283     job->flags &= ~JOB_TRACED;
   1284 
   1285     if (DEBUG(JOB)) {
   1286 	int 	  i;
   1287 
   1288 	(void) fprintf(stdout, "Running %s %sly\n", job->node->name,
   1289 		       job->flags&JOB_REMOTE?"remote":"local");
   1290 	(void) fprintf(stdout, "\tCommand: ");
   1291 	for (i = 0; argv[i] != NULL; i++) {
   1292 	    (void) fprintf(stdout, "%s ", argv[i]);
   1293 	}
   1294  	(void) fprintf(stdout, "\n");
   1295  	(void) fflush(stdout);
   1296     }
   1297 
   1298     /*
   1299      * Some jobs produce no output and it's disconcerting to have
   1300      * no feedback of their running (since they produce no output, the
   1301      * banner with their name in it never appears). This is an attempt to
   1302      * provide that feedback, even if nothing follows it.
   1303      */
   1304     if ((lastNode != job->node) && (job->flags & JOB_FIRST) &&
   1305 	!(job->flags & JOB_SILENT)) {
   1306 	MESSAGE(stdout, job->node);
   1307 	lastNode = job->node;
   1308     }
   1309 
   1310 #ifdef RMT_NO_EXEC
   1311     if (job->flags & JOB_REMOTE) {
   1312 	goto jobExecFinish;
   1313     }
   1314 #endif /* RMT_NO_EXEC */
   1315 
   1316     if ((cpid = vfork()) == -1) {
   1317 	Punt("Cannot vfork: %s", strerror(errno));
   1318     } else if (cpid == 0) {
   1319 
   1320 	/*
   1321 	 * Must duplicate the input stream down to the child's input and
   1322 	 * reset it to the beginning (again). Since the stream was marked
   1323 	 * close-on-exec, we must clear that bit in the new input.
   1324 	 */
   1325 	if (dup2(FILENO(job->cmdFILE), 0) == -1)
   1326 	    Punt("Cannot dup2: %s", strerror(errno));
   1327 	(void) fcntl(0, F_SETFD, 0);
   1328 	(void) lseek(0, (off_t)0, SEEK_SET);
   1329 
   1330 	if (job->node->type & OP_MAKE) {
   1331 		/*
   1332 		 * Pass job token pipe to submakes.
   1333 		 */
   1334 		fcntl(job_pipe[0], F_SETFD, 0);
   1335 		fcntl(job_pipe[1], F_SETFD, 0);
   1336 	}
   1337 
   1338 	if (usePipes) {
   1339 	    /*
   1340 	     * Set up the child's output to be routed through the pipe
   1341 	     * we've created for it.
   1342 	     */
   1343 	    if (dup2(job->outPipe, 1) == -1)
   1344 		Punt("Cannot dup2: %s", strerror(errno));
   1345 	} else {
   1346 	    /*
   1347 	     * We're capturing output in a file, so we duplicate the
   1348 	     * descriptor to the temporary file into the standard
   1349 	     * output.
   1350 	     */
   1351 	    if (dup2(job->outFd, 1) == -1)
   1352 		Punt("Cannot dup2: %s", strerror(errno));
   1353 	}
   1354 	/*
   1355 	 * The output channels are marked close on exec. This bit was
   1356 	 * duplicated by the dup2 (on some systems), so we have to clear
   1357 	 * it before routing the shell's error output to the same place as
   1358 	 * its standard output.
   1359 	 */
   1360 	(void) fcntl(1, F_SETFD, 0);
   1361 	if (dup2(1, 2) == -1)
   1362 	    Punt("Cannot dup2: %s", strerror(errno));
   1363 
   1364 #ifdef USE_PGRP
   1365 	/*
   1366 	 * We want to switch the child into a different process family so
   1367 	 * we can kill it and all its descendants in one fell swoop,
   1368 	 * by killing its process family, but not commit suicide.
   1369 	 */
   1370 # if defined(SYSV)
   1371 	(void) setsid();
   1372 # else
   1373 	(void) setpgid(0, getpid());
   1374 # endif
   1375 #endif /* USE_PGRP */
   1376 
   1377 #ifdef REMOTE
   1378 	if (job->flags & JOB_REMOTE) {
   1379 	    Rmt_Exec(shellPath, argv, FALSE);
   1380 	} else
   1381 #endif /* REMOTE */
   1382 	{
   1383 	   (void) execv(shellPath, argv);
   1384 	   execError(shellPath);
   1385 	}
   1386 	_exit(1);
   1387     } else {
   1388 #ifdef REMOTE
   1389 	sigset_t nmask, omask;
   1390 	sigemptyset(&nmask);
   1391 	sigaddset(&nmask, SIGCHLD);
   1392 	sigprocmask(SIG_BLOCK, &nmask, &omask);
   1393 #endif
   1394 	job->pid = cpid;
   1395 
   1396 	Trace_Log(JOBSTART, job);
   1397 
   1398 	if (usePipes && (job->flags & JOB_FIRST)) {
   1399 	    /*
   1400 	     * The first time a job is run for a node, we set the current
   1401 	     * position in the buffer to the beginning and mark another
   1402 	     * stream to watch in the outputs mask
   1403 	     */
   1404 	    job->curPos = 0;
   1405 
   1406 #ifdef RMT_WILL_WATCH
   1407 	    Rmt_Watch(job->inPipe, JobLocalInput, job);
   1408 #else
   1409 #ifdef USE_SELECT
   1410 	    FD_SET(job->inPipe, &outputs);
   1411 #else
   1412 	    watchfd(job);
   1413 #endif
   1414 #endif /* RMT_WILL_WATCH */
   1415 	}
   1416 
   1417 	if (job->flags & JOB_REMOTE) {
   1418 #ifndef REMOTE
   1419 	    job->rmtID = 0;
   1420 #else
   1421 	    job->rmtID = Rmt_LastID(job->pid);
   1422 #endif /* REMOTE */
   1423 	} else {
   1424 	    nLocal += 1;
   1425 	    /*
   1426 	     * XXX: Used to not happen if REMOTE. Why?
   1427 	     */
   1428 	    if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
   1429 		(void) fclose(job->cmdFILE);
   1430 		job->cmdFILE = NULL;
   1431 	    }
   1432 	}
   1433 #ifdef REMOTE
   1434 	sigprocmask(SIG_SETMASK, &omask, NULL);
   1435 #endif
   1436     }
   1437 
   1438 #ifdef RMT_NO_EXEC
   1439 jobExecFinish:
   1440 #endif
   1441     /*
   1442      * Now the job is actually running, add it to the table.
   1443      */
   1444     nJobs += 1;
   1445     (void) Lst_AtEnd(jobs, (ClientData)job);
   1446 }
   1447 
   1448 /*-
   1449  *-----------------------------------------------------------------------
   1450  * JobMakeArgv --
   1451  *	Create the argv needed to execute the shell for a given job.
   1452  *
   1453  *
   1454  * Results:
   1455  *
   1456  * Side Effects:
   1457  *
   1458  *-----------------------------------------------------------------------
   1459  */
   1460 static void
   1461 JobMakeArgv(job, argv)
   1462     Job	    	  *job;
   1463     char	  **argv;
   1464 {
   1465     int	    	  argc;
   1466     static char	  args[10]; 	/* For merged arguments */
   1467 
   1468     argv[0] = shellName;
   1469     argc = 1;
   1470 
   1471     if ((commandShell->exit && (*commandShell->exit != '-')) ||
   1472 	(commandShell->echo && (*commandShell->echo != '-')))
   1473     {
   1474 	/*
   1475 	 * At least one of the flags doesn't have a minus before it, so
   1476 	 * merge them together. Have to do this because the *(&(@*#*&#$#
   1477 	 * Bourne shell thinks its second argument is a file to source.
   1478 	 * Grrrr. Note the ten-character limitation on the combined arguments.
   1479 	 */
   1480 	(void)snprintf(args, sizeof(args), "-%s%s",
   1481 		      ((job->flags & JOB_IGNERR) ? "" :
   1482 		       (commandShell->exit ? commandShell->exit : "")),
   1483 		      ((job->flags & JOB_SILENT) ? "" :
   1484 		       (commandShell->echo ? commandShell->echo : "")));
   1485 
   1486 	if (args[1]) {
   1487 	    argv[argc] = args;
   1488 	    argc++;
   1489 	}
   1490     } else {
   1491 	if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
   1492 	    argv[argc] = commandShell->exit;
   1493 	    argc++;
   1494 	}
   1495 	if (!(job->flags & JOB_SILENT) && commandShell->echo) {
   1496 	    argv[argc] = commandShell->echo;
   1497 	    argc++;
   1498 	}
   1499     }
   1500     argv[argc] = NULL;
   1501 }
   1502 
   1503 /*-
   1504  *-----------------------------------------------------------------------
   1505  * JobRestart --
   1506  *	Restart a job that stopped for some reason.
   1507  *
   1508  * Results:
   1509  *	None.
   1510  *
   1511  *-----------------------------------------------------------------------
   1512  */
   1513 static void
   1514 JobRestart(job)
   1515     Job 	  *job;    	/* Job to restart */
   1516 {
   1517 #ifdef REMOTE
   1518     int host;
   1519 #endif
   1520 
   1521     if (job->flags & JOB_REMIGRATE) {
   1522 	if (
   1523 #ifdef REMOTE
   1524 	    verboseRemigrates ||
   1525 #endif
   1526 	    DEBUG(JOB)) {
   1527 	   (void) fprintf(stdout, "*** remigrating %x(%s)\n",
   1528 			   job->pid, job->node->name);
   1529 	   (void) fflush(stdout);
   1530 	}
   1531 
   1532 #ifdef REMOTE
   1533 	if (!Rmt_ReExport(job->pid, job->node, &host)) {
   1534 	    if (verboseRemigrates || DEBUG(JOB)) {
   1535 		(void) fprintf(stdout, "*** couldn't migrate...\n");
   1536 		(void) fflush(stdout);
   1537 	    }
   1538 #endif
   1539 	    if (nLocal != maxLocal) {
   1540 		/*
   1541 		 * Job cannot be remigrated, but there's room on the local
   1542 		 * machine, so resume the job and note that another
   1543 		 * local job has started.
