Home | History | Annotate | Line # | Download | only in make
job.c revision 1.51
      1 /*	$NetBSD: job.c,v 1.51 2001/06/12 23:36:17 sjg 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 1.51 2001/06/12 23:36:17 sjg 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 1.51 2001/06/12 23:36:17 sjg 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     LstNode 	  cmdNode;  	    /* Node for replacing the command */
    600     char     	  *cmd = (char *) cmdp;
    601     Job           *job = (Job *) jobp;
    602     char	*cp;
    603 
    604     noSpecials = NoExecute(job->node);
    605 
    606     if (strcmp(cmd, "...") == 0) {
    607 	job->node->type |= OP_SAVE_CMDS;
    608 	if ((job->flags & JOB_IGNDOTS) == 0) {
    609 	    job->tailCmds = Lst_Succ(Lst_Member(job->node->commands,
    610 						(ClientData)cmd));
    611 	    return 1;
    612 	}
    613 	return 0;
    614     }
    615 
    616 #define DBPRINTF(fmt, arg) if (DEBUG(JOB)) {	\
    617 	(void) fprintf(stdout, fmt, arg); 	\
    618 	(void) fflush(stdout); 			\
    619     }						\
    620    (void) fprintf(job->cmdFILE, fmt, arg);	\
    621    (void) fflush(job->cmdFILE);
    622 
    623     numCommands += 1;
    624 
    625     /*
    626      * For debugging, we replace each command with the result of expanding
    627      * the variables in the command.
    628      */
    629     cmdNode = Lst_Member(job->node->commands, (ClientData)cmd);
    630     cmdStart = cmd = Var_Subst(NULL, cmd, job->node, FALSE);
    631     Lst_Replace(cmdNode, (ClientData)cmdStart);
    632 
    633     cmdTemplate = "%s\n";
    634 
    635     /*
    636      * Check for leading @' and -'s to control echoing and error checking.
    637      */
    638     while (*cmd == '@' || *cmd == '-') {
    639 	if (*cmd == '@') {
    640 	    shutUp = TRUE;
    641 	} else {
    642 	    errOff = TRUE;
    643 	}
    644 	cmd++;
    645     }
    646 
    647     while (isspace((unsigned char) *cmd))
    648 	cmd++;
    649 
    650     if (shutUp) {
    651 	if (!(job->flags & JOB_SILENT) && !noSpecials &&
    652 	    commandShell->hasEchoCtl) {
    653 		DBPRINTF("%s\n", commandShell->echoOff);
    654 	} else {
    655 	    shutUp = FALSE;
    656 	}
    657     }
    658 
    659     if (errOff) {
    660 	if ( !(job->flags & JOB_IGNERR) && !noSpecials) {
    661 	    if (commandShell->hasErrCtl) {
    662 		/*
    663 		 * we don't want the error-control commands showing
    664 		 * up either, so we turn off echoing while executing
    665 		 * them. We could put another field in the shell
    666 		 * structure to tell JobDoOutput to look for this
    667 		 * string too, but why make it any more complex than
    668 		 * it already is?
    669 		 */
    670 		if (!(job->flags & JOB_SILENT) && !shutUp &&
    671 		    commandShell->hasEchoCtl) {
    672 			DBPRINTF("%s\n", commandShell->echoOff);
    673 			DBPRINTF("%s\n", commandShell->ignErr);
    674 			DBPRINTF("%s\n", commandShell->echoOn);
    675 		} else {
    676 		    DBPRINTF("%s\n", commandShell->ignErr);
    677 		}
    678 	    } else if (commandShell->ignErr &&
    679 		      (*commandShell->ignErr != '\0'))
    680 	    {
    681 		/*
    682 		 * The shell has no error control, so we need to be
    683 		 * weird to get it to ignore any errors from the command.
    684 		 * If echoing is turned on, we turn it off and use the
    685 		 * errCheck template to echo the command. Leave echoing
    686 		 * off so the user doesn't see the weirdness we go through
    687 		 * to ignore errors. Set cmdTemplate to use the weirdness
    688 		 * instead of the simple "%s\n" template.
    689 		 */
    690 		if (!(job->flags & JOB_SILENT) && !shutUp &&
    691 		    commandShell->hasEchoCtl) {
    692 			DBPRINTF("%s\n", commandShell->echoOff);
    693 			DBPRINTF(commandShell->errCheck, cmd);
    694 			shutUp = TRUE;
    695 		}
    696 		cmdTemplate = commandShell->ignErr;
    697 		/*
    698 		 * The error ignoration (hee hee) is already taken care
    699 		 * of by the ignErr template, so pretend error checking
    700 		 * is still on.
    701 		 */
    702 		errOff = FALSE;
    703 	    } else {
    704 		errOff = FALSE;
    705 	    }
    706 	} else {
    707 	    errOff = FALSE;
    708 	}
    709     }
    710 
    711     if (DEBUG(SHELL) && strcmp(shellName, "sh") == 0 &&
    712 	(job->flags & JOB_TRACED) == 0) {
    713 	    DBPRINTF("set -%s\n", "x");
    714 	    job->flags |= JOB_TRACED;
    715     }
    716 
    717     if ((cp = Check_Cwd_Cmd(cmd)) != NULL) {
    718 	    DBPRINTF("cd %s; ", cp);
    719     }
    720     DBPRINTF(cmdTemplate, cmd);
    721     free(cmdStart);
    722 
    723     if (errOff) {
    724 	/*
    725 	 * If echoing is already off, there's no point in issuing the
    726 	 * echoOff command. Otherwise we issue it and pretend it was on
    727 	 * for the whole command...
    728 	 */
    729 	if (!shutUp && !(job->flags & JOB_SILENT) && commandShell->hasEchoCtl){
    730 	    DBPRINTF("%s\n", commandShell->echoOff);
    731 	    shutUp = TRUE;
    732 	}
    733 	DBPRINTF("%s\n", commandShell->errCheck);
    734     }
    735     if (shutUp) {
    736 	DBPRINTF("%s\n", commandShell->echoOn);
    737     }
    738     return 0;
    739 }
    740 
    741 /*-
    742  *-----------------------------------------------------------------------
    743  * JobSaveCommand --
    744  *	Save a command to be executed when everything else is done.
    745  *	Callback function for JobFinish...
    746  *
    747  * Results:
    748  *	Always returns 0
    749  *
    750  * Side Effects:
    751  *	The command is tacked onto the end of postCommands's commands list.
    752  *
    753  *-----------------------------------------------------------------------
    754  */
    755 static int
    756 JobSaveCommand(cmd, gn)
    757     ClientData   cmd;
    758     ClientData   gn;
    759 {
    760     cmd = (ClientData) Var_Subst(NULL, (char *) cmd, (GNode *) gn, FALSE);
    761     (void) Lst_AtEnd(postCommands->commands, cmd);
    762     return(0);
    763 }
    764 
    765 
    766 /*-
    767  *-----------------------------------------------------------------------
    768  * JobClose --
    769  *	Called to close both input and output pipes when a job is finished.
    770  *
    771  * Results:
    772  *	Nada
    773  *
    774  * Side Effects:
    775  *	The file descriptors associated with the job are closed.
    776  *
    777  *-----------------------------------------------------------------------
    778  */
    779 static void
    780 JobClose(job)
    781     Job *job;
    782 {
    783     if (usePipes && (job->flags & JOB_FIRST)) {
    784 #ifdef RMT_WILL_WATCH
    785 	Rmt_Ignore(job->inPipe);
    786 #else
    787 #ifdef USE_SELECT
    788 	FD_CLR(job->inPipe, &outputs);
    789 #else
    790 	clearfd(job);
    791 #endif
    792 #endif
    793 	if (job->outPipe != job->inPipe) {
    794 	   (void) close(job->outPipe);
    795 	}
    796 	JobDoOutput(job, TRUE);
    797 	(void) close(job->inPipe);
    798     } else {
    799 	(void) close(job->outFd);
    800 	JobDoOutput(job, TRUE);
    801     }
    802 }
    803 
    804 /*-
    805  *-----------------------------------------------------------------------
    806  * JobFinish  --
    807  *	Do final processing for the given job including updating
    808  *	parents and starting new jobs as available/necessary. Note
    809  *	that we pay no attention to the JOB_IGNERR flag here.
    810  *	This is because when we're called because of a noexecute flag
    811  *	or something, jstat.w_status is 0 and when called from
    812  *	Job_CatchChildren, the status is zeroed if it s/b ignored.
    813  *
    814  * Results:
    815  *	None
    816  *
    817  * Side Effects:
    818  *	Some nodes may be put on the toBeMade queue.
    819  *	Final commands for the job are placed on postCommands.
    820  *
    821  *	If we got an error and are aborting (aborting == ABORT_ERROR) and
    822  *	the job list is now empty, we are done for the day.
    823  *	If we recognized an error (errors !=0), we set the aborting flag
    824  *	to ABORT_ERROR so no more jobs will be started.
    825  *-----------------------------------------------------------------------
    826  */
    827 /*ARGSUSED*/
    828 static void
    829 JobFinish(job, status)
    830     Job         *job;	      	  /* job to finish */
    831     int	  	*status;     	  /* sub-why job went away */
    832 {
    833     Boolean 	 done;
    834 
    835     if ((WIFEXITED(*status) &&
    836 	 (((WEXITSTATUS(*status) != 0) && !(job->flags & JOB_IGNERR)))) ||
    837 	WIFSIGNALED(*status))
    838     {
    839 	/*
    840 	 * If it exited non-zero and either we're doing things our
    841 	 * way or we're not ignoring errors, the job is finished.
    842 	 * Similarly, if the shell died because of a signal
    843 	 * the job is also finished. In these
    844 	 * cases, finish out the job's output before printing the exit
    845 	 * status...
    846 	 */
    847 #ifdef REMOTE
    848 	KILL(job->pid, SIGCONT);
    849 #endif
    850 	JobClose(job);
    851 	if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
    852 	   (void) fclose(job->cmdFILE);
    853 	   job->cmdFILE = NULL;
    854 	}
    855 	done = TRUE;
    856 #ifdef REMOTE
    857 	if (job->flags & JOB_REMOTE)
    858 	    Rmt_Done(job->rmtID, job->node);
    859 #endif
    860     } else if (WIFEXITED(*status)) {
    861 	/*
    862 	 * Deal with ignored errors in -B mode. We need to print a message
    863 	 * telling of the ignored error as well as setting status.w_status
    864 	 * to 0 so the next command gets run. To do this, we set done to be
    865 	 * TRUE if in -B mode and the job exited non-zero.
    866 	 */
    867 	done = WEXITSTATUS(*status) != 0;
    868 	/*
    869 	 * Old comment said: "Note we don't
    870 	 * want to close down any of the streams until we know we're at the
    871 	 * end."
    872 	 * But we do. Otherwise when are we going to print the rest of the
    873 	 * stuff?
    874 	 */
    875 	JobClose(job);
    876 #ifdef REMOTE
    877 	if (job->flags & JOB_REMOTE)
    878 	    Rmt_Done(job->rmtID, job->node);
    879 #endif /* REMOTE */
    880     } else {
    881 	/*
    882 	 * No need to close things down or anything.
    883 	 */
    884 	done = FALSE;
    885     }
    886 
    887     if (done ||
    888 	WIFSTOPPED(*status) ||
    889 	(WIFSIGNALED(*status) && (WTERMSIG(*status) == SIGCONT)))
    890     {
    891 	FILE	  *out;
    892 
    893 	if (compatMake && !usePipes && (job->flags & JOB_IGNERR)) {
    894 	    /*
    895 	     * If output is going to a file and this job is ignoring
    896 	     * errors, arrange to have the exit status sent to the
    897 	     * output file as well.
    898 	     */
    899 	    out = fdopen(job->outFd, "w");
    900 	    if (out == NULL)
    901 		Punt("Cannot fdopen");
    902 	} else {
    903 	    out = stdout;
    904 	}
    905 
    906 	if (WIFEXITED(*status)) {
    907 	    if (DEBUG(JOB)) {
    908 		(void) fprintf(stdout, "Process %d exited.\n", job->pid);
    909 		(void) fflush(stdout);
    910 	    }
    911 	    if (WEXITSTATUS(*status) != 0) {
    912 		if (usePipes && job->node != lastNode) {
    913 		    MESSAGE(out, job->node);
    914 		    lastNode = job->node;
    915 		}
    916 		(void) fprintf(out, "*** Error code %d%s\n",
    917 			       WEXITSTATUS(*status),
    918 			       (job->flags & JOB_IGNERR) ? "(ignored)" : "");
    919 
    920 		if (job->flags & JOB_IGNERR) {
    921 		    *status = 0;
    922 		}
    923 	    } else if (DEBUG(JOB)) {
    924 		if (usePipes && job->node != lastNode) {
    925 		    MESSAGE(out, job->node);
    926 		    lastNode = job->node;
    927 		}
    928 		(void) fprintf(out, "*** Completed successfully\n");
    929 	    }
    930 	} else if (WIFSTOPPED(*status) && WSTOPSIG(*status) != SIGCONT) {
    931 	    if (DEBUG(JOB)) {
    932 		(void) fprintf(stdout, "Process %d stopped.\n", job->pid);
    933 		(void) fflush(stdout);
    934 	    }
    935 	    if (usePipes && job->node != lastNode) {
    936 		MESSAGE(out, job->node);
    937 		lastNode = job->node;
    938 	    }
    939 	    if (!(job->flags & JOB_REMIGRATE)) {
    940 		switch (WSTOPSIG(*status)) {
    941 		case SIGTSTP:
    942 		    (void) fprintf(out, "*** Suspended\n");
    943 		    break;
    944 		case SIGSTOP:
    945 		    (void) fprintf(out, "*** Stopped\n");
    946 		    break;
    947 		default:
    948 		    (void) fprintf(out, "*** Stopped -- signal %d\n",
    949 			WSTOPSIG(*status));
    950 		}
    951 	    }
    952 	    job->flags |= JOB_RESUME;
    953 	    (void)Lst_AtEnd(stoppedJobs, (ClientData)job);
    954 #ifdef REMOTE
    955 	    if (job->flags & JOB_REMIGRATE)
    956 		JobRestart(job);
    957 #endif
    958 	    (void) fflush(out);
    959 	    return;
    960 	} else if (WIFSTOPPED(*status) &&  WSTOPSIG(*status) == SIGCONT) {
    961 	    /*
    962 	     * If the beastie has continued, shift the Job from the stopped
    963 	     * list to the running one (or re-stop it if concurrency is
    964 	     * exceeded) and go and get another child.
    965 	     */
    966 	    if (job->flags & (JOB_RESUME|JOB_REMIGRATE|JOB_RESTART)) {
    967 		if (usePipes && job->node != lastNode) {
    968 		    MESSAGE(out, job->node);
    969 		    lastNode = job->node;
    970 		}
    971 		(void) fprintf(out, "*** Continued\n");
    972 	    }
    973 	    if (!(job->flags & JOB_CONTINUING)) {
    974 		if (DEBUG(JOB)) {
    975 		    (void) fprintf(stdout,
    976 				   "Warning: process %d was not continuing.\n",
    977 				   job->pid);
    978 		    (void) fflush(stdout);
    979 		}
    980 #ifdef notdef
    981 		/*
    982 		 * We don't really want to restart a job from scratch just
    983 		 * because it continued, especially not without killing the
    984 		 * continuing process!  That's why this is ifdef'ed out.
    985 		 * FD - 9/17/90
    986 		 */
    987 		JobRestart(job);
    988 #endif
    989 	    }
    990 	    job->flags &= ~JOB_CONTINUING;
    991  	    Lst_AtEnd(jobs, (ClientData)job);
    992 	    nJobs += 1;
    993 	    if (!(job->flags & JOB_REMOTE)) {
    994 		if (DEBUG(JOB)) {
    995 		    (void) fprintf(stdout,
    996 				   "Process %d is continuing locally.\n",
    997 				   job->pid);
    998 		    (void) fflush(stdout);
    999   		}
   1000 		nLocal += 1;
   1001 	    }
   1002 	    (void) fflush(out);
   1003   	    return;
   1004 	} else {
   1005 	    if (usePipes && job->node != lastNode) {
   1006 		MESSAGE(out, job->node);
   1007 		lastNode = job->node;
   1008 	    }
   1009 	    (void) fprintf(out, "*** Signal %d\n", WTERMSIG(*status));
   1010 	}
   1011 
   1012 	(void) fflush(out);
   1013     }
   1014 
   1015     /*
   1016      * Now handle the -B-mode stuff. If the beast still isn't finished,
   1017      * try and restart the job on the next command. If JobStart says it's
   1018      * ok, it's ok. If there's an error, this puppy is done.
   1019      */
   1020     if (compatMake && (WIFEXITED(*status) &&
   1021 	!Lst_IsAtEnd(job->node->commands))) {
   1022 	switch (JobStart(job->node, job->flags & JOB_IGNDOTS, job)) {
   1023 	case JOB_RUNNING:
   1024 	    done = FALSE;
   1025 	    break;
   1026 	case JOB_ERROR:
   1027 	    done = TRUE;
   1028 	    *status = W_EXITCODE(1, 0);
   1029 	    break;
   1030 	case JOB_FINISHED:
   1031 	    /*
   1032 	     * If we got back a JOB_FINISHED code, JobStart has already
   1033 	     * called Make_Update and freed the job descriptor. We set
   1034 	     * done to false here to avoid fake cycles and double frees.
   1035 	     * JobStart needs to do the update so we can proceed up the
   1036 	     * graph when given the -n flag..
   1037 	     */
   1038 	    done = FALSE;
   1039 	    break;
   1040 	}
   1041     } else {
   1042 	done = TRUE;
   1043     }
   1044 
   1045     if (done) {
   1046 	Trace_Log(JOBEND, job);
   1047 	if (!compatMake && !(job->flags & JOB_SPECIAL)) {
   1048 	    if ((*status != 0) ||
   1049 	        (aborting == ABORT_ERROR) ||
   1050 	        (aborting == ABORT_INTERRUPT))
   1051 		Job_TokenReturn();
   1052 	}
   1053 
   1054     }
   1055 
   1056     if (done &&
   1057 	(aborting != ABORT_ERROR) &&
   1058 	(aborting != ABORT_INTERRUPT) &&
   1059 	(*status == 0))
   1060     {
   1061 	/*
   1062 	 * As long as we aren't aborting and the job didn't return a non-zero
   1063 	 * status that we shouldn't ignore, we call Make_Update to update
   1064 	 * the parents. In addition, any saved commands for the node are placed
   1065 	 * on the .END target.
   1066 	 */
   1067 	if (job->tailCmds != NILLNODE) {
   1068 	    Lst_ForEachFrom(job->node->commands, job->tailCmds,
   1069 			     JobSaveCommand,
   1070 			    (ClientData)job->node);
   1071 	}
   1072 	job->node->made = MADE;
   1073 	if (!(job->flags & JOB_SPECIAL))
   1074 	    Job_TokenReturn();
   1075 	Make_Update(job->node);
   1076 	free((Address)job);
   1077     } else if (*status != 0) {
   1078 	errors += 1;
   1079 	free((Address)job);
   1080     }
   1081     JobRestartJobs();
   1082 
   1083     /*
   1084      * Set aborting if any error.