   1544 		 */
   1545 		if (
   1546 #ifdef REMOTE
   1547 		    verboseRemigrates ||
   1548 #endif
   1549 		    DEBUG(JOB)) {
   1550 		    (void) fprintf(stdout, "*** resuming on local machine\n");
   1551 		    (void) fflush(stdout);
   1552 		}
   1553 		KILL(job->pid, SIGCONT);
   1554 		nLocal +=1;
   1555 #ifdef REMOTE
   1556 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE);
   1557 		job->flags |= JOB_CONTINUING;
   1558 #else
   1559 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
   1560 #endif
   1561 	} else {
   1562 		/*
   1563 		 * Job cannot be restarted. Mark the table as full and
   1564 		 * place the job back on the list of stopped jobs.
   1565 		 */
   1566 		if (
   1567 #ifdef REMOTE
   1568 		    verboseRemigrates ||
   1569 #endif
   1570 		    DEBUG(JOB)) {
   1571 		   (void) fprintf(stdout, "*** holding\n");
   1572 		   (void) fflush(stdout);
   1573   		}
   1574 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
   1575 		return;
   1576 	    }
   1577 #ifdef REMOTE
   1578 	} else {
   1579 	    /*
   1580 	     * Clear out the remigrate and resume flags. Set the continuing
   1581 	     * flag so we know later on that the process isn't exiting just
   1582 	     * because of a signal.
   1583 	     */
   1584 	    job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
   1585 	    job->flags |= JOB_CONTINUING;
   1586 	    job->rmtID = host;
   1587 	}
   1588 #endif
   1589 
   1590 	(void)Lst_AtEnd(jobs, (ClientData)job);
   1591 	nJobs += 1;
   1592     } else if (job->flags & JOB_RESTART) {
   1593 	/*
   1594 	 * Set up the control arguments to the shell. This is based on the
   1595 	 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
   1596 	 * the 'exit' flag of the commandShell is used to cause it to exit
   1597 	 * upon receiving an error. If the JOB_SILENT flag is clear, the
   1598 	 * 'echo' flag of the commandShell is used to get it to start echoing
   1599 	 * as soon as it starts processing commands.
   1600 	 */
   1601 	char	  *argv[10];
   1602 
   1603 	JobMakeArgv(job, argv);
   1604 
   1605 	if (DEBUG(JOB)) {
   1606 	    (void) fprintf(stdout, "Restarting %s...", job->node->name);
   1607 	    (void) fflush(stdout);
   1608 	}
   1609 #ifdef REMOTE
   1610 	if ((job->node->type&OP_NOEXPORT) ||
   1611  	    (nLocal < maxLocal && runLocalFirst)
   1612 # ifdef RMT_NO_EXEC
   1613 	    || !Rmt_Export(shellPath, argv, job)
   1614 # else
   1615 	    || !Rmt_Begin(shellPath, argv, job->node)
   1616 # endif
   1617 #endif
   1618 	{
   1619 	    if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) {
   1620 		/*
   1621 		 * Can't be exported and not allowed to run locally -- put it
   1622 		 * back on the hold queue and mark the table full
   1623 		 */
   1624 		if (DEBUG(JOB)) {
   1625 		    (void) fprintf(stdout, "holding\n");
   1626 		    (void) fflush(stdout);
   1627 		}
   1628 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
   1629 		return;
   1630 	    } else {
   1631 		/*
   1632 		 * Job may be run locally.
   1633 		 */
   1634 		if (DEBUG(JOB)) {
   1635 		    (void) fprintf(stdout, "running locally\n");
   1636 		    (void) fflush(stdout);
   1637 		}
   1638 		job->flags &= ~JOB_REMOTE;
   1639 	    }
   1640 	}
   1641 #ifdef REMOTE
   1642 	else {
   1643 	    /*
   1644 	     * Can be exported. Hooray!
   1645 	     */
   1646 	    if (DEBUG(JOB)) {
   1647 		(void) fprintf(stdout, "exporting\n");
   1648 		(void) fflush(stdout);
   1649 	    }
   1650 	    job->flags |= JOB_REMOTE;
   1651 	}
   1652 #endif
   1653 	JobExec(job, argv);
   1654     } else {
   1655 	/*
   1656 	 * The job has stopped and needs to be restarted. Why it stopped,
   1657 	 * we don't know...
   1658 	 */
   1659 	if (DEBUG(JOB)) {
   1660 	   (void) fprintf(stdout, "Resuming %s...", job->node->name);
   1661 	   (void) fflush(stdout);
   1662 	}
   1663 	if (((job->flags & JOB_REMOTE) ||
   1664 	    (nLocal < maxLocal) ||
   1665 #ifdef REMOTE
   1666 	    (((job->flags & JOB_SPECIAL) &&
   1667 	      (job->node->type & OP_NOEXPORT)) &&
   1668 	     (maxLocal == 0))) &&
   1669 #else
   1670 	    ((job->flags & JOB_SPECIAL) &&
   1671 	     (maxLocal == 0))) &&
   1672 #endif
   1673 	   (nJobs != maxJobs))
   1674 	{
   1675 	    /*
   1676 	     * If the job is remote, it's ok to resume it as long as the
   1677 	     * maximum concurrency won't be exceeded. If it's local and
   1678 	     * we haven't reached the local concurrency limit already (or the
   1679 	     * job must be run locally and maxLocal is 0), it's also ok to
   1680 	     * resume it.
   1681 	     */
   1682 	    Boolean error;
   1683 	    int status;
   1684 
   1685 #ifdef RMT_WANTS_SIGNALS
   1686 	    if (job->flags & JOB_REMOTE) {
   1687 		error = !Rmt_Signal(job, SIGCONT);
   1688 	    } else
   1689 #endif	/* RMT_WANTS_SIGNALS */
   1690 		error = (KILL(job->pid, SIGCONT) != 0);
   1691 
   1692 	    if (!error) {
   1693 		/*
   1694 		 * Make sure the user knows we've continued the beast and
   1695 		 * actually put the thing in the job table.
   1696 		 */
   1697 		job->flags |= JOB_CONTINUING;
   1698 		status = W_STOPCODE(SIGCONT);
   1699 		JobFinish(job, &status);
   1700 
   1701 		job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
   1702 		if (DEBUG(JOB)) {
   1703 		   (void) fprintf(stdout, "done\n");
   1704 		   (void) fflush(stdout);
   1705 		}
   1706 	    } else {
   1707 		Error("couldn't resume %s: %s",
   1708 		    job->node->name, strerror(errno));
   1709 		status = W_EXITCODE(1, 0);
   1710 		JobFinish(job, &status);
   1711 	    }
   1712 	} else {
   1713 	    /*
   1714 	     * Job cannot be restarted. Mark the table as full and
   1715 	     * place the job back on the list of stopped jobs.
   1716 	     */
   1717 	    if (DEBUG(JOB)) {
   1718 		(void) fprintf(stdout, "table full\n");
   1719 		(void) fflush(stdout);
   1720 	    }
   1721 	    (void) Lst_AtFront(stoppedJobs, (ClientData)job);
   1722 	}
   1723     }
   1724 }
   1725 
   1726 /*-
   1727  *-----------------------------------------------------------------------
   1728  * JobStart  --
   1729  *	Start a target-creation process going for the target described
   1730  *	by the graph node gn.
   1731  *
   1732  * Results:
   1733  *	JOB_ERROR if there was an error in the commands, JOB_FINISHED
   1734  *	if there isn't actually anything left to do for the job and
   1735  *	JOB_RUNNING if the job has been started.
   1736  *
   1737  * Side Effects:
   1738  *	A new Job node is created and added to the list of running
   1739  *	jobs. PMake is forked and a child shell created.
   1740  *-----------------------------------------------------------------------
   1741  */
   1742 static int
   1743 JobStart(gn, flags, previous)
   1744     GNode         *gn;	      /* target to create */
   1745     int	  	   flags;      /* flags for the job to override normal ones.
   1746 			       * e.g. JOB_SPECIAL or JOB_IGNDOTS */
   1747     Job 	  *previous;  /* The previous Job structure for this node,
   1748 			       * if any. */
   1749 {
   1750     register Job  *job;       /* new job descriptor */
   1751     char	  *argv[10];  /* Argument vector to shell */
   1752     Boolean	  cmdsOK;     /* true if the nodes commands were all right */
   1753     Boolean 	  local;      /* Set true if the job was run locally */
   1754     Boolean 	  noExec;     /* Set true if we decide not to run the job */
   1755     int		  tfd;	      /* File descriptor to the temp file */
   1756 
   1757     if (previous != NULL) {
   1758 	previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
   1759 	job = previous;
   1760     } else {
   1761 	job = (Job *) emalloc(sizeof(Job));
   1762 	if (job == NULL) {
   1763 	    Punt("JobStart out of memory");
   1764 	}
   1765 	flags |= JOB_FIRST;
   1766     }
   1767 
   1768     job->node = gn;
   1769     job->tailCmds = NILLNODE;
   1770 
   1771     /*
   1772      * Set the initial value of the flags for this job based on the global
   1773      * ones and the node's attributes... Any flags supplied by the caller
   1774      * are also added to the field.
   1775      */
   1776     job->flags = 0;
   1777     if (Targ_Ignore(gn)) {
   1778 	job->flags |= JOB_IGNERR;
   1779     }
   1780     if (Targ_Silent(gn)) {
   1781 	job->flags |= JOB_SILENT;
   1782     }
   1783     job->flags |= flags;
   1784 
   1785     /*
   1786      * Check the commands now so any attributes from .DEFAULT have a chance
   1787      * to migrate to the node
   1788      */
   1789     if (!compatMake && job->flags & JOB_FIRST) {
   1790 	cmdsOK = Job_CheckCommands(gn, Error);
   1791     } else {
   1792 	cmdsOK = TRUE;
   1793     }
   1794 
   1795 #ifndef RMT_WILL_WATCH
   1796 #ifndef USE_SELECT
   1797     job->inPollfd = NULL;
   1798 #endif
   1799 #endif
   1800     /*
   1801      * If the -n flag wasn't given, we open up OUR (not the child's)
   1802      * temporary file to stuff commands in it. The thing is rd/wr so we don't
   1803      * need to reopen it to feed it to the shell. If the -n flag *was* given,
   1804      * we just set the file to be stdout. Cute, huh?
   1805      */
   1806     if (((gn->type & OP_MAKE) && !(noRecursiveExecute)) ||
   1807 	(!noExecute && !touchFlag)) {
   1808 	/*
   1809 	 * tfile is the name of a file into which all shell commands are
   1810 	 * put. It is used over by removing it before the child shell is
   1811 	 * executed. The XXXXXX in the string are replaced by the pid of
   1812 	 * the make process in a 6-character field with leading zeroes.
   1813 	 */
   1814 	char     tfile[sizeof(TMPPAT)];
   1815 	/*
   1816 	 * We're serious here, but if the commands were bogus, we're
   1817 	 * also dead...