   1085      */
   1086     if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) {
   1087 	/*
   1088 	 * If we found any errors in this batch of children and the -k flag
   1089 	 * wasn't given, we set the aborting flag so no more jobs get
   1090 	 * started.
   1091 	 */
   1092 	aborting = ABORT_ERROR;
   1093     }
   1094 
   1095     if ((aborting == ABORT_ERROR) && Job_Empty()) {
   1096 	/*
   1097 	 * If we are aborting and the job table is now empty, we finish.
   1098 	 */
   1099 	Finish(errors);
   1100     }
   1101 }
   1102 
   1103 /*-
   1104  *-----------------------------------------------------------------------
   1105  * Job_Touch --
   1106  *	Touch the given target. Called by JobStart when the -t flag was
   1107  *	given
   1108  *
   1109  * Results:
   1110  *	None
   1111  *
   1112  * Side Effects:
   1113  *	The data modification of the file is changed. In addition, if the
   1114  *	file did not exist, it is created.
   1115  *-----------------------------------------------------------------------
   1116  */
   1117 void
   1118 Job_Touch(gn, silent)
   1119     GNode         *gn;	      	/* the node of the file to touch */
   1120     Boolean 	  silent;   	/* TRUE if should not print messages */
   1121 {
   1122     int		  streamID;   	/* ID of stream opened to do the touch */
   1123     struct utimbuf times;	/* Times for utime() call */
   1124 
   1125     if (gn->type & (OP_JOIN|OP_USE|OP_EXEC|OP_OPTIONAL|OP_PHONY)) {
   1126 	/*
   1127 	 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets
   1128 	 * and, as such, shouldn't really be created.
   1129 	 */
   1130 	return;
   1131     }
   1132 
   1133     if (!silent || NoExecute(gn)) {
   1134 	(void) fprintf(stdout, "touch %s\n", gn->name);
   1135 	(void) fflush(stdout);
   1136     }
   1137 
   1138     if (NoExecute(gn)) {
   1139 	return;
   1140     }
   1141 
   1142     if (gn->type & OP_ARCHV) {
   1143 	Arch_Touch(gn);
   1144     } else if (gn->type & OP_LIB) {
   1145 	Arch_TouchLib(gn);
   1146     } else {
   1147 	char	*file = gn->path ? gn->path : gn->name;
   1148 
   1149 	times.actime = times.modtime = now;
   1150 	if (utime(file, &times) < 0){
   1151 	    streamID = open(file, O_RDWR | O_CREAT, 0666);
   1152 
   1153 	    if (streamID >= 0) {
   1154 		char	c;
   1155 
   1156 		/*
   1157 		 * Read and write a byte to the file to change the
   1158 		 * modification time, then close the file.
   1159 		 */
   1160 		if (read(streamID, &c, 1) == 1) {
   1161 		    (void) lseek(streamID, (off_t)0, SEEK_SET);
   1162 		    (void) write(streamID, &c, 1);
   1163 		}
   1164 
   1165 		(void) close(streamID);
   1166 	    } else {
   1167 		(void) fprintf(stdout, "*** couldn't touch %s: %s",
   1168 			       file, strerror(errno));
   1169 		(void) fflush(stdout);
   1170 	    }
   1171 	}
   1172     }
   1173 }
   1174 
   1175 /*-
   1176  *-----------------------------------------------------------------------
   1177  * Job_CheckCommands --
   1178  *	Make sure the given node has all the commands it needs.
   1179  *
   1180  * Results:
   1181  *	TRUE if the commands list is/was ok.
   1182  *
   1183  * Side Effects:
   1184  *	The node will have commands from the .DEFAULT rule added to it
   1185  *	if it needs them.
   1186  *-----------------------------------------------------------------------
   1187  */
   1188 Boolean
   1189 Job_CheckCommands(gn, abortProc)
   1190     GNode          *gn;	    	    /* The target whose commands need
   1191 				     * verifying */
   1192     void    	 (*abortProc) __P((char *, ...));
   1193 			/* Function to abort with message */
   1194 {
   1195     if (OP_NOP(gn->type) && Lst_IsEmpty(gn->commands) &&
   1196 	(gn->type & OP_LIB) == 0) {
   1197 	/*
   1198 	 * No commands. Look for .DEFAULT rule from which we might infer
   1199 	 * commands
   1200 	 */
   1201 	if ((DEFAULT != NILGNODE) && !Lst_IsEmpty(DEFAULT->commands)) {
   1202 	    char *p1;
   1203 	    /*
   1204 	     * Make only looks for a .DEFAULT if the node was never the
   1205 	     * target of an operator, so that's what we do too. If
   1206 	     * a .DEFAULT was given, we substitute its commands for gn's
   1207 	     * commands and set the IMPSRC variable to be the target's name
   1208 	     * The DEFAULT node acts like a transformation rule, in that
   1209 	     * gn also inherits any attributes or sources attached to
   1210 	     * .DEFAULT itself.
   1211 	     */
   1212 	    Make_HandleUse(DEFAULT, gn);
   1213 	    Var_Set(IMPSRC, Var_Value(TARGET, gn, &p1), gn, 0);
   1214 	    if (p1)
   1215 		free(p1);
   1216 	} else if (Dir_MTime(gn) == 0) {
   1217 	    /*
   1218 	     * The node wasn't the target of an operator we have no .DEFAULT
   1219 	     * rule to go on and the target doesn't already exist. There's
   1220 	     * nothing more we can do for this branch. If the -k flag wasn't
   1221 	     * given, we stop in our tracks, otherwise we just don't update
   1222 	     * this node's parents so they never get examined.
   1223 	     */
   1224 	    static const char msg[] = ": don't know how to make";
   1225 
   1226 	    if (gn->type & OP_OPTIONAL) {
   1227 		(void) fprintf(stdout, "%s%s %s(ignored)\n", progname,
   1228 		    msg, gn->name);
   1229 		(void) fflush(stdout);
   1230 	    } else if (keepgoing) {
   1231 		(void) fprintf(stdout, "%s%s %s(continuing)\n", progname,
   1232 		    msg, gn->name);
   1233 		(void) fflush(stdout);
   1234   		return FALSE;
   1235 	    } else {
   1236 		(*abortProc)("%s%s %s. Stop", progname, msg, gn->name);
   1237 		return FALSE;
   1238 	    }
   1239 	}
   1240     }
   1241     return TRUE;
   1242 }
   1243 #ifdef RMT_WILL_WATCH
   1244 /*-
   1245  *-----------------------------------------------------------------------
   1246  * JobLocalInput --
   1247  *	Handle a pipe becoming readable. Callback function for Rmt_Watch
   1248  *
   1249  * Results:
   1250  *	None
   1251  *
   1252  * Side Effects:
   1253  *	JobDoOutput is called.
   1254  *
   1255  *-----------------------------------------------------------------------
   1256  */
   1257 /*ARGSUSED*/
   1258 static void
   1259 JobLocalInput(stream, job)
   1260     int	    stream; 	/* Stream that's ready (ignored) */
   1261     Job	    *job;   	/* Job to which the stream belongs */
   1262 {
   1263     JobDoOutput(job, FALSE);
   1264 }
   1265 #endif /* RMT_WILL_WATCH */
   1266 
   1267 /*-
   1268  *-----------------------------------------------------------------------
   1269  * JobExec --
   1270  *	Execute the shell for the given job. Called from JobStart and
   1271  *	JobRestart.
   1272  *
   1273  * Results:
   1274  *	None.
   1275  *
   1276  * Side Effects:
   1277  *	A shell is executed, outputs is altered and the Job structure added
   1278  *	to the job table.
   1279  *
   1280  *-----------------------------------------------------------------------
   1281  */
   1282 static void
   1283 JobExec(job, argv)
   1284     Job	    	  *job; 	/* Job to execute */
   1285     char    	  **argv;
   1286 {
   1287     int	    	  cpid;	    	/* ID of new child */
   1288 
   1289     job->flags &= ~JOB_TRACED;
   1290 
   1291     if (DEBUG(JOB)) {
   1292 	int 	  i;
   1293 
   1294 	(void) fprintf(stdout, "Running %s %sly\n", job->node->name,
   1295 		       job->flags&JOB_REMOTE?"remote":"local");
   1296 	(void) fprintf(stdout, "\tCommand: ");
   1297 	for (i = 0; argv[i] != NULL; i++) {
   1298 	    (void) fprintf(stdout, "%s ", argv[i]);
   1299 	}
   1300  	(void) fprintf(stdout, "\n");
   1301  	(void) fflush(stdout);
   1302     }
   1303 
   1304     /*
   1305      * Some jobs produce no output and it's disconcerting to have
   1306      * no feedback of their running (since they produce no output, the
   1307      * banner with their name in it never appears). This is an attempt to
   1308      * provide that feedback, even if nothing follows it.
   1309      */
   1310     if ((lastNode != job->node) && (job->flags & JOB_FIRST) &&
   1311 	!(job->flags & JOB_SILENT)) {
   1312 	MESSAGE(stdout, job->node);
   1313 	lastNode = job->node;
   1314     }
   1315 
   1316 #ifdef RMT_NO_EXEC
   1317     if (job->flags & JOB_REMOTE) {
   1318 	goto jobExecFinish;
   1319     }
   1320 #endif /* RMT_NO_EXEC */
   1321 
   1322     if ((cpid = vfork()) == -1) {
   1323 	Punt("Cannot vfork");
   1324     } else if (cpid == 0) {
   1325 
   1326 	/*
   1327 	 * Must duplicate the input stream down to the child's input and
   1328 	 * reset it to the beginning (again). Since the stream was marked
   1329 	 * close-on-exec, we must clear that bit in the new input.