   1818 	 */
   1819 	if (!cmdsOK) {
   1820 	    DieHorribly();
   1821 	}
   1822 
   1823 	(void)strcpy(tfile, TMPPAT);
   1824 	if ((tfd = mkstemp(tfile)) == -1)
   1825 	    Punt("Could not create temporary file %s", strerror(errno));
   1826 	(void) eunlink(tfile);
   1827 
   1828 	job->cmdFILE = fdopen(tfd, "w+");
   1829 	if (job->cmdFILE == NULL) {
   1830 	    Punt("Could not fdopen %s", tfile);
   1831 	}
   1832 	(void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1);
   1833 	/*
   1834 	 * Send the commands to the command file, flush all its buffers then
   1835 	 * rewind and remove the thing.
   1836 	 */
   1837 	noExec = FALSE;
   1838 
   1839 	/*
   1840 	 * used to be backwards; replace when start doing multiple commands
   1841 	 * per shell.
   1842 	 */
   1843 	if (compatMake) {
   1844 	    /*
   1845 	     * Be compatible: If this is the first time for this node,
   1846 	     * verify its commands are ok and open the commands list for
   1847 	     * sequential access by later invocations of JobStart.
   1848 	     * Once that is done, we take the next command off the list
   1849 	     * and print it to the command file. If the command was an
   1850 	     * ellipsis, note that there's nothing more to execute.
   1851 	     */
   1852 	    if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
   1853 		cmdsOK = FALSE;
   1854 	    } else {
   1855 		LstNode	ln = Lst_Next(gn->commands);
   1856 
   1857 		if ((ln == NILLNODE) ||
   1858 		    JobPrintCommand((ClientData) Lst_Datum(ln),
   1859 				    (ClientData) job))
   1860 		{
   1861 		    noExec = TRUE;
   1862 		    Lst_Close(gn->commands);
   1863 		}
   1864 		if (noExec && !(job->flags & JOB_FIRST)) {
   1865 		    /*
   1866 		     * If we're not going to execute anything, the job
   1867 		     * is done and we need to close down the various
   1868 		     * file descriptors we've opened for output, then
   1869 		     * call JobDoOutput to catch the final characters or
   1870 		     * send the file to the screen... Note that the i/o streams
   1871 		     * are only open if this isn't the first job.
   1872 		     * Note also that this could not be done in
   1873 		     * Job_CatchChildren b/c it wasn't clear if there were
   1874 		     * more commands to execute or not...
   1875 		     */
   1876 		    JobClose(job);
   1877 		}
   1878 	    }
   1879 	} else {
   1880 	    /*
   1881 	     * We can do all the commands at once. hooray for sanity
   1882 	     */
   1883 	    numCommands = 0;
   1884 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
   1885 
   1886 	    /*
   1887 	     * If we didn't print out any commands to the shell script,
   1888 	     * there's not much point in executing the shell, is there?
   1889 	     */
   1890 	    if (numCommands == 0) {
   1891 		noExec = TRUE;
   1892 	    }
   1893 	}
   1894     } else if (NoExecute(gn)) {
   1895 	/*
   1896 	 * Not executing anything -- just print all the commands to stdout
   1897 	 * in one fell swoop. This will still set up job->tailCmds correctly.
   1898 	 */
   1899 	if (lastNode != gn) {
   1900 	    MESSAGE(stdout, gn);
   1901 	    lastNode = gn;
   1902 	}
   1903 	job->cmdFILE = stdout;
   1904 	/*
   1905 	 * Only print the commands if they're ok, but don't die if they're
   1906 	 * not -- just let the user know they're bad and keep going. It
   1907 	 * doesn't do any harm in this case and may do some good.
   1908 	 */
   1909 	if (cmdsOK) {
   1910 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
   1911 	}
   1912 	/*
   1913 	 * Don't execute the shell, thank you.
   1914 	 */
   1915 	noExec = TRUE;
   1916     } else {
   1917 	/*
   1918 	 * Just touch the target and note that no shell should be executed.
   1919 	 * Set cmdFILE to stdout to make life easier. Check the commands, too,
   1920 	 * but don't die if they're no good -- it does no harm to keep working
   1921 	 * up the graph.
   1922 	 */
   1923 	job->cmdFILE = stdout;
   1924     	Job_Touch(gn, job->flags&JOB_SILENT);
   1925 	noExec = TRUE;
   1926     }
   1927 
   1928     /*
   1929      * If we're not supposed to execute a shell, don't.
   1930      */
   1931     if (noExec) {
   1932 	/*
   1933 	 * Unlink and close the command file if we opened one
   1934 	 */
   1935 	if (job->cmdFILE != stdout) {
   1936 	    if (job->cmdFILE != NULL) {
   1937 		(void) fclose(job->cmdFILE);
   1938 		job->cmdFILE = NULL;
   1939 	    }
   1940 	} else {
   1941 	     (void) fflush(stdout);
   1942 	}
   1943 
   1944 	/*
   1945 	 * We only want to work our way up the graph if we aren't here because
   1946 	 * the commands for the job were no good.
   1947 	 */
   1948 	if (cmdsOK) {
   1949 	    if (aborting == 0) {
   1950 		if (job->tailCmds != NILLNODE) {
   1951 		    Lst_ForEachFrom(job->node->commands, job->tailCmds,
   1952 				    JobSaveCommand,
   1953 				   (ClientData)job->node);
   1954 		}
   1955 		if (!(job->flags & JOB_SPECIAL))
   1956 		    Job_TokenReturn();
   1957 		Make_Update(job->node);
   1958 	    }
   1959 	    free((Address)job);
   1960 	    return(JOB_FINISHED);
   1961 	} else {
   1962 	    free((Address)job);
   1963 	    return(JOB_ERROR);
   1964 	}
   1965     } else {
   1966 	(void) fflush(job->cmdFILE);
   1967     }
   1968 
   1969     /*
   1970      * Set up the control arguments to the shell. This is based on the flags
   1971      * set earlier for this job.
   1972      */
   1973     JobMakeArgv(job, argv);
   1974 
   1975     /*
   1976      * If we're using pipes to catch output, create the pipe by which we'll
   1977      * get the shell's output. If we're using files, print out that we're
   1978      * starting a job and then set up its temporary-file name.
   1979      */
   1980     if (!compatMake || (job->flags & JOB_FIRST)) {
   1981 	if (usePipes) {
   1982 	    int fd[2];
   1983 	    if (pipe(fd) == -1)
   1984 		Punt("Cannot create pipe: %s", strerror(errno));
   1985 	    job->inPipe = fd[0];
   1986 #ifdef USE_SELECT
   1987 	    if (job->inPipe >= FD_SETSIZE)
   1988 		Punt("Ran out of fd_set slots; "
   1989 		    "recompile with a larger FD_SETSIZE.");
   1990 #endif
   1991 	    job->outPipe = fd[1];
   1992 	    (void) fcntl(job->inPipe, F_SETFD, 1);
   1993 	    (void) fcntl(job->outPipe, F_SETFD, 1);
   1994 	} else {
   1995 	    (void) fprintf(stdout, "Remaking `%s'\n", gn->name);
   1996   	    (void) fflush(stdout);
   1997 	    (void) strcpy(job->outFile, TMPPAT);
   1998 	    job->outFd = mkstemp(job->outFile);
   1999 	    (void) fcntl(job->outFd, F_SETFD, 1);
   2000 	}
   2001     }
   2002 
   2003 #ifdef REMOTE
   2004     if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) {
   2005 #ifdef RMT_NO_EXEC
   2006 	local = !Rmt_Export(shellPath, argv, job);
   2007 #else
   2008 	local = !Rmt_Begin(shellPath, argv, job->node);
   2009 #endif /* RMT_NO_EXEC */
   2010 	if (!local) {
   2011 	    job->flags |= JOB_REMOTE;
   2012 	}
   2013     } else
   2014 #endif
   2015 	local = TRUE;
   2016 
   2017     if (local && (((nLocal >= maxLocal) &&
   2018 	!(job->flags & JOB_SPECIAL) &&
   2019 #ifdef REMOTE
   2020 	(!(gn->type & OP_NOEXPORT) || (maxLocal != 0))
   2021 #else
   2022 	(maxLocal != 0)
   2023 #endif
   2024 	)))
   2025     {
   2026 	/*
   2027 	 * The job can only be run locally, but we've hit the limit of
   2028 	 * local concurrency, so put the job on hold until some other job
   2029 	 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
   2030 	 * may be run locally even when the local limit has been reached
   2031 	 * (e.g. when maxLocal == 0), though they will be exported if at
   2032 	 * all possible. In addition, any target marked with .NOEXPORT will
   2033 	 * be run locally if maxLocal is 0.
   2034 	 */
   2035 	job->flags |= JOB_RESTART;
   2036 	(void) Lst_AtEnd(stoppedJobs, (ClientData)job);
   2037     } else {
   2038 	JobExec(job, argv);
   2039     }
   2040     return(JOB_RUNNING);
   2041 }
   2042 
   2043 static char *
   2044 JobOutput(job, cp, endp, msg)
   2045     register Job *job;
   2046     register char *cp, *endp;
   2047     int msg;
   2048 {
   2049     register char *ecp;
   2050 
   2051     if (commandShell->noPrint) {
   2052 	ecp = Str_FindSubstring(cp, commandShell->noPrint);
   2053 	while (ecp != NULL) {
   2054 	    if (cp != ecp) {
   2055 		*ecp = '\0';
   2056 		if (msg && job->node != lastNode) {
   2057 		    MESSAGE(stdout, job->node);
   2058 		    lastNode = job->node;
   2059 		}
   2060 		/*
   2061 		 * The only way there wouldn't be a newline after
   2062 		 * this line is if it were the last in the buffer.
   2063 		 * however, since the non-printable comes after it,
   2064 		 * there must be a newline, so we don't print one.
   2065 		 */
   2066 		(void) fprintf(stdout, "%s", cp);
   2067 		(void) fflush(stdout);
   2068 	    }
   2069 	    cp = ecp + commandShell->noPLen;
   2070 	    if (cp != endp) {
   2071 		/*
   2072 		 * Still more to print, look again after skipping
   2073 		 * the whitespace following the non-printable
   2074 		 * command....