   1330 	 */
   1331 	if (dup2(FILENO(job->cmdFILE), 0) == -1)
   1332 	    Punt("Cannot dup2: %s", strerror(errno));
   1333 	(void) fcntl(0, F_SETFD, 0);
   1334 	(void) lseek(0, (off_t)0, SEEK_SET);
   1335 
   1336 	if (job->node->type & OP_MAKE) {
   1337 		/*
   1338 		 * Pass job token pipe to submakes.
   1339 		 */
   1340 		fcntl(job_pipe[0], F_SETFD, 0);
   1341 		fcntl(job_pipe[1], F_SETFD, 0);
   1342 	}
   1343 
   1344 	if (usePipes) {
   1345 	    /*
   1346 	     * Set up the child's output to be routed through the pipe
   1347 	     * we've created for it.
   1348 	     */
   1349 	    if (dup2(job->outPipe, 1) == -1)
   1350 		Punt("Cannot dup2: %s", strerror(errno));
   1351 	} else {
   1352 	    /*
   1353 	     * We're capturing output in a file, so we duplicate the
   1354 	     * descriptor to the temporary file into the standard
   1355 	     * output.
   1356 	     */
   1357 	    if (dup2(job->outFd, 1) == -1)
   1358 		Punt("Cannot dup2: %s", strerror(errno));
   1359 	}
   1360 	/*
   1361 	 * The output channels are marked close on exec. This bit was
   1362 	 * duplicated by the dup2 (on some systems), so we have to clear
   1363 	 * it before routing the shell's error output to the same place as
   1364 	 * its standard output.
   1365 	 */
   1366 	(void) fcntl(1, F_SETFD, 0);
   1367 	if (dup2(1, 2) == -1)
   1368 	    Punt("Cannot dup2: %s", strerror(errno));
   1369 
   1370 #ifdef USE_PGRP
   1371 	/*
   1372 	 * We want to switch the child into a different process family so
   1373 	 * we can kill it and all its descendants in one fell swoop,
   1374 	 * by killing its process family, but not commit suicide.
   1375 	 */
   1376 # if defined(SYSV)
   1377 	(void) setsid();
   1378 # else
   1379 	(void) setpgid(0, getpid());
   1380 # endif
   1381 #endif /* USE_PGRP */
   1382 
   1383 #ifdef REMOTE
   1384 	if (job->flags & JOB_REMOTE) {
   1385 	    Rmt_Exec(shellPath, argv, FALSE);
   1386 	} else
   1387 #endif /* REMOTE */
   1388 	{
   1389 	   (void) execv(shellPath, argv);
   1390 	   execError(shellPath);
   1391 	}
   1392 	_exit(1);
   1393     } else {
   1394 #ifdef REMOTE
   1395 	sigset_t nmask, omask;
   1396 	sigemptyset(&nmask);
   1397 	sigaddset(&nmask, SIGCHLD);
   1398 	sigprocmask(SIG_BLOCK, &nmask, &omask);
   1399 #endif
   1400 	job->pid = cpid;
   1401 
   1402 	Trace_Log(JOBSTART, job);
   1403 
   1404 	if (usePipes && (job->flags & JOB_FIRST)) {
   1405 	    /*
   1406 	     * The first time a job is run for a node, we set the current
   1407 	     * position in the buffer to the beginning and mark another
   1408 	     * stream to watch in the outputs mask
   1409 	     */
   1410 	    job->curPos = 0;
   1411 
   1412 #ifdef RMT_WILL_WATCH
   1413 	    Rmt_Watch(job->inPipe, JobLocalInput, job);
   1414 #else
   1415 #ifdef USE_SELECT
   1416 	    FD_SET(job->inPipe, &outputs);
   1417 #else
   1418 	    watchfd(job);
   1419 #endif
   1420 #endif /* RMT_WILL_WATCH */
   1421 	}
   1422 
   1423 	if (job->flags & JOB_REMOTE) {
   1424 #ifndef REMOTE
   1425 	    job->rmtID = 0;
   1426 #else
   1427 	    job->rmtID = Rmt_LastID(job->pid);
   1428 #endif /* REMOTE */
   1429 	} else {
   1430 	    nLocal += 1;
   1431 	    /*
   1432 	     * XXX: Used to not happen if REMOTE. Why?
   1433 	     */
   1434 	    if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
   1435 		(void) fclose(job->cmdFILE);
   1436 		job->cmdFILE = NULL;
   1437 	    }
   1438 	}
   1439 #ifdef REMOTE
   1440 	sigprocmask(SIG_SETMASK, &omask, NULL);
   1441 #endif
   1442     }
   1443 
   1444 #ifdef RMT_NO_EXEC
   1445 jobExecFinish:
   1446 #endif
   1447     /*
   1448      * Now the job is actually running, add it to the table.
   1449      */
   1450     nJobs += 1;
   1451     (void) Lst_AtEnd(jobs, (ClientData)job);
   1452 }
   1453 
   1454 /*-
   1455  *-----------------------------------------------------------------------
   1456  * JobMakeArgv --
   1457  *	Create the argv needed to execute the shell for a given job.
   1458  *
   1459  *
   1460  * Results:
   1461  *
   1462  * Side Effects:
   1463  *
   1464  *-----------------------------------------------------------------------
   1465  */
   1466 static void
   1467 JobMakeArgv(job, argv)
   1468     Job	    	  *job;
   1469     char	  **argv;
   1470 {
   1471     int	    	  argc;
   1472     static char	  args[10]; 	/* For merged arguments */
   1473 
   1474     argv[0] = shellName;
   1475     argc = 1;
   1476 
   1477     if ((commandShell->exit && (*commandShell->exit != '-')) ||
   1478 	(commandShell->echo && (*commandShell->echo != '-')))
   1479     {
   1480 	/*
   1481 	 * At least one of the flags doesn't have a minus before it, so
   1482 	 * merge them together. Have to do this because the *(&(@*#*&#$#
   1483 	 * Bourne shell thinks its second argument is a file to source.
   1484 	 * Grrrr. Note the ten-character limitation on the combined arguments.
   1485 	 */
   1486 	(void)snprintf(args, sizeof(args), "-%s%s",
   1487 		      ((job->flags & JOB_IGNERR) ? "" :
   1488 		       (commandShell->exit ? commandShell->exit : "")),
   1489 		      ((job->flags & JOB_SILENT) ? "" :
   1490 		       (commandShell->echo ? commandShell->echo : "")));
   1491 
   1492 	if (args[1]) {
   1493 	    argv[argc] = args;
   1494 	    argc++;
   1495 	}
   1496     } else {
   1497 	if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
   1498 	    argv[argc] = commandShell->exit;
   1499 	    argc++;
   1500 	}
   1501 	if (!(job->flags & JOB_SILENT) && commandShell->echo) {
   1502 	    argv[argc] = commandShell->echo;
   1503 	    argc++;
   1504 	}
   1505     }
   1506     argv[argc] = NULL;
   1507 }
   1508 
   1509 /*-
   1510  *-----------------------------------------------------------------------
   1511  * JobRestart --
   1512  *	Restart a job that stopped for some reason.
   1513  *
   1514  * Results:
   1515  *	None.
   1516  *
   1517  *-----------------------------------------------------------------------
   1518  */
   1519 static void
   1520 JobRestart(job)
   1521     Job 	  *job;    	/* Job to restart */
   1522 {
   1523 #ifdef REMOTE
   1524     int host;
   1525 #endif
   1526 
   1527     if (job->flags & JOB_REMIGRATE) {
   1528 	if (
   1529 #ifdef REMOTE
   1530 	    verboseRemigrates ||
   1531 #endif
   1532 	    DEBUG(JOB)) {
   1533 	   (void) fprintf(stdout, "*** remigrating %x(%s)\n",
   1534 			   job->pid, job->node->name);
   1535 	   (void) fflush(stdout);
   1536 	}
   1537 
   1538 #ifdef REMOTE
   1539 	if (!Rmt_ReExport(job->pid, job->node, &host)) {
   1540 	    if (verboseRemigrates || DEBUG(JOB)) {
   1541 		(void) fprintf(stdout, "*** couldn't migrate...\n");
   1542 		(void) fflush(stdout);
   1543 	    }
   1544 #endif
   1545 	    if (nLocal != maxLocal) {
   1546 		/*
   1547 		 * Job cannot be remigrated, but there's room on the local
   1548 		 * machine, so resume the job and note that another
   1549 		 * local job has started.
   1550 		 */
   1551 		if (
   1552 #ifdef REMOTE
   1553 		    verboseRemigrates ||
   1554 #endif
   1555 		    DEBUG(JOB)) {
   1556 		    (void) fprintf(stdout, "*** resuming on local machine\n");
   1557 		    (void) fflush(stdout);
   1558 		}
   1559 		KILL(job->pid, SIGCONT);
   1560 		nLocal +=1;
   1561 #ifdef REMOTE
   1562 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE);
   1563 		job->flags |= JOB_CONTINUING;
   1564 #else
   1565 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
   1566 #endif
   1567 	} else {
   1568 		/*
   1569 		 * Job cannot be restarted. Mark the table as full and
   1570 		 * place the job back on the list of stopped jobs.
   1571 		 */
   1572 		if (
   1573 #ifdef REMOTE
   1574 		    verboseRemigrates ||
   1575 #endif
   1576 		    DEBUG(JOB)) {
   1577 		   (void) fprintf(stdout, "*** holding\n");
   1578 		   (void) fflush(stdout);
   1579   		}
   1580 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
   1581 		return;
   1582 	    }
   1583 #ifdef REMOTE
   1584 	} else {
   1585 	    /*
   1586 	     * Clear out the remigrate and resume flags. Set the continuing
   1587 	     * flag so we know later on that the process isn't exiting just
   1588 	     * because of a signal.
   1589 	     */
   1590 	    job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
   1591 	    job->flags |= JOB_CONTINUING;
   1592 	    job->rmtID = host;
   1593 	}
   1594 #endif
   1595 
   1596 	(void)Lst_AtEnd(jobs, (ClientData)job);
   1597 	nJobs += 1;
   1598     } else if (job->flags & JOB_RESTART) {
   1599 	/*
   1600 	 * Set up the control arguments to the shell. This is based on the
   1601 	 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
   1602 	 * the 'exit' flag of the commandShell is used to cause it to exit
   1603 	 * upon receiving an error. If the JOB_SILENT flag is clear, the
   1604 	 * 'echo' flag of the commandShell is used to get it to start echoing
   1605 	 * as soon as it starts processing commands.
   1606 	 */
   1607 	char	  *argv[10];
   1608 
   1609 	JobMakeArgv(job, argv);
   1610 
   1611 	if (DEBUG(JOB)) {
   1612 	    (void) fprintf(stdout, "Restarting %s...", job->node->name);
   1613 	    (void) fflush(stdout);
   1614 	}
   1615 #ifdef REMOTE
   1616 	if ((job->node->type&OP_NOEXPORT) ||
   1617  	    (nLocal < maxLocal && runLocalFirst)
   1618 # ifdef RMT_NO_EXEC
   1619 	    || !Rmt_Export(shellPath, argv, job)
   1620 # else
   1621 	    || !Rmt_Begin(shellPath, argv, job->node)
   1622 # endif
   1623 #endif
   1624 	{
   1625 	    if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) {
   1626 		/*
   1627 		 * Can't be exported and not allowed to run locally -- put it
   1628 		 * back on the hold queue and mark the table full
   1629 		 */
   1630 		if (DEBUG(JOB)) {
   1631 		    (void) fprintf(stdout, "holding\n");
   1632 		    (void) fflush(stdout);
   1633 		}
   1634 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
   1635 		return;
   1636 	    } else {
   1637 		/*
   1638 		 * Job may be run locally.
   1639 		 */
   1640 		if (DEBUG(JOB)) {
   1641 		    (void) fprintf(stdout, "running locally\n");
   1642 		    (void) fflush(stdout);
   1643 		}
   1644 		job->flags &= ~JOB_REMOTE;
   1645 	    }
   1646 	}
   1647 #ifdef REMOTE
   1648 	else {
   1649 	    /*
   1650 	     * Can be exported. Hooray!
   1651 	     */
   1652 	    if (DEBUG(JOB)) {
   1653 		(void) fprintf(stdout, "exporting\n");
   1654 		(void) fflush(stdout);
   1655 	    }
   1656 	    job->flags |= JOB_REMOTE;
   1657 	}
   1658 #endif
   1659 	JobExec(job, argv);
   1660     } else {
   1661 	/*
   1662 	 * The job has stopped and needs to be restarted. Why it stopped,
   1663 	 * we don't know...
   1664 	 */
   1665 	if (DEBUG(JOB)) {
   1666 	   (void) fprintf(stdout, "Resuming %s...", job->node->name);
   1667 	   (void) fflush(stdout);
   1668 	}
   1669 	if (((job->flags & JOB_REMOTE) ||
   1670 	    (nLocal < maxLocal) ||
   1671 #ifdef REMOTE
   1672 	    (((job->flags & JOB_SPECIAL) &&
   1673 	      (job->node->type & OP_NOEXPORT)) &&
   1674 	     (maxLocal == 0))) &&
   1675 #else
   1676 	    ((job->flags & JOB_SPECIAL) &&
   1677 	     (maxLocal == 0))) &&
   1678 #endif
   1679 	   (nJobs != maxJobs))
   1680 	{
   1681 	    /*
   1682 	     * If the job is remote, it's ok to resume it as long as the
   1683 	     * maximum concurrency won't be exceeded. If it's local and
   1684 	     * we haven't reached the local concurrency limit already (or the
   1685 	     * job must be run locally and maxLocal is 0), it's also ok to
   1686 	     * resume it.
   1687 	     */
   1688 	    Boolean error;
   1689 	    int status;
   1690 
   1691 #ifdef RMT_WANTS_SIGNALS
   1692 	    if (job->flags & JOB_REMOTE) {
   1693 		error = !Rmt_Signal(job, SIGCONT);
   1694 	    } else
   1695 #endif	/* RMT_WANTS_SIGNALS */
   1696 		error = (KILL(job->pid, SIGCONT) != 0);
   1697 
   1698 	    if (!error) {
   1699 		/*
   1700 		 * Make sure the user knows we've continued the beast and
   1701 		 * actually put the thing in the job table.
   1702 		 */
   1703 		job->flags |= JOB_CONTINUING;
   1704 		status = W_STOPCODE(SIGCONT);
   1705 		JobFinish(job, &status);
   1706 
   1707 		job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
   1708 		if (DEBUG(JOB)) {
   1709 		   (void) fprintf(stdout, "done\n");
   1710 		   (void) fflush(stdout);
   1711 		}
   1712 	    } else {
   1713 		Error("couldn't resume %s: %s",
   1714 		    job->node->name, strerror(errno));
   1715 		status = W_EXITCODE(1, 0);
   1716 		JobFinish(job, &status);
   1717 	    }
   1718 	} else {
   1719 	    /*
   1720 	     * Job cannot be restarted. Mark the table as full and
   1721 	     * place the job back on the list of stopped jobs.
   1722 	     */
   1723 	    if (DEBUG(JOB)) {
   1724 		(void) fprintf(stdout, "table full\n");
   1725 		(void) fflush(stdout);
   1726 	    }
   1727 	    (void) Lst_AtFront(stoppedJobs, (ClientData)job);
   1728 	}
   1729     }
   1730 }
   1731 
   1732 /*-
   1733  *-----------------------------------------------------------------------
   1734  * JobStart  --
   1735  *	Start a target-creation process going for the target described
   1736  *	by the graph node gn.
   1737  *
   1738  * Results:
   1739  *	JOB_ERROR if there was an error in the commands, JOB_FINISHED
   1740  *	if there isn't actually anything left to do for the job and
   1741  *	JOB_RUNNING if the job has been started.
   1742  *
   1743  * Side Effects:
   1744  *	A new Job node is created and added to the list of running
   1745  *	jobs. PMake is forked and a child shell created.
   1746  *-----------------------------------------------------------------------
   1747  */
   1748 static int
   1749 JobStart(gn, flags, previous)
   1750     GNode         *gn;	      /* target to create */
   1751     int	  	   flags;      /* flags for the job to override normal ones.
   1752 			       * e.g. JOB_SPECIAL or JOB_IGNDOTS */
   1753     Job 	  *previous;  /* The previous Job structure for this node,
   1754 			       * if any. */
   1755 {
   1756     register Job  *job;       /* new job descriptor */
   1757     char	  *argv[10];  /* Argument vector to shell */
   1758     Boolean	  cmdsOK;     /* true if the nodes commands were all right */
   1759     Boolean 	  local;      /* Set true if the job was run locally */
   1760     Boolean 	  noExec;     /* Set true if we decide not to run the job */
   1761     int		  tfd;	      /* File descriptor to the temp file */
   1762 
   1763     if (previous != NULL) {
   1764 	previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
   1765 	job = previous;
   1766     } else {
   1767 	job = (Job *) emalloc(sizeof(Job));
   1768 	if (job == NULL) {
   1769 	    Punt("JobStart out of memory");
   1770 	}
   1771 	flags |= JOB_FIRST;
   1772     }
   1773 
   1774     job->node = gn;
   1775     job->tailCmds = NILLNODE;
   1776 
   1777     /*
   1778      * Set the initial value of the flags for this job based on the global
   1779      * ones and the node's attributes... Any flags supplied by the caller
   1780      * are also added to the field.
   1781      */
   1782     job->flags = 0;
   1783     if (Targ_Ignore(gn)) {
   1784 	job->flags |= JOB_IGNERR;
   1785     }
   1786     if (Targ_Silent(gn)) {
   1787 	job->flags |= JOB_SILENT;
   1788     }
   1789     job->flags |= flags;
   1790 
   1791     /*
   1792      * Check the commands now so any attributes from .DEFAULT have a chance
   1793      * to migrate to the node
   1794      */
   1795     if (!compatMake && job->flags & JOB_FIRST) {
   1796 	cmdsOK = Job_CheckCommands(gn, Error);
   1797     } else {
   1798 	cmdsOK = TRUE;
   1799     }
   1800 
   1801 #ifndef RMT_WILL_WATCH
   1802 #ifndef USE_SELECT
   1803     job->inPollfd = NULL;
   1804 #endif
   1805 #endif
   1806     /*
   1807      * If the -n flag wasn't given, we open up OUR (not the child's)
   1808      * temporary file to stuff commands in it. The thing is rd/wr so we don't
   1809      * need to reopen it to feed it to the shell. If the -n flag *was* given,
   1810      * we just set the file to be stdout. Cute, huh?
   1811      */
   1812     if (((gn->type & OP_MAKE) && !(noRecursiveExecute)) ||
   1813 	(!noExecute && !touchFlag)) {
   1814 	/*
   1815 	 * tfile is the name of a file into which all shell commands are
   1816 	 * put. It is used over by removing it before the child shell is
   1817 	 * executed. The XXXXXX in the string are replaced by the pid of
   1818 	 * the make process in a 6-character field with leading zeroes.
   1819 	 */
   1820 	char     tfile[sizeof(TMPPAT)];
   1821 	/*
   1822 	 * We're serious here, but if the commands were bogus, we're
   1823 	 * also dead...