   2075 		 */
   2076 		cp++;
   2077 		while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
   2078 		    cp++;
   2079 		}
   2080 		ecp = Str_FindSubstring(cp, commandShell->noPrint);
   2081 	    } else {
   2082 		return cp;
   2083 	    }
   2084 	}
   2085     }
   2086     return cp;
   2087 }
   2088 
   2089 /*-
   2090  *-----------------------------------------------------------------------
   2091  * JobDoOutput  --
   2092  *	This function is called at different times depending on
   2093  *	whether the user has specified that output is to be collected
   2094  *	via pipes or temporary files. In the former case, we are called
   2095  *	whenever there is something to read on the pipe. We collect more
   2096  *	output from the given job and store it in the job's outBuf. If
   2097  *	this makes up a line, we print it tagged by the job's identifier,
   2098  *	as necessary.
   2099  *	If output has been collected in a temporary file, we open the
   2100  *	file and read it line by line, transfering it to our own
   2101  *	output channel until the file is empty. At which point we
   2102  *	remove the temporary file.
   2103  *	In both cases, however, we keep our figurative eye out for the
   2104  *	'noPrint' line for the shell from which the output came. If
   2105  *	we recognize a line, we don't print it. If the command is not
   2106  *	alone on the line (the character after it is not \0 or \n), we
   2107  *	do print whatever follows it.
   2108  *
   2109  * Results:
   2110  *	None
   2111  *
   2112  * Side Effects:
   2113  *	curPos may be shifted as may the contents of outBuf.
   2114  *-----------------------------------------------------------------------
   2115  */
   2116 STATIC void
   2117 JobDoOutput(job, finish)
   2118     register Job   *job;	  /* the job whose output needs printing */
   2119     Boolean	   finish;	  /* TRUE if this is the last time we'll be
   2120 				   * called for this job */
   2121 {
   2122     Boolean       gotNL = FALSE;  /* true if got a newline */
   2123     Boolean       fbuf;  	  /* true if our buffer filled up */
   2124     register int  nr;	      	  /* number of bytes read */
   2125     register int  i;	      	  /* auxiliary index into outBuf */
   2126     register int  max;	      	  /* limit for i (end of current data) */
   2127     int		  nRead;      	  /* (Temporary) number of bytes read */
   2128 
   2129     FILE      	  *oFILE;	  /* Stream pointer to shell's output file */
   2130     char          inLine[132];
   2131 
   2132 
   2133     if (usePipes) {
   2134 	/*
   2135 	 * Read as many bytes as will fit in the buffer.
   2136 	 */
   2137 end_loop:
   2138 	gotNL = FALSE;
   2139 	fbuf = FALSE;
   2140 
   2141 	nRead = read(job->inPipe, &job->outBuf[job->curPos],
   2142 			 JOB_BUFSIZE - job->curPos);
   2143 	if (nRead < 0) {
   2144 	    if (DEBUG(JOB)) {
   2145 		perror("JobDoOutput(piperead)");
   2146 	    }
   2147 	    nr = 0;
   2148 	} else {
   2149 	    nr = nRead;
   2150 	}
   2151 
   2152 	/*
   2153 	 * If we hit the end-of-file (the job is dead), we must flush its
   2154 	 * remaining output, so pretend we read a newline if there's any
   2155 	 * output remaining in the buffer.
   2156 	 * Also clear the 'finish' flag so we stop looping.
   2157 	 */
   2158 	if ((nr == 0) && (job->curPos != 0)) {
   2159 	    job->outBuf[job->curPos] = '\n';
   2160 	    nr = 1;
   2161 	    finish = FALSE;
   2162 	} else if (nr == 0) {
   2163 	    finish = FALSE;
   2164 	}
   2165 
   2166 	/*
   2167 	 * Look for the last newline in the bytes we just got. If there is
   2168 	 * one, break out of the loop with 'i' as its index and gotNL set
   2169 	 * TRUE.
   2170 	 */
   2171 	max = job->curPos + nr;
   2172 	for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
   2173 	    if (job->outBuf[i] == '\n') {
   2174 		gotNL = TRUE;
   2175 		break;
   2176 	    } else if (job->outBuf[i] == '\0') {
   2177 		/*
   2178 		 * Why?
   2179 		 */
   2180 		job->outBuf[i] = ' ';
   2181 	    }
   2182 	}
   2183 
   2184 	if (!gotNL) {
   2185 	    job->curPos += nr;
   2186 	    if (job->curPos == JOB_BUFSIZE) {
   2187 		/*
   2188 		 * If we've run out of buffer space, we have no choice
   2189 		 * but to print the stuff. sigh.
   2190 		 */
   2191 		fbuf = TRUE;
   2192 		i = job->curPos;
   2193 	    }
   2194 	}
   2195 	if (gotNL || fbuf) {
   2196 	    /*
   2197 	     * Need to send the output to the screen. Null terminate it
   2198 	     * first, overwriting the newline character if there was one.
   2199 	     * So long as the line isn't one we should filter (according
   2200 	     * to the shell description), we print the line, preceded
   2201 	     * by a target banner if this target isn't the same as the
   2202 	     * one for which we last printed something.
   2203 	     * The rest of the data in the buffer are then shifted down
   2204 	     * to the start of the buffer and curPos is set accordingly.
   2205 	     */
   2206 	    job->outBuf[i] = '\0';
   2207 	    if (i >= job->curPos) {
   2208 		char *cp;
   2209 
   2210 		cp = JobOutput(job, job->outBuf, &job->outBuf[i], FALSE);
   2211 
   2212 		/*
   2213 		 * There's still more in that thar buffer. This time, though,
   2214 		 * we know there's no newline at the end, so we add one of
   2215 		 * our own free will.
   2216 		 */
   2217 		if (*cp != '\0') {
   2218 		    if (job->node != lastNode) {
   2219 			MESSAGE(stdout, job->node);
   2220 			lastNode = job->node;
   2221 		    }
   2222 		    (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : "");
   2223 		    (void) fflush(stdout);
   2224 		}
   2225 	    }
   2226 	    if (i < max - 1) {
   2227 		/* shift the remaining characters down */
   2228 		(void) memcpy(job->outBuf, &job->outBuf[i + 1], max - (i + 1));
   2229 		job->curPos = max - (i + 1);
   2230 
   2231 	    } else {
   2232 		/*
   2233 		 * We have written everything out, so we just start over
   2234 		 * from the start of the buffer. No copying. No nothing.
   2235 		 */
   2236 		job->curPos = 0;
   2237 	    }
   2238 	}
   2239 	if (finish) {
   2240 	    /*
   2241 	     * If the finish flag is true, we must loop until we hit
   2242 	     * end-of-file on the pipe. This is guaranteed to happen
   2243 	     * eventually since the other end of the pipe is now closed
   2244 	     * (we closed it explicitly and the child has exited). When
   2245 	     * we do get an EOF, finish will be set FALSE and we'll fall
   2246 	     * through and out.
   2247 	     */
   2248 	    goto end_loop;
   2249 	}
   2250     } else {
   2251 	/*
   2252 	 * We've been called to retrieve the output of the job from the
   2253 	 * temporary file where it's been squirreled away. This consists of
   2254 	 * opening the file, reading the output line by line, being sure not
   2255 	 * to print the noPrint line for the shell we used, then close and
   2256 	 * remove the temporary file. Very simple.
   2257 	 *
   2258 	 * Change to read in blocks and do FindSubString type things as for
   2259 	 * pipes? That would allow for "@echo -n..."
   2260 	 */
   2261 	oFILE = fopen(job->outFile, "r");
   2262 	if (oFILE != NULL) {
   2263 	    (void) fprintf(stdout, "Results of making %s:\n", job->node->name);
   2264 	    (void) fflush(stdout);
   2265 	    while (fgets(inLine, sizeof(inLine), oFILE) != NULL) {
   2266 		register char	*cp, *endp, *oendp;
   2267 
   2268 		cp = inLine;
   2269 		oendp = endp = inLine + strlen(inLine);
   2270 		if (endp[-1] == '\n') {
   2271 		    *--endp = '\0';
   2272 		}
   2273 		cp = JobOutput(job, inLine, endp, FALSE);
   2274 
   2275 		/*
   2276 		 * There's still more in that thar buffer. This time, though,
   2277 		 * we know there's no newline at the end, so we add one of
   2278 		 * our own free will.
   2279 		 */
   2280 		(void) fprintf(stdout, "%s", cp);
   2281 		(void) fflush(stdout);
   2282 		if (endp != oendp) {
   2283 		    (void) fprintf(stdout, "\n");
   2284 		    (void) fflush(stdout);
   2285 		}
   2286 	    }
   2287 	    (void) fclose(oFILE);
   2288 	    (void) eunlink(job->outFile);
   2289 	} else {
   2290 	    Punt("Cannot open `%s'", job->outFile);
   2291 	}
   2292     }
   2293 }
   2294 
   2295 /*-
   2296  *-----------------------------------------------------------------------
   2297  * Job_CatchChildren --
   2298  *	Handle the exit of a child. Called from Make_Make.
   2299  *
   2300  * Results:
   2301  *	none.
   2302  *
   2303  * Side Effects:
   2304  *	The job descriptor is removed from the list of children.
   2305  *
   2306  * Notes:
   2307  *	We do waits, blocking or not, according to the wisdom of our
   2308  *	caller, until there are no more children to report. For each
   2309  *	job, call JobFinish to finish things off. This will take care of
   2310  *	putting jobs on the stoppedJobs queue.
   2311  *
   2312  *-----------------------------------------------------------------------
   2313  */
   2314 void
   2315 Job_CatchChildren(block)
   2316     Boolean	  block;    	/* TRUE if should block on the wait. */
   2317 {
   2318     int    	  pid;	    	/* pid of dead child */
   2319     register Job  *job;	    	/* job descriptor for dead child */
   2320     LstNode       jnode;    	/* list element for finding job */
   2321     int	  	  status;   	/* Exit/termination status */
   2322 
   2323     /*
   2324      * Don't even bother if we know there's no one around.
   2325      */
   2326     if (nLocal == 0) {
   2327 	return;
   2328     }
   2329 
   2330     while ((pid = waitpid((pid_t) -1, &status,
   2331 			  (block?0:WNOHANG)|WUNTRACED)) > 0)
   2332     {
   2333 	if (DEBUG(JOB)) {
   2334 	    (void) fprintf(stdout, "Process %d exited or stopped %x.\n", pid,
   2335 	      status);
   2336 	    (void) fflush(stdout);
   2337 	}
   2338 
   2339 
   2340 	jnode = Lst_Find(jobs, (ClientData)&pid, JobCmpPid);
   2341 
   2342 	if (jnode == NILLNODE) {
   2343 	    if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGCONT)) {
   2344 		jnode = Lst_Find(stoppedJobs, (ClientData) &pid, JobCmpPid);
   2345 		if (jnode == NILLNODE) {
   2346 		    Error("Resumed child (%d) not in table", pid);
   2347 		    continue;
   2348 		}
   2349 		job = (Job *)Lst_Datum(jnode);
   2350 		(void) Lst_Remove(stoppedJobs, jnode);
   2351 	    } else {
   2352 		Error("Child (%d) not in table?", pid);
   2353 		continue;
   2354 	    }
   2355 	} else {
   2356 	    job = (Job *) Lst_Datum(jnode);
   2357 	    (void) Lst_Remove(jobs, jnode);
   2358 	    nJobs -= 1;
   2359 #ifdef REMOTE
   2360 	    if (!(job->flags & JOB_REMOTE)) {
   2361 		if (DEBUG(JOB)) {
   2362 		    (void) fprintf(stdout,
   2363 				   "Job queue has one fewer local process.\n");
   2364 		    (void) fflush(stdout);
   2365 		}
   2366 		nLocal -= 1;
   2367 	    }
   2368 #else
   2369 	    nLocal -= 1;
   2370 #endif
   2371 	}
   2372 
   2373 	JobFinish(job, &status);
   2374     }
   2375 }
   2376 
   2377 /*-
   2378  *-----------------------------------------------------------------------
   2379  * Job_CatchOutput --
   2380  *	Catch the output from our children, if we're using
   2381  *	pipes do so. Otherwise just block time until we get a
   2382  *	signal (most likely a SIGCHLD) since there's no point in
   2383  *	just spinning when there's nothing to do and the reaping
   2384  *	of a child can wait for a while.
   2385  *
   2386  * Results:
   2387  *	None
   2388  *
   2389  * Side Effects:
   2390  *	Output is read from pipes if we're piping.
   2391  * -----------------------------------------------------------------------
   2392  */
   2393 void
   2394 Job_CatchOutput()
   2395 {
   2396     int           	  nready;
   2397     register LstNode	  ln;
   2398     register Job   	  *job;
   2399 #ifdef RMT_WILL_WATCH
   2400     int	    	  	  pnJobs;   	/* Previous nJobs */
   2401 #endif
   2402 
   2403     (void) fflush(stdout);
   2404     Job_TokenFlush();
   2405 #ifdef RMT_WILL_WATCH
   2406     pnJobs = nJobs;
   2407 
   2408     /*
   2409      * It is possible for us to be called with nJobs equal to 0. This happens
   2410      * if all the jobs finish and a job that is stopped cannot be run
   2411      * locally (eg if maxLocal is 0) and cannot be exported. The job will
   2412      * be placed back on the stoppedJobs queue, Job_Empty() will return false,
   2413      * Make_Run will call us again when there's nothing for which to wait.
   2414      * nJobs never changes, so we loop forever. Hence the check. It could
   2415      * be argued that we should sleep for a bit so as not to swamp the
   2416      * exportation system with requests. Perhaps we should.
   2417      *
   2418      * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
   2419      * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
   2420      * It may use the variable nLocal to determine if it needs to call
   2421      * Job_CatchChildren (if nLocal is 0, there's nothing for which to
   2422      * wait...)
   2423      */
   2424     while (nJobs != 0 && pnJobs == nJobs) {
   2425 	Rmt_Wait();
   2426     }
   2427 #else
   2428     if (usePipes) {
   2429 #ifdef USE_SELECT
   2430 	struct timeval	  timeout;
   2431 	fd_set         	  readfds;
   2432 
   2433 	readfds = outputs;
   2434 	timeout.tv_sec = SEL_SEC;
   2435 	timeout.tv_usec = SEL_USEC;
   2436 
   2437 	if ((nready = select(FD_SETSIZE, &readfds, (fd_set *) 0,
   2438 			   (fd_set *) 0, &timeout)) <= 0)
   2439 	    return;
   2440 #else
   2441 	if ((nready = poll((wantToken ? fds : (fds + 1)),
   2442 	  		   (wantToken ? nfds : (nfds - 1)), POLL_MSEC)) <= 0)
   2443 	    return;
   2444 #endif
   2445 	else {
   2446 	    if (Lst_Open(jobs) == FAILURE) {
   2447 		Punt("Cannot open job table");
   2448 	    }
   2449 	    while (nready && (ln = Lst_Next(jobs)) != NILLNODE) {
   2450 		job = (Job *) Lst_Datum(ln);
   2451 #ifdef USE_SELECT
   2452 		if (FD_ISSET(job->inPipe, &readfds))
   2453 #else
   2454 		if (readyfd(job))
   2455 #endif
   2456 		{
   2457 		    JobDoOutput(job, FALSE);
   2458 		    nready -= 1;
   2459 		}
   2460 
   2461 	    }
   2462 	    Lst_Close(jobs);
   2463 	}
   2464     }
   2465 #endif /* RMT_WILL_WATCH */
   2466 }
   2467 
   2468 /*-
   2469  *-----------------------------------------------------------------------
   2470  * Job_Make --
   2471  *	Start the creation of a target. Basically a front-end for
   2472  *	JobStart used by the Make module.
   2473  *
   2474  * Results:
   2475  *	None.
   2476  *
   2477  * Side Effects:
   2478  *	Another job is started.
   2479  *
   2480  *-----------------------------------------------------------------------
   2481  */
   2482 void
   2483 Job_Make(gn)
   2484     GNode   *gn;
   2485 {
   2486     (void) JobStart(gn, 0, NULL);
   2487 }
   2488 
   2489 /*-
   2490  *-----------------------------------------------------------------------
   2491  * Job_Init --
   2492  *	Initialize the process module
   2493  *
   2494  * Results:
   2495  *	none
   2496  *
   2497  * Side Effects:
   2498  *	lists and counters are initialized
   2499  *-----------------------------------------------------------------------
   2500  */
   2501 void
   2502 Job_Init(maxproc, maxlocal)
   2503     int           maxproc;  /* the greatest number of jobs which may be
   2504 			     * running at one time */
   2505     int	    	  maxlocal; /* the greatest number of local jobs which may
   2506 			     * be running at once. */
   2507 {
   2508     GNode         *begin;     /* node for commands to do at the very start */
   2509 
   2510     jobs =  	  Lst_Init(FALSE);
   2511     stoppedJobs = Lst_Init(FALSE);
   2512     maxJobs = 	  maxproc;
   2513     maxLocal = 	  maxlocal;
   2514     nJobs = 	  0;
   2515     nLocal = 	  0;
   2516     wantToken =	  FALSE;
   2517 
   2518     aborting = 	  0;
   2519     errors = 	  0;
   2520 
   2521     lastNode =	  NILGNODE;
   2522 
   2523     if (maxJobs == 1
   2524 #ifdef REMOTE
   2525 	|| noMessages
   2526 #endif
   2527 		     ) {
   2528 	/*
   2529 	 * If only one job can run at a time, there's no need for a banner,
   2530 	 * is there?
   2531 	 */
   2532 	targFmt = "";
   2533     } else {
   2534 	targFmt = TARG_FMT;
   2535     }
   2536 
   2537     if (shellPath == NULL) {
   2538 	/*
   2539 	 * The user didn't specify a shell to use, so we are using the
   2540 	 * default one... Both the absolute path and the last component
   2541 	 * must be set. The last component is taken from the 'name' field
   2542 	 * of the default shell description pointed-to by commandShell.
   2543 	 * All default shells are located in _PATH_DEFSHELLDIR.
   2544 	 */
   2545 	shellName = commandShell->name;
   2546 	shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
   2547     }
   2548 
   2549     if (commandShell->exit == NULL) {
   2550 	commandShell->exit = "";
   2551     }
   2552     if (commandShell->echo == NULL) {
   2553 	commandShell->echo = "";
   2554     }
   2555 
   2556     /*
   2557      * Catch the four signals that POSIX specifies if they aren't ignored.
   2558      * JobPassSig will take care of calling JobInterrupt if appropriate.
   2559      */
   2560     if (signal(SIGINT, SIG_IGN) != SIG_IGN) {
   2561 	(void) signal(SIGINT, JobPassSig);
   2562     }
   2563     if (signal(SIGHUP, SIG_IGN) != SIG_IGN) {
   2564 	(void) signal(SIGHUP, JobPassSig);
   2565     }
   2566     if (signal(SIGQUIT, SIG_IGN) != SIG_IGN) {
   2567 	(void) signal(SIGQUIT, JobPassSig);
   2568     }
   2569     if (signal(SIGTERM, SIG_IGN) != SIG_IGN) {
   2570 	(void) signal(SIGTERM, JobPassSig);
   2571     }
   2572     /*
   2573      * Install a NOOP  SIGCHLD handler so we are woken up if we're blocked.
   2574      */
   2575     signal(SIGCHLD, JobIgnoreSig);
   2576 
   2577     /*
   2578      * There are additional signals that need to be caught and passed if
   2579      * either the export system wants to be told directly of signals or if
   2580      * we're giving each job its own process group (since then it won't get
   2581      * signals from the terminal driver as we own the terminal)
   2582      */
   2583 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
   2584     if (signal(SIGTSTP, SIG_IGN) != SIG_IGN) {
   2585 	(void) signal(SIGTSTP, JobPassSig);
   2586     }
   2587     if (signal(SIGTTOU, SIG_IGN) != SIG_IGN) {
   2588 	(void) signal(SIGTTOU, JobPassSig);
   2589     }
   2590     if (signal(SIGTTIN, SIG_IGN) != SIG_IGN) {
   2591 	(void) signal(SIGTTIN, JobPassSig);
   2592     }
   2593     if (signal(SIGWINCH, SIG_IGN) != SIG_IGN) {
   2594 	(void) signal(SIGWINCH, JobPassSig);
   2595     }
   2596     if (signal(SIGCONT, SIG_IGN) != SIG_IGN) {
   2597 	(void) signal(SIGCONT, JobContinueSig);
   2598     }
   2599 #endif
   2600 
   2601     begin = Targ_FindNode(".BEGIN", TARG_NOCREATE);
   2602 
   2603     if (begin != NILGNODE) {
   2604 	JobStart(begin, JOB_SPECIAL, (Job *)0);
   2605 	while (nJobs) {
   2606 	    Job_CatchOutput();
   2607 #ifndef RMT_WILL_WATCH
   2608 	    Job_CatchChildren(!usePipes);
   2609 #endif /* RMT_WILL_WATCH */
   2610 	}
   2611     }
   2612     postCommands = Targ_FindNode(".END", TARG_CREATE);
   2613 }
   2614 
   2615 /*-
   2616  *-----------------------------------------------------------------------
   2617  * Job_Empty --
   2618  *	See if the job table is empty.  Because the local concurrency may
   2619  *	be set to 0, it is possible for the job table to become empty,
   2620  *	while the list of stoppedJobs remains non-empty. In such a case,
   2621  *	we want to restart as many jobs as we can.