   1824 	 */
   1825 	if (!cmdsOK) {
   1826 	    DieHorribly();
   1827 	}
   1828 
   1829 	(void)strcpy(tfile, TMPPAT);
   1830 	if ((tfd = mkstemp(tfile)) == -1)
   1831 	    Punt("Could not create temporary file %s", strerror(errno));
   1832 	(void) eunlink(tfile);
   1833 
   1834 	job->cmdFILE = fdopen(tfd, "w+");
   1835 	if (job->cmdFILE == NULL) {
   1836 	    Punt("Could not fdopen %s", tfile);
   1837 	}
   1838 	(void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1);
   1839 	/*
   1840 	 * Send the commands to the command file, flush all its buffers then
   1841 	 * rewind and remove the thing.
   1842 	 */
   1843 	noExec = FALSE;
   1844 
   1845 	/*
   1846 	 * used to be backwards; replace when start doing multiple commands
   1847 	 * per shell.
   1848 	 */
   1849 	if (compatMake) {
   1850 	    /*
   1851 	     * Be compatible: If this is the first time for this node,
   1852 	     * verify its commands are ok and open the commands list for
   1853 	     * sequential access by later invocations of JobStart.
   1854 	     * Once that is done, we take the next command off the list
   1855 	     * and print it to the command file. If the command was an
   1856 	     * ellipsis, note that there's nothing more to execute.
   1857 	     */
   1858 	    if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
   1859 		cmdsOK = FALSE;
   1860 	    } else {
   1861 		LstNode	ln = Lst_Next(gn->commands);
   1862 
   1863 		if ((ln == NILLNODE) ||
   1864 		    JobPrintCommand((ClientData) Lst_Datum(ln),
   1865 				    (ClientData) job))
   1866 		{
   1867 		    noExec = TRUE;
   1868 		    Lst_Close(gn->commands);
   1869 		}
   1870 		if (noExec && !(job->flags & JOB_FIRST)) {
   1871 		    /*
   1872 		     * If we're not going to execute anything, the job
   1873 		     * is done and we need to close down the various
   1874 		     * file descriptors we've opened for output, then
   1875 		     * call JobDoOutput to catch the final characters or
   1876 		     * send the file to the screen... Note that the i/o streams
   1877 		     * are only open if this isn't the first job.
   1878 		     * Note also that this could not be done in
   1879 		     * Job_CatchChildren b/c it wasn't clear if there were
   1880 		     * more commands to execute or not...
   1881 		     */
   1882 		    JobClose(job);
   1883 		}
   1884 	    }
   1885 	} else {
   1886 	    /*
   1887 	     * We can do all the commands at once. hooray for sanity
   1888 	     */
   1889 	    numCommands = 0;
   1890 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
   1891 
   1892 	    /*
   1893 	     * If we didn't print out any commands to the shell script,
   1894 	     * there's not much point in executing the shell, is there?
   1895 	     */
   1896 	    if (numCommands == 0) {
   1897 		noExec = TRUE;
   1898 	    }
   1899 	}
   1900     } else if (NoExecute(gn)) {
   1901 	/*
   1902 	 * Not executing anything -- just print all the commands to stdout
   1903 	 * in one fell swoop. This will still set up job->tailCmds correctly.
   1904 	 */
   1905 	if (lastNode != gn) {
   1906 	    MESSAGE(stdout, gn);
   1907 	    lastNode = gn;
   1908 	}
   1909 	job->cmdFILE = stdout;
   1910 	/*
   1911 	 * Only print the commands if they're ok, but don't die if they're
   1912 	 * not -- just let the user know they're bad and keep going. It
   1913 	 * doesn't do any harm in this case and may do some good.
   1914 	 */
   1915 	if (cmdsOK) {
   1916 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
   1917 	}
   1918 	/*
   1919 	 * Don't execute the shell, thank you.
   1920 	 */
   1921 	noExec = TRUE;
   1922     } else {
   1923 	/*
   1924 	 * Just touch the target and note that no shell should be executed.
   1925 	 * Set cmdFILE to stdout to make life easier. Check the commands, too,
   1926 	 * but don't die if they're no good -- it does no harm to keep working
   1927 	 * up the graph.
   1928 	 */
   1929 	job->cmdFILE = stdout;
   1930     	Job_Touch(gn, job->flags&JOB_SILENT);
   1931 	noExec = TRUE;
   1932     }
   1933 
   1934     /*
   1935      * If we're not supposed to execute a shell, don't.
   1936      */
   1937     if (noExec) {
   1938 	/*
   1939 	 * Unlink and close the command file if we opened one
   1940 	 */
   1941 	if (job->cmdFILE != stdout) {
   1942 	    if (job->cmdFILE != NULL) {
   1943 		(void) fclose(job->cmdFILE);
   1944 		job->cmdFILE = NULL;
   1945 	    }
   1946 	} else {
   1947 	     (void) fflush(stdout);
   1948 	}
   1949 
   1950 	/*
   1951 	 * We only want to work our way up the graph if we aren't here because
   1952 	 * the commands for the job were no good.
   1953 	 */
   1954 	if (cmdsOK) {
   1955 	    if (aborting == 0) {
   1956 		if (job->tailCmds != NILLNODE) {
   1957 		    Lst_ForEachFrom(job->node->commands, job->tailCmds,
   1958 				    JobSaveCommand,
   1959 				   (ClientData)job->node);
   1960 		}
   1961 		if (!(job->flags & JOB_SPECIAL))
   1962 		    Job_TokenReturn();
   1963 		Make_Update(job->node);
   1964 	    }
   1965 	    free((Address)job);
   1966 	    return(JOB_FINISHED);
   1967 	} else {
   1968 	    free((Address)job);
   1969 	    return(JOB_ERROR);
   1970 	}
   1971     } else {
   1972 	(void) fflush(job->cmdFILE);
   1973     }
   1974 
   1975     /*
   1976      * Set up the control arguments to the shell. This is based on the flags
   1977      * set earlier for this job.
   1978      */
   1979     JobMakeArgv(job, argv);
   1980 
   1981     /*
   1982      * If we're using pipes to catch output, create the pipe by which we'll
   1983      * get the shell's output. If we're using files, print out that we're
   1984      * starting a job and then set up its temporary-file name.
   1985      */
   1986     if (!compatMake || (job->flags & JOB_FIRST)) {
   1987 	if (usePipes) {
   1988 	    int fd[2];
   1989 	    if (pipe(fd) == -1)
   1990 		Punt("Cannot create pipe: %s", strerror(errno));
   1991 	    job->inPipe = fd[0];
   1992 #ifdef USE_SELECT
   1993 	    if (job->inPipe >= FD_SETSIZE)
   1994 		Punt("Ran out of fd_set slots; "
   1995 		    "recompile with a larger FD_SETSIZE.");
   1996 #endif
   1997 	    job->outPipe = fd[1];
   1998 	    (void) fcntl(job->inPipe, F_SETFD, 1);
   1999 	    (void) fcntl(job->outPipe, F_SETFD, 1);
   2000 	} else {
   2001 	    (void) fprintf(stdout, "Remaking `%s'\n", gn->name);
   2002   	    (void) fflush(stdout);
   2003 	    (void) strcpy(job->outFile, TMPPAT);
   2004 	    job->outFd = mkstemp(job->outFile);
   2005 	    (void) fcntl(job->outFd, F_SETFD, 1);
   2006 	}
   2007     }
   2008 
   2009 #ifdef REMOTE
   2010     if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) {
   2011 #ifdef RMT_NO_EXEC
   2012 	local = !Rmt_Export(shellPath, argv, job);
   2013 #else
   2014 	local = !Rmt_Begin(shellPath, argv, job->node);
   2015 #endif /* RMT_NO_EXEC */
   2016 	if (!local) {
   2017 	    job->flags |= JOB_REMOTE;
   2018 	}
   2019     } else
   2020 #endif
   2021 	local = TRUE;
   2022 
   2023     if (local && (((nLocal >= maxLocal) &&
   2024 	!(job->flags & JOB_SPECIAL) &&
   2025 #ifdef REMOTE
   2026 	(!(gn->type & OP_NOEXPORT) || (maxLocal != 0))
   2027 #else
   2028 	(maxLocal != 0)
   2029 #endif
   2030 	)))
   2031     {
   2032 	/*
   2033 	 * The job can only be run locally, but we've hit the limit of
   2034 	 * local concurrency, so put the job on hold until some other job
   2035 	 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
   2036 	 * may be run locally even when the local limit has been reached
   2037 	 * (e.g. when maxLocal == 0), though they will be exported if at
   2038 	 * all possible. In addition, any target marked with .NOEXPORT will
   2039 	 * be run locally if maxLocal is 0.
   2040 	 */
   2041 	job->flags |= JOB_RESTART;
   2042 	(void) Lst_AtEnd(stoppedJobs, (ClientData)job);
   2043     } else {
   2044 	JobExec(job, argv);
   2045     }
   2046     return(JOB_RUNNING);
   2047 }
   2048 
   2049 static char *
   2050 JobOutput(job, cp, endp, msg)
   2051     register Job *job;
   2052     register char *cp, *endp;
   2053     int msg;
   2054 {
   2055     register char *ecp;
   2056 
   2057     if (commandShell->noPrint) {
   2058 	ecp = Str_FindSubstring(cp, commandShell->noPrint);
   2059 	while (ecp != NULL) {
   2060 	    if (cp != ecp) {
   2061 		*ecp = '\0';
   2062 		if (msg && job->node != lastNode) {
   2063 		    MESSAGE(stdout, job->node);
   2064 		    lastNode = job->node;
   2065 		}
   2066 		/*
   2067 		 * The only way there wouldn't be a newline after
   2068 		 * this line is if it were the last in the buffer.