   2622  *
   2623  * Results:
   2624  *	TRUE if it is. FALSE if it ain't.
   2625  *
   2626  * Side Effects:
   2627  *	None.
   2628  *
   2629  * -----------------------------------------------------------------------
   2630  */
   2631 Boolean
   2632 Job_Empty()
   2633 {
   2634     if (nJobs == 0) {
   2635 	if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
   2636 	    /*
   2637 	     * The job table is obviously not full if it has no jobs in
   2638 	     * it...Try and restart the stopped jobs.
   2639 	     */
   2640 	    JobRestartJobs();
   2641 	    return(FALSE);
   2642 	} else {
   2643 	    return(TRUE);
   2644 	}
   2645     } else {
   2646 	return(FALSE);
   2647     }
   2648 }
   2649 
   2650 /*-
   2651  *-----------------------------------------------------------------------
   2652  * JobMatchShell --
   2653  *	Find a matching shell in 'shells' given its final component.
   2654  *
   2655  * Results:
   2656  *	A pointer to the Shell structure.
   2657  *
   2658  * Side Effects:
   2659  *	None.
   2660  *
   2661  *-----------------------------------------------------------------------
   2662  */
   2663 static Shell *
   2664 JobMatchShell(name)
   2665     char	  *name;      /* Final component of shell path */
   2666 {
   2667     register Shell *sh;	      /* Pointer into shells table */
   2668     Shell	   *match;    /* Longest-matching shell */
   2669     register char *cp1,
   2670 		  *cp2;
   2671     char	  *eoname;
   2672 
   2673     eoname = name + strlen(name);
   2674 
   2675     match = NULL;
   2676 
   2677     for (sh = shells; sh->name != NULL; sh++) {
   2678 	for (cp1 = eoname - strlen(sh->name), cp2 = sh->name;
   2679 	     *cp1 != '\0' && *cp1 == *cp2;
   2680 	     cp1++, cp2++) {
   2681 		 continue;
   2682 	}
   2683 	if (*cp1 != *cp2) {
   2684 	    continue;
   2685 	} else if (match == NULL || strlen(match->name) < strlen(sh->name)) {
   2686 	   match = sh;
   2687 	}
   2688     }
   2689     return(match == NULL ? sh : match);
   2690 }
   2691 
   2692 /*-
   2693  *-----------------------------------------------------------------------
   2694  * Job_ParseShell --
   2695  *	Parse a shell specification and set up commandShell, shellPath
   2696  *	and shellName appropriately.
   2697  *
   2698  * Results:
   2699  *	FAILURE if the specification was incorrect.
   2700  *
   2701  * Side Effects:
   2702  *	commandShell points to a Shell structure (either predefined or
   2703  *	created from the shell spec), shellPath is the full path of the
   2704  *	shell described by commandShell, while shellName is just the
   2705  *	final component of shellPath.
   2706  *
   2707  * Notes:
   2708  *	A shell specification consists of a .SHELL target, with dependency
   2709  *	operator, followed by a series of blank-separated words. Double
   2710  *	quotes can be used to use blanks in words. A backslash escapes
   2711  *	anything (most notably a double-quote and a space) and
   2712  *	provides the functionality it does in C. Each word consists of
   2713  *	keyword and value separated by an equal sign. There should be no
   2714  *	unnecessary spaces in the word. The keywords are as follows:
   2715  *	    name  	    Name of shell.
   2716  *	    path  	    Location of shell. Overrides "name" if given
   2717  *	    quiet 	    Command to turn off echoing.
   2718  *	    echo  	    Command to turn echoing on
   2719  *	    filter	    Result of turning off echoing that shouldn't be
   2720  *	    	  	    printed.
   2721  *	    echoFlag	    Flag to turn echoing on at the start
   2722  *	    errFlag	    Flag to turn error checking on at the start
   2723  *	    hasErrCtl	    True if shell has error checking control
   2724  *	    check 	    Command to turn on error checking if hasErrCtl
   2725  *	    	  	    is TRUE or template of command to echo a command
   2726  *	    	  	    for which error checking is off if hasErrCtl is
   2727  *	    	  	    FALSE.
   2728  *	    ignore	    Command to turn off error checking if hasErrCtl
   2729  *	    	  	    is TRUE or template of command to execute a
   2730  *	    	  	    command so as to ignore any errors it returns if
   2731  *	    	  	    hasErrCtl is FALSE.
   2732  *
   2733  *-----------------------------------------------------------------------
   2734  */
   2735 ReturnStatus
   2736 Job_ParseShell(line)
   2737     char	  *line;  /* The shell spec */
   2738 {
   2739     char    	  **words;
   2740     int	    	  wordCount;
   2741     register char **argv;
   2742     register int  argc;
   2743     char    	  *path;
   2744     Shell   	  newShell;
   2745     Boolean 	  fullSpec = FALSE;
   2746 
   2747     while (isspace((unsigned char)*line)) {
   2748 	line++;
   2749     }
   2750 
   2751     if (shellArgv)
   2752 	free(shellArgv);
   2753 
   2754     words = brk_string(line, &wordCount, TRUE, &shellArgv);
   2755 
   2756     memset((Address)&newShell, 0, sizeof(newShell));
   2757 
   2758     /*
   2759      * Parse the specification by keyword
   2760      */
   2761     for (path = NULL, argc = wordCount - 1, argv = words;
   2762 	argc != 0;
   2763 	argc--, argv++) {
   2764 	    if (strncmp(*argv, "path=", 5) == 0) {
   2765 		path = &argv[0][5];
   2766 	    } else if (strncmp(*argv, "name=", 5) == 0) {
   2767 		newShell.name = &argv[0][5];
   2768 	    } else {
   2769 		if (strncmp(*argv, "quiet=", 6) == 0) {
   2770 		    newShell.echoOff = &argv[0][6];
   2771 		} else if (strncmp(*argv, "echo=", 5) == 0) {
   2772 		    newShell.echoOn = &argv[0][5];
   2773 		} else if (strncmp(*argv, "filter=", 7) == 0) {
   2774 		    newShell.noPrint = &argv[0][7];
   2775 		    newShell.noPLen = strlen(newShell.noPrint);
   2776 		} else if (strncmp(*argv, "echoFlag=", 9) == 0) {
   2777 		    newShell.echo = &argv[0][9];
   2778 		} else if (strncmp(*argv, "errFlag=", 8) == 0) {
   2779 		    newShell.exit = &argv[0][8];
   2780 		} else if (strncmp(*argv, "hasErrCtl=", 10) == 0) {
   2781 		    char c = argv[0][10];
   2782 		    newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
   2783 					   (c != 'T') && (c != 't'));
   2784 		} else if (strncmp(*argv, "check=", 6) == 0) {
   2785 		    newShell.errCheck = &argv[0][6];
   2786 		} else if (strncmp(*argv, "ignore=", 7) == 0) {
   2787 		    newShell.ignErr = &argv[0][7];
   2788 		} else {
   2789 		    Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"",
   2790 				*argv);
   2791 		    free(words);
   2792 		    return(FAILURE);
   2793 		}
   2794 		fullSpec = TRUE;
   2795 	    }
   2796     }
   2797 
   2798     if (path == NULL) {
   2799 	/*
   2800 	 * If no path was given, the user wants one of the pre-defined shells,
   2801 	 * yes? So we find the one s/he wants with the help of JobMatchShell
   2802 	 * and set things up the right way. shellPath will be set up by
   2803 	 * Job_Init.
   2804 	 */
   2805 	if (newShell.name == NULL) {
   2806 	    Parse_Error(PARSE_FATAL, "Neither path nor name specified");
   2807 	    return(FAILURE);
   2808 	} else {
   2809 	    commandShell = JobMatchShell(newShell.name);
   2810 	    shellName = newShell.name;
   2811 	}
   2812     } else {
   2813 	/*
   2814 	 * The user provided a path. If s/he gave nothing else (fullSpec is
   2815 	 * FALSE), try and find a matching shell in the ones we know of.
   2816 	 * Else we just take the specification at its word and copy it
   2817 	 * to a new location. In either case, we need to record the
   2818 	 * path the user gave for the shell.
   2819 	 */
   2820 	shellPath = path;
   2821 	path = strrchr(path, '/');
   2822 	if (path == NULL) {
   2823 	    path = shellPath;
   2824 	} else {
   2825 	    path += 1;
   2826 	}
   2827 	if (newShell.name != NULL) {
   2828 	    shellName = newShell.name;
   2829 	} else {
   2830 	    shellName = path;
   2831 	}
   2832 	if (!fullSpec) {
   2833 	    commandShell = JobMatchShell(shellName);
   2834 	} else {
   2835 	    commandShell = (Shell *) emalloc(sizeof(Shell));
   2836 	    *commandShell = newShell;
   2837 	}
   2838     }
   2839 
   2840     if (commandShell->echoOn && commandShell->echoOff) {
   2841 	commandShell->hasEchoCtl = TRUE;
   2842     }
   2843 
   2844     if (!commandShell->hasErrCtl) {
   2845 	if (commandShell->errCheck == NULL) {
   2846 	    commandShell->errCheck = "";
   2847 	}
   2848 	if (commandShell->ignErr == NULL) {
   2849 	    commandShell->ignErr = "%s\n";
   2850 	}
   2851     }
   2852 
   2853     /*
   2854      * Do not free up the words themselves, since they might be in use by the
   2855      * shell specification.
   2856      */
   2857     free(words);
   2858     return SUCCESS;
   2859 }
   2860 
   2861 /*-
   2862  *-----------------------------------------------------------------------
   2863  * JobInterrupt --
   2864  *	Handle the receipt of an interrupt.
   2865  *
   2866  * Results:
   2867  *	None
   2868  *
   2869  * Side Effects:
   2870  *	All children are killed. Another job will be started if the
   2871  *	.INTERRUPT target was given.