   2069 		 * however, since the non-printable comes after it,
   2070 		 * there must be a newline, so we don't print one.
   2071 		 */
   2072 		(void) fprintf(stdout, "%s", cp);
   2073 		(void) fflush(stdout);
   2074 	    }
   2075 	    cp = ecp + commandShell->noPLen;
   2076 	    if (cp != endp) {
   2077 		/*
   2078 		 * Still more to print, look again after skipping
   2079 		 * the whitespace following the non-printable
   2080 		 * command....
   2081 		 */
   2082 		cp++;
   2083 		while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
   2084 		    cp++;
   2085 		}
   2086 		ecp = Str_FindSubstring(cp, commandShell->noPrint);
   2087 	    } else {
   2088 		return cp;
   2089 	    }
   2090 	}
   2091     }
   2092     return cp;
   2093 }
   2094 
   2095 /*-
   2096  *-----------------------------------------------------------------------
   2097  * JobDoOutput  --
   2098  *	This function is called at different times depending on
   2099  *	whether the user has specified that output is to be collected
   2100  *	via pipes or temporary files. In the former case, we are called
   2101  *	whenever there is something to read on the pipe. We collect more
   2102  *	output from the given job and store it in the job's outBuf. If
   2103  *	this makes up a line, we print it tagged by the job's identifier,
   2104  *	as necessary.
   2105  *	If output has been collected in a temporary file, we open the
   2106  *	file and read it line by line, transfering it to our own
   2107  *	output channel until the file is empty. At which point we
   2108  *	remove the temporary file.
   2109  *	In both cases, however, we keep our figurative eye out for the
   2110  *	'noPrint' line for the shell from which the output came. If
   2111  *	we recognize a line, we don't print it. If the command is not
   2112  *	alone on the line (the character after it is not \0 or \n), we
   2113  *	do print whatever follows it.
   2114  *
   2115  * Results:
   2116  *	None
   2117  *
   2118  * Side Effects:
   2119  *	curPos may be shifted as may the contents of outBuf.
   2120  *-----------------------------------------------------------------------
   2121  */
   2122 STATIC void
   2123 JobDoOutput(job, finish)
   2124     register Job   *job;	  /* the job whose output needs printing */
   2125     Boolean	   finish;	  /* TRUE if this is the last time we'll be
   2126 				   * called for this job */
   2127 {
   2128     Boolean       gotNL = FALSE;  /* true if got a newline */
   2129     Boolean       fbuf;  	  /* true if our buffer filled up */
   2130     register int  nr;	      	  /* number of bytes read */
   2131     register int  i;	      	  /* auxiliary index into outBuf */
   2132     register int  max;	      	  /* limit for i (end of current data) */
   2133     int		  nRead;      	  /* (Temporary) number of bytes read */
   2134 
   2135     FILE      	  *oFILE;	  /* Stream pointer to shell's output file */
   2136     char          inLine[132];
   2137 
   2138 
   2139     if (usePipes) {
   2140 	/*
   2141 	 * Read as many bytes as will fit in the buffer.
   2142 	 */
   2143 end_loop:
   2144 	gotNL = FALSE;
   2145 	fbuf = FALSE;
   2146 
   2147 	nRead = read(job->inPipe, &job->outBuf[job->curPos],
   2148 			 JOB_BUFSIZE - job->curPos);
   2149 	if (nRead < 0) {
   2150 	    if (DEBUG(JOB)) {
   2151 		perror("JobDoOutput(piperead)");
   2152 	    }
   2153 	    nr = 0;
   2154 	} else {
   2155 	    nr = nRead;
   2156 	}
   2157 
   2158 	/*
   2159 	 * If we hit the end-of-file (the job is dead), we must flush its
   2160 	 * remaining output, so pretend we read a newline if there's any
   2161 	 * output remaining in the buffer.
   2162 	 * Also clear the 'finish' flag so we stop looping.
   2163 	 */
   2164 	if ((nr == 0) && (job->curPos != 0)) {
   2165 	    job->outBuf[job->curPos] = '\n';
   2166 	    nr = 1;
   2167 	    finish = FALSE;
   2168 	} else if (nr == 0) {
   2169 	    finish = FALSE;
   2170 	}
   2171 
   2172 	/*
   2173 	 * Look for the last newline in the bytes we just got. If there is
   2174 	 * one, break out of the loop with 'i' as its index and gotNL set
   2175 	 * TRUE.
   2176 	 */
   2177 	max = job->curPos + nr;
   2178 	for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
   2179 	    if (job->outBuf[i] == '\n') {
   2180 		gotNL = TRUE;
   2181 		break;
   2182 	    } else if (job->outBuf[i] == '\0') {
   2183 		/*
   2184 		 * Why?
   2185 		 */
   2186 		job->outBuf[i] = ' ';
   2187 	    }
   2188 	}
   2189 
   2190 	if (!gotNL) {
   2191 	    job->curPos += nr;
   2192 	    if (job->curPos == JOB_BUFSIZE) {
   2193 		/*
   2194 		 * If we've run out of buffer space, we have no choice
   2195 		 * but to print the stuff. sigh.
   2196 		 */
   2197 		fbuf = TRUE;
   2198 		i = job->curPos;
   2199 	    }
   2200 	}
   2201 	if (gotNL || fbuf) {
   2202 	    /*
   2203 	     * Need to send the output to the screen. Null terminate it
   2204 	     * first, overwriting the newline character if there was one.
   2205 	     * So long as the line isn't one we should filter (according
   2206 	     * to the shell description), we print the line, preceeded
   2207 	     * by a target banner if this target isn't the same as the
   2208 	     * one for which we last printed something.
   2209 	     * The rest of the data in the buffer are then shifted down
   2210 	     * to the start of the buffer and curPos is set accordingly.
   2211 	     */
   2212 	    job->outBuf[i] = '\0';
   2213 	    if (i >= job->curPos) {
   2214 		char *cp;
   2215 
   2216 		cp = JobOutput(job, job->outBuf, &job->outBuf[i], FALSE);
   2217 
   2218 		/*
   2219 		 * There's still more in that thar buffer. This time, though,
   2220 		 * we know there's no newline at the end, so we add one of
   2221 		 * our own free will.
   2222 		 */
   2223 		if (*cp != '\0') {
   2224 		    if (job->node != lastNode) {
   2225 			MESSAGE(stdout, job->node);
   2226 			lastNode = job->node;
   2227 		    }
   2228 		    (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : "");
   2229 		    (void) fflush(stdout);
   2230 		}
   2231 	    }
   2232 	    if (i < max - 1) {
   2233 		/* shift the remaining characters down */
   2234 		(void) memcpy(job->outBuf, &job->outBuf[i + 1], max - (i + 1));
   2235 		job->curPos = max - (i + 1);
   2236 
   2237 	    } else {
   2238 		/*
   2239 		 * We have written everything out, so we just start over
   2240 		 * from the start of the buffer. No copying. No nothing.
   2241 		 */
   2242 		job->curPos = 0;
   2243 	    }
   2244 	}
   2245 	if (finish) {
   2246 	    /*
   2247 	     * If the finish flag is true, we must loop until we hit
   2248 	     * end-of-file on the pipe. This is guaranteed to happen
   2249 	     * eventually since the other end of the pipe is now closed
   2250 	     * (we closed it explicitly and the child has exited). When
   2251 	     * we do get an EOF, finish will be set FALSE and we'll fall
   2252 	     * through and out.
   2253 	     */
   2254 	    goto end_loop;
   2255 	}
   2256     } else {
   2257 	/*
   2258 	 * We've been called to retrieve the output of the job from the
   2259 	 * temporary file where it's been squirreled away. This consists of
   2260 	 * opening the file, reading the output line by line, being sure not
   2261 	 * to print the noPrint line for the shell we used, then close and
   2262 	 * remove the temporary file. Very simple.
   2263 	 *
   2264 	 * Change to read in blocks and do FindSubString type things as for
   2265 	 * pipes? That would allow for "@echo -n..."
   2266 	 */
   2267 	oFILE = fopen(job->outFile, "r");
   2268 	if (oFILE != NULL) {
   2269 	    (void) fprintf(stdout, "Results of making %s:\n", job->node->name);
   2270 	    (void) fflush(stdout);
   2271 	    while (fgets(inLine, sizeof(inLine), oFILE) != NULL) {
   2272 		register char	*cp, *endp, *oendp;
   2273 
   2274 		cp = inLine;
   2275 		oendp = endp = inLine + strlen(inLine);
   2276 		if (endp[-1] == '\n') {
   2277 		    *--endp = '\0';
   2278 		}
   2279 		cp = JobOutput(job, inLine, endp, FALSE);
   2280 
   2281 		/*
   2282 		 * There's still more in that thar buffer. This time, though,
   2283 		 * we know there's no newline at the end, so we add one of
   2284 		 * our own free will.
   2285 		 */
   2286 		(void) fprintf(stdout, "%s", cp);
   2287 		(void) fflush(stdout);
   2288 		if (endp != oendp) {
   2289 		    (void) fprintf(stdout, "\n");
   2290 		    (void) fflush(stdout);
   2291 		}
   2292 	    }
   2293 	    (void) fclose(oFILE);
   2294 	    (void) eunlink(job->outFile);
   2295 	} else {
   2296 	    Punt("Cannot open `%s'", job->outFile);
   2297 	}
   2298     }
   2299 }
   2300 
   2301 /*-
   2302  *-----------------------------------------------------------------------
   2303  * Job_CatchChildren --
   2304  *	Handle the exit of a child. Called from Make_Make.
   2305  *
   2306  * Results:
   2307  *	none.