   2872  *-----------------------------------------------------------------------
   2873  */
   2874 static void
   2875 JobInterrupt(runINTERRUPT, signo)
   2876     int	    runINTERRUPT;   	/* Non-zero if commands for the .INTERRUPT
   2877 				 * target should be executed */
   2878     int	    signo;		/* signal received */
   2879 {
   2880     LstNode 	  ln;		/* element in job table */
   2881     Job           *job;	    	/* job descriptor in that element */
   2882     GNode         *interrupt;	/* the node describing the .INTERRUPT target */
   2883 
   2884     aborting = ABORT_INTERRUPT;
   2885 
   2886    (void) Lst_Open(jobs);
   2887     while ((ln = Lst_Next(jobs)) != NILLNODE) {
   2888 	job = (Job *) Lst_Datum(ln);
   2889 
   2890 	if (!Targ_Precious(job->node)) {
   2891 	    char  	*file = (job->node->path == NULL ?
   2892 				 job->node->name :
   2893 				 job->node->path);
   2894 	    if (!noExecute && eunlink(file) != -1) {
   2895 		Error("*** %s removed", file);
   2896 	    }
   2897 	}
   2898 #ifdef RMT_WANTS_SIGNALS
   2899 	if (job->flags & JOB_REMOTE) {
   2900 	    /*
   2901 	     * If job is remote, let the Rmt module do the killing.
   2902 	     */
   2903 	    if (!Rmt_Signal(job, signo)) {
   2904 		/*
   2905 		 * If couldn't kill the thing, finish it out now with an
   2906 		 * error code, since no exit report will come in likely.
   2907 		 */
   2908 		int status;
   2909 
   2910 		status.w_status = 0;
   2911 		status.w_retcode = 1;
   2912 		JobFinish(job, &status);
   2913 	    }
   2914 	} else if (job->pid) {
   2915 	    KILL(job->pid, signo);
   2916 	}
   2917 #else
   2918 	if (job->pid) {
   2919 	    if (DEBUG(JOB)) {
   2920 		(void) fprintf(stdout,
   2921 			       "JobInterrupt passing signal to child %d.\n",
   2922 			       job->pid);
   2923 		(void) fflush(stdout);
   2924 	    }
   2925 	    KILL(job->pid, signo);
   2926 	}
   2927 #endif /* RMT_WANTS_SIGNALS */
   2928     }
   2929 
   2930 #ifdef REMOTE
   2931    (void)Lst_Open(stoppedJobs);
   2932     while ((ln = Lst_Next(stoppedJobs)) != NILLNODE) {
   2933 	job = (Job *) Lst_Datum(ln);
   2934 
   2935 	if (job->flags & JOB_RESTART) {
   2936 	    if (DEBUG(JOB)) {
   2937 		(void) fprintf(stdout, "%s%s",
   2938 			       "JobInterrupt skipping job on stopped queue",
   2939 			       "-- it was waiting to be restarted.\n");
   2940 		(void) fflush(stdout);
   2941 	    }
   2942 	    continue;
   2943 	}
   2944 	if (!Targ_Precious(job->node)) {
   2945 	    char  	*file = (job->node->path == NULL ?
   2946 				 job->node->name :
   2947 				 job->node->path);
   2948 	    if (eunlink(file) == 0) {
   2949 		Error("*** %s removed", file);
   2950 	    }
   2951 	}
   2952 	/*
   2953 	 * Resume the thing so it will take the signal.
   2954 	 */
   2955 	if (DEBUG(JOB)) {
   2956 	    (void) fprintf(stdout,
   2957 			   "JobInterrupt passing CONT to stopped child %d.\n",
   2958 			   job->pid);
   2959 	    (void) fflush(stdout);
   2960 	}
   2961 	KILL(job->pid, SIGCONT);
   2962 #ifdef RMT_WANTS_SIGNALS
   2963 	if (job->flags & JOB_REMOTE) {
   2964 	    /*
   2965 	     * If job is remote, let the Rmt module do the killing.
   2966 	     */
   2967 	    if (!Rmt_Signal(job, SIGINT)) {
   2968 		/*
   2969 		 * If couldn't kill the thing, finish it out now with an
   2970 		 * error code, since no exit report will come in likely.
   2971 		 */
   2972 		int status;
   2973 		status.w_status = 0;
   2974 		status.w_retcode = 1;
   2975 		JobFinish(job, &status);
   2976 	    }
   2977 	} else if (job->pid) {
   2978 	    if (DEBUG(JOB)) {
   2979 		(void) fprintf(stdout,
   2980 		       "JobInterrupt passing interrupt to stopped child %d.\n",
   2981 			       job->pid);
   2982 		(void) fflush(stdout);
   2983 	    }
   2984 	    KILL(job->pid, SIGINT);
   2985 	}
   2986 #endif /* RMT_WANTS_SIGNALS */
   2987     }
   2988 #endif
   2989     Lst_Close(stoppedJobs);
   2990 
   2991     if (runINTERRUPT && !touchFlag) {
   2992 	interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE);
   2993 	if (interrupt != NILGNODE) {
   2994 	    ignoreErrors = FALSE;
   2995 
   2996 	    JobStart(interrupt, JOB_IGNDOTS, (Job *)0);
   2997 	    while (nJobs) {
   2998 		Job_CatchOutput();
   2999 #ifndef RMT_WILL_WATCH
   3000 		Job_CatchChildren(!usePipes);
   3001 #endif /* RMT_WILL_WATCH */
   3002 	    }
   3003 	}
   3004     }
   3005     Trace_Log(MAKEINTR, 0);
   3006     exit(signo);
   3007 }
   3008 
   3009 /*
   3010  *-----------------------------------------------------------------------
   3011  * Job_Finish --
   3012  *	Do final processing such as the running of the commands
   3013  *	attached to the .END target.
   3014  *
   3015  * Results:
   3016  *	Number of errors reported.
   3017  *
   3018  * Side Effects:
   3019  *	None.
   3020  *-----------------------------------------------------------------------
   3021  */
   3022 int
   3023 Job_Finish()
   3024 {
   3025     if (postCommands != NILGNODE && !Lst_IsEmpty(postCommands->commands)) {
   3026 	if (errors) {
   3027 	    Error("Errors reported so .END ignored");
   3028 	} else {
   3029 	    JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL);
   3030 
   3031 	    while (nJobs) {
   3032 		Job_CatchOutput();
   3033 #ifndef RMT_WILL_WATCH
   3034 		Job_CatchChildren(!usePipes);
   3035 #endif /* RMT_WILL_WATCH */
   3036 	    }
   3037 	}
   3038     }
   3039     Job_TokenFlush();
   3040     return(errors);
   3041 }
   3042 
   3043 /*-
   3044  *-----------------------------------------------------------------------
   3045  * Job_End --
   3046  *	Cleanup any memory used by the jobs module
   3047  *
   3048  * Results:
   3049  *	None.
   3050  *
   3051  * Side Effects:
   3052  *	Memory is freed
   3053  *-----------------------------------------------------------------------
   3054  */
   3055 void
   3056 Job_End()
   3057 {
   3058 #ifdef CLEANUP
   3059     if (shellArgv)
   3060 	free(shellArgv);
   3061 #endif
   3062 }
   3063 
   3064 /*-
   3065  *-----------------------------------------------------------------------
   3066  * Job_Wait --
   3067  *	Waits for all running jobs to finish and returns. Sets 'aborting'
   3068  *	to ABORT_WAIT to prevent other jobs from starting.
   3069  *
   3070  * Results:
   3071  *	None.
   3072  *
   3073  * Side Effects:
   3074  *	Currently running jobs finish.
   3075  *
   3076  *-----------------------------------------------------------------------
   3077  */
   3078 void
   3079 Job_Wait()
   3080 {
   3081     aborting = ABORT_WAIT;
   3082     while (nJobs != 0) {
   3083 	Job_CatchOutput();
   3084 #ifndef RMT_WILL_WATCH
   3085 	Job_CatchChildren(!usePipes);
   3086 #endif /* RMT_WILL_WATCH */
   3087     }
   3088     Job_TokenFlush();
   3089     aborting = 0;
   3090 }
   3091 
   3092 /*-
   3093  *-----------------------------------------------------------------------
   3094  * Job_AbortAll --
   3095  *	Abort all currently running jobs without handling output or anything.
   3096  *	This function is to be called only in the event of a major
   3097  *	error. Most definitely NOT to be called from JobInterrupt.
   3098  *
   3099  * Results:
   3100  *	None
   3101  *
   3102  * Side Effects:
   3103  *	All children are killed, not just the firstborn
   3104  *-----------------------------------------------------------------------
   3105  */
   3106 void
   3107 Job_AbortAll()
   3108 {
   3109     LstNode           	ln;	/* element in job table */
   3110     Job            	*job;	/* the job descriptor in that element */
   3111     int     	  	foo;
   3112 
   3113     aborting = ABORT_ERROR;
   3114 
   3115     if (nJobs) {
   3116 
   3117 	(void) Lst_Open(jobs);
   3118 	while ((ln = Lst_Next(jobs)) != NILLNODE) {
   3119 	    job = (Job *) Lst_Datum(ln);
   3120 
   3121 	    /*
   3122 	     * kill the child process with increasingly drastic signals to make
   3123 	     * darn sure it's dead.
   3124 	     */
   3125 #ifdef RMT_WANTS_SIGNALS
   3126 	    if (job->flags & JOB_REMOTE) {
   3127 		Rmt_Signal(job, SIGINT);
   3128 		Rmt_Signal(job, SIGKILL);
   3129 	    } else {
   3130 		KILL(job->pid, SIGINT);
   3131 		KILL(job->pid, SIGKILL);
   3132 	    }
   3133 #else
   3134 	    KILL(job->pid, SIGINT);
   3135 	    KILL(job->pid, SIGKILL);
   3136 #endif /* RMT_WANTS_SIGNALS */
   3137 	}
   3138     }
   3139 
   3140     /*
   3141      * Catch as many children as want to report in at first, then give up
   3142      */
   3143     while (waitpid((pid_t) -1, &foo, WNOHANG) > 0)
   3144 	continue;
   3145 }
   3146 
   3147 #ifdef REMOTE
   3148 /*-
   3149  *-----------------------------------------------------------------------
   3150  * JobFlagForMigration --
   3151  *	Handle the eviction of a child. Called from RmtStatusChange.
   3152  *	Flags the child as remigratable and then suspends it.
   3153  *
   3154  * Results:
   3155  *	none.
   3156  *
   3157  * Side Effects:
   3158  *	The job descriptor is flagged for remigration.