   2308  *
   2309  * Side Effects:
   2310  *	The job descriptor is removed from the list of children.
   2311  *
   2312  * Notes:
   2313  *	We do waits, blocking or not, according to the wisdom of our
   2314  *	caller, until there are no more children to report. For each
   2315  *	job, call JobFinish to finish things off. This will take care of
   2316  *	putting jobs on the stoppedJobs queue.
   2317  *
   2318  *-----------------------------------------------------------------------
   2319  */
   2320 void
   2321 Job_CatchChildren(block)
   2322     Boolean	  block;    	/* TRUE if should block on the wait. */
   2323 {
   2324     int    	  pid;	    	/* pid of dead child */
   2325     register Job  *job;	    	/* job descriptor for dead child */
   2326     LstNode       jnode;    	/* list element for finding job */
   2327     int	  	  status;   	/* Exit/termination status */
   2328 
   2329     /*
   2330      * Don't even bother if we know there's no one around.
   2331      */
   2332     if (nLocal == 0) {
   2333 	return;
   2334     }
   2335 
   2336     while ((pid = waitpid((pid_t) -1, &status,
   2337 			  (block?0:WNOHANG)|WUNTRACED)) > 0)
   2338     {
   2339 	if (DEBUG(JOB)) {
   2340 	    (void) fprintf(stdout, "Process %d exited or stopped %x.\n", pid,
   2341 	      status);
   2342 	    (void) fflush(stdout);
   2343 	}
   2344 
   2345 
   2346 	jnode = Lst_Find(jobs, (ClientData)&pid, JobCmpPid);
   2347 
   2348 	if (jnode == NILLNODE) {
   2349 	    if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGCONT)) {
   2350 		jnode = Lst_Find(stoppedJobs, (ClientData) &pid, JobCmpPid);
   2351 		if (jnode == NILLNODE) {
   2352 		    Error("Resumed child (%d) not in table", pid);
   2353 		    continue;
   2354 		}
   2355 		job = (Job *)Lst_Datum(jnode);
   2356 		(void) Lst_Remove(stoppedJobs, jnode);
   2357 	    } else {
   2358 		Error("Child (%d) not in table?", pid);
   2359 		continue;
   2360 	    }
   2361 	} else {
   2362 	    job = (Job *) Lst_Datum(jnode);
   2363 	    (void) Lst_Remove(jobs, jnode);
   2364 	    nJobs -= 1;
   2365 #ifdef REMOTE
   2366 	    if (!(job->flags & JOB_REMOTE)) {
   2367 		if (DEBUG(JOB)) {
   2368 		    (void) fprintf(stdout,
   2369 				   "Job queue has one fewer local process.\n");
   2370 		    (void) fflush(stdout);
   2371 		}
   2372 		nLocal -= 1;
   2373 	    }
   2374 #else
   2375 	    nLocal -= 1;
   2376 #endif
   2377 	}
   2378 
   2379 	JobFinish(job, &status);
   2380     }
   2381 }
   2382 
   2383 /*-
   2384  *-----------------------------------------------------------------------
   2385  * Job_CatchOutput --
   2386  *	Catch the output from our children, if we're using
   2387  *	pipes do so. Otherwise just block time until we get a
   2388  *	signal (most likely a SIGCHLD) since there's no point in
   2389  *	just spinning when there's nothing to do and the reaping
   2390  *	of a child can wait for a while.
   2391  *
   2392  * Results:
   2393  *	None
   2394  *
   2395  * Side Effects:
   2396  *	Output is read from pipes if we're piping.
   2397  * -----------------------------------------------------------------------
   2398  */
   2399 void
   2400 Job_CatchOutput()
   2401 {
   2402     int           	  nready;
   2403     register LstNode	  ln;
   2404     register Job   	  *job;
   2405 #ifdef RMT_WILL_WATCH
   2406     int	    	  	  pnJobs;   	/* Previous nJobs */
   2407 #endif
   2408 
   2409     (void) fflush(stdout);
   2410     Job_TokenFlush();
   2411 #ifdef RMT_WILL_WATCH
   2412     pnJobs = nJobs;
   2413 
   2414     /*
   2415      * It is possible for us to be called with nJobs equal to 0. This happens
   2416      * if all the jobs finish and a job that is stopped cannot be run
   2417      * locally (eg if maxLocal is 0) and cannot be exported. The job will
   2418      * be placed back on the stoppedJobs queue, Job_Empty() will return false,
   2419      * Make_Run will call us again when there's nothing for which to wait.
   2420      * nJobs never changes, so we loop forever. Hence the check. It could
   2421      * be argued that we should sleep for a bit so as not to swamp the
   2422      * exportation system with requests. Perhaps we should.
   2423      *
   2424      * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
   2425      * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
   2426      * It may use the variable nLocal to determine if it needs to call
   2427      * Job_CatchChildren (if nLocal is 0, there's nothing for which to
   2428      * wait...)
   2429      */
   2430     while (nJobs != 0 && pnJobs == nJobs) {
   2431 	Rmt_Wait();
   2432     }
   2433 #else
   2434     if (usePipes) {
   2435 #ifdef USE_SELECT
   2436 	struct timeval	  timeout;
   2437 	fd_set         	  readfds;
   2438 
   2439 	readfds = outputs;
   2440 	timeout.tv_sec = SEL_SEC;
   2441 	timeout.tv_usec = SEL_USEC;
   2442 
   2443 	if ((nready = select(FD_SETSIZE, &readfds, (fd_set *) 0,
   2444 			   (fd_set *) 0, &timeout)) <= 0)
   2445 	    return;
   2446 #else
   2447 	if ((nready = poll((wantToken ? fds : (fds + 1)),
   2448 	  		   (wantToken ? nfds : (nfds - 1)), POLL_MSEC)) <= 0)
   2449 	    return;
   2450 #endif
   2451 	else {
   2452 	    if (Lst_Open(jobs) == FAILURE) {
   2453 		Punt("Cannot open job table");
   2454 	    }
   2455 	    while (nready && (ln = Lst_Next(jobs)) != NILLNODE) {
   2456 		job = (Job *) Lst_Datum(ln);
   2457 #ifdef USE_SELECT
   2458 		if (FD_ISSET(job->inPipe, &readfds))
   2459 #else
   2460 		if (readyfd(job))
   2461 #endif
   2462 		{
   2463 		    JobDoOutput(job, FALSE);
   2464 		    nready -= 1;
   2465 		}
   2466 
   2467 	    }
   2468 	    Lst_Close(jobs);
   2469 	}
   2470     }
   2471 #endif /* RMT_WILL_WATCH */
   2472 }
   2473 
   2474 /*-
   2475  *-----------------------------------------------------------------------
   2476  * Job_Make --
   2477  *	Start the creation of a target. Basically a front-end for
   2478  *	JobStart used by the Make module.
   2479  *
   2480  * Results:
   2481  *	None.
   2482  *
   2483  * Side Effects:
   2484  *	Another job is started.
   2485  *
   2486  *-----------------------------------------------------------------------
   2487  */
   2488 void
   2489 Job_Make(gn)
   2490     GNode   *gn;
   2491 {
   2492     (void) JobStart(gn, 0, NULL);
   2493 }
   2494 
   2495 /*-
   2496  *-----------------------------------------------------------------------
   2497  * Job_Init --
   2498  *	Initialize the process module
   2499  *
   2500  * Results:
   2501  *	none
   2502  *
   2503  * Side Effects:
   2504  *	lists and counters are initialized
   2505  *-----------------------------------------------------------------------
   2506  */
   2507 void
   2508 Job_Init(maxproc, maxlocal)
   2509     int           maxproc;  /* the greatest number of jobs which may be
   2510 			     * running at one time */
   2511     int	    	  maxlocal; /* the greatest number of local jobs which may
   2512 			     * be running at once. */
   2513 {
   2514     GNode         *begin;     /* node for commands to do at the very start */
   2515 
   2516     jobs =  	  Lst_Init(FALSE);
   2517     stoppedJobs = Lst_Init(FALSE);
   2518     maxJobs = 	  maxproc;
   2519     maxLocal = 	  maxlocal;
   2520     nJobs = 	  0;
   2521     nLocal = 	  0;
   2522     wantToken =	  FALSE;
   2523 
   2524     aborting = 	  0;
   2525     errors = 	  0;
   2526 
   2527     lastNode =	  NILGNODE;
   2528 
   2529     if (maxJobs == 1
   2530 #ifdef REMOTE
   2531 	|| noMessages
   2532 #endif
   2533 		     ) {
   2534 	/*
   2535 	 * If only one job can run at a time, there's no need for a banner,
   2536 	 * is there?
   2537 	 */
   2538 	targFmt = "";
   2539     } else {
   2540 	targFmt = TARG_FMT;
   2541     }
   2542 
   2543     if (shellPath == NULL) {
   2544 	/*
   2545 	 * The user didn't specify a shell to use, so we are using the
   2546 	 * default one... Both the absolute path and the last component
   2547 	 * must be set. The last component is taken from the 'name' field
   2548 	 * of the default shell description pointed-to by commandShell.
   2549 	 * All default shells are located in _PATH_DEFSHELLDIR.
   2550 	 */
   2551 	shellName = commandShell->name;
   2552 	shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
   2553     }
   2554 
   2555     if (commandShell->exit == NULL) {
   2556 	commandShell->exit = "";
   2557     }
   2558     if (commandShell->echo == NULL) {
   2559 	commandShell->echo = "";
   2560     }
   2561 
   2562     /*
   2563      * Catch the four signals that POSIX specifies if they aren't ignored.
   2564      * JobPassSig will take care of calling JobInterrupt if appropriate.
   2565      */
   2566     if (signal(SIGINT, SIG_IGN) != SIG