   3159  *
   3160  *-----------------------------------------------------------------------
   3161  */
   3162 void
   3163 JobFlagForMigration(hostID)
   3164     int 	  hostID;    	/* ID of host we used, for matching children. */
   3165 {
   3166     register Job  *job;	    	/* job descriptor for dead child */
   3167     LstNode       jnode;    	/* list element for finding job */
   3168 
   3169     if (DEBUG(JOB)) {
   3170 	(void) fprintf(stdout, "JobFlagForMigration(%d) called.\n", hostID);
   3171 	(void) fflush(stdout);
   3172     }
   3173     jnode = Lst_Find(jobs, (ClientData)hostID, JobCmpRmtID);
   3174 
   3175     if (jnode == NILLNODE) {
   3176 	jnode = Lst_Find(stoppedJobs, (ClientData)hostID, JobCmpRmtID);
   3177 		if (jnode == NILLNODE) {
   3178 		    if (DEBUG(JOB)) {
   3179 			Error("Evicting host(%d) not in table", hostID);
   3180 		    }
   3181 		    return;
   3182 		}
   3183     }
   3184     job = (Job *) Lst_Datum(jnode);
   3185 
   3186     if (DEBUG(JOB)) {
   3187 	(void) fprintf(stdout,
   3188 		       "JobFlagForMigration(%d) found job '%s'.\n", hostID,
   3189 		       job->node->name);
   3190 	(void) fflush(stdout);
   3191     }
   3192 
   3193     KILL(job->pid, SIGSTOP);
   3194 
   3195     job->flags |= JOB_REMIGRATE;
   3196 }
   3197 
   3198 #endif
   3199 
   3200 /*-
   3202  *-----------------------------------------------------------------------
   3203  * JobRestartJobs --
   3204  *	Tries to restart stopped jobs if there are slots available.
   3205  *	Note that this tries to restart them regardless of pending errors.
   3206  *	It's not good to leave stopped jobs lying around!
   3207  *
   3208  * Results:
   3209  *	None.
   3210  *
   3211  * Side Effects:
   3212  *	Resumes(and possibly migrates) jobs.
   3213  *
   3214  *-----------------------------------------------------------------------
   3215  */
   3216 static void
   3217 JobRestartJobs()
   3218 {
   3219     while (!Lst_IsEmpty(stoppedJobs)) {
   3220 	if (DEBUG(JOB)) {
   3221 	    (void) fprintf(stdout, "Restarting a stopped job.\n");
   3222 	    (void) fflush(stdout);
   3223 	}
   3224 	JobRestart((Job *)Lst_DeQueue(stoppedJobs));
   3225     }
   3226 }
   3227 
   3228 #ifndef RMT_WILL_WATCH
   3229 #ifndef USE_SELECT
   3230 static void
   3231 watchfd(job)
   3232     Job *job;
   3233 {
   3234     int i;
   3235     if (job->inPollfd != NULL)
   3236 	Punt("Watching watched job");
   3237     if (fds == NULL) {
   3238 	maxfds = JBSTART;
   3239 	fds = emalloc(sizeof(struct pollfd) * maxfds);
   3240 	jobfds = emalloc(sizeof(Job **) * maxfds);
   3241 
   3242 	fds[0].fd = job_pipe[0];
   3243 	fds[0].events = POLLIN;
   3244 	jobfds[0] = &tokenWaitJob;
   3245 	tokenWaitJob.inPollfd = &fds[0];
   3246 	nfds++;
   3247     } else if (nfds == maxfds) {
   3248 	maxfds *= JBFACTOR;
   3249 	fds = erealloc(fds, sizeof(struct pollfd) * maxfds);
   3250 	jobfds = erealloc(jobfds, sizeof(Job **) * maxfds);
   3251 	for (i = 0; i < nfds; i++)
   3252 	    jobfds[i]->inPollfd = &fds[i];
   3253     }
   3254 
   3255     fds[nfds].fd = job->inPipe;
   3256     fds[nfds].events = POLLIN;
   3257     jobfds[nfds] = job;
   3258     job->inPollfd = &fds[nfds];
   3259     nfds++;
   3260 }
   3261 
   3262 static void
   3263 clearfd(job)
   3264     Job *job;
   3265 {
   3266     int i;
   3267     if (job->inPollfd == NULL)
   3268 	Punt("Unwatching unwatched job");
   3269     i = job->inPollfd - fds;
   3270     nfds--;
   3271     /*
   3272      * Move last job in table into hole made by dead job.
   3273      */
   3274     if (nfds != i) {
   3275 	fds[i] = fds[nfds];
   3276 	jobfds[i] = jobfds[nfds];
   3277 	jobfds[i]->inPollfd = &fds[i];
   3278     }
   3279     job->inPollfd = NULL;
   3280 }
   3281 
   3282 static int
   3283 readyfd(job)
   3284     Job *job;
   3285 {
   3286     if (job->inPollfd == NULL)
   3287 	Punt("Polling unwatched job");
   3288     return (job->inPollfd->revents & POLLIN) != 0;
   3289 }
   3290 #endif
   3291 #endif
   3292 
   3293 /*-
   3294  *-----------------------------------------------------------------------
   3295  * JobTokenAdd --
   3296  *	Put a token into the job pipe so that some make process can start
   3297  *	another job.
   3298  *
   3299  * Side Effects:
   3300  *	Allows more build jobs to be spawned somewhere.
   3301  *
   3302  *-----------------------------------------------------------------------
   3303  */
   3304 
   3305 static void
   3306 JobTokenAdd()
   3307 {
   3308 
   3309     if (DEBUG(JOB))
   3310 	printf("deposit token\n");
   3311     write(job_pipe[1], "+", 1);
   3312 }
   3313 
   3314 /*-
   3315  *-----------------------------------------------------------------------
   3316  * Job_ServerStartTokenAdd --
   3317  *	Prep the job token pipe in the root make process.
   3318  *
   3319  *-----------------------------------------------------------------------
   3320  */
   3321 
   3322 void Job_ServerStart(maxproc)
   3323     int maxproc;
   3324 {
   3325     int i, flags;
   3326     char jobarg[64];
   3327 
   3328     if (pipe(job_pipe) < 0)
   3329 	Fatal ("error in pipe: %s", strerror(errno));
   3330 
   3331     /*
   3332      * We mark the input side of the pipe non-blocking; we poll(2) the
   3333      * pipe when we're waiting for a job token, but we might lose the
   3334      * race for the token when a new one becomes available, so the read
   3335      * from the pipe should not block.
   3336      */
   3337     flags = fcntl(job_pipe[0], F_GETFL, 0);
   3338     flags |= O_NONBLOCK;
   3339     fcntl(job_pipe[0], F_SETFL, flags);
   3340 
   3341     /*
   3342      * Mark job pipes as close-on-exec.
   3343      * Note that we will clear this when executing submakes.
   3344      */
   3345     fcntl(job_pipe[0], F_SETFD, 1);
   3346     fcntl(job_pipe[1], F_SETFD, 1);
   3347 
   3348     snprintf(jobarg, sizeof(jobarg), "%d,%d", job_pipe[0], job_pipe[1]);
   3349 
   3350     Var_Append(MAKEFLAGS, "-J", VAR_GLOBAL);
   3351     Var_Append(MAKEFLAGS, jobarg, VAR_GLOBAL);
   3352 
   3353     /*
   3354      * Preload job_pipe with one token per job, save the one
   3355      * "extra" token for the primary job.
   3356      *
   3357      * XXX should clip maxJobs against PIPE_BUF -- if maxJobs is
   3358      * larger than the write buffer size of the pipe, we will
   3359      * deadlock here.
   3360      */
   3361     for (i=1; i < maxproc; i++)
   3362 	JobTokenAdd();
   3363 }
   3364 
   3365 /*
   3366  * this tracks the number of tokens currently "out" to build jobs.
   3367  */
   3368 int jobTokensRunning = 0;
   3369 int jobTokensFree = 0;
   3370 /*-
   3371  *-----------------------------------------------------------------------
   3372  * Job_TokenReturn --
   3373  *	Return a withdrawn token to the pool.
   3374  *
   3375  *-----------------------------------------------------------------------
   3376  */
   3377 
   3378 void
   3379 Job_TokenReturn()
   3380 {
   3381     jobTokensRunning--;
   3382     if (jobTokensRunning < 0)
   3383 	Punt("token botch");
   3384     if (jobTokensRunning)
   3385 	jobTokensFree++;
   3386 }
   3387 
   3388 /*-
   3389  *-----------------------------------------------------------------------
   3390  * Job_TokenWithdraw --
   3391  *	Attempt to withdraw a token from the pool.
   3392  *
   3393  * Results:
   3394  *	Returns TRUE if a token was withdrawn, and FALSE if the pool
   3395  *	is currently empty.
   3396  *
   3397  * Side Effects:
   3398  * 	If pool is empty, set wantToken so that we wake up
   3399  *	when a token is released.
   3400  *
   3401  *-----------------------------------------------------------------------
   3402  */
   3403 
   3404 
   3405 Boolean
   3406 Job_TokenWithdraw()
   3407 {
   3408     char tok;
   3409     int count;
   3410 
   3411     if (aborting)
   3412 	    return FALSE;
   3413 
   3414     if (jobTokensRunning == 0) {
   3415 	if (DEBUG(JOB))
   3416 	    printf("first one's free\n");
   3417 	jobTokensRunning++;
   3418 	wantToken = FALSE;
   3419 	return TRUE;
   3420     }
   3421     if (jobTokensFree > 0) {
   3422 	jobTokensFree--;
   3423 	jobTokensRunning++;
   3424 	wantToken = FALSE;
   3425 	return TRUE;
   3426     }
   3427     count = read(job_pipe[0], &tok, 1);
   3428     if (count == 0)
   3429 	Fatal("eof on job pipe!");
   3430     else if (count < 0) {
   3431 	if (errno != EAGAIN) {
   3432 	    Fatal("job pipe read: %s", strerror(errno));
   3433 	}
   3434 	if (DEBUG(JOB))
   3435 	    printf("blocked for token\n");
   3436 	wantToken = TRUE;
   3437 	return FALSE;
   3438     }
   3439     wantToken = FALSE;
   3440     jobTokensRunning++;
   3441     if (DEBUG(JOB))
   3442 	printf("withdrew token\n");
   3443     return TRUE;
   3444 }
   3445 
   3446 /*-
   3447  *-----------------------------------------------------------------------
   3448  * Job_TokenFlush --
   3449  *	Return free tokens to the pool.
   3450  *
   3451  *-----------------------------------------------------------------------
   3452  */
   3453 
   3454 void
   3455 Job_TokenFlush()
   3456 {
   3457     if (compatMake) return;
   3458 
   3459     while (jobTokensFree > 0) {
   3460 	JobTokenAdd();
   3461 	jobTokensFree--;
   3462     }
   3463 }
   3464 
   3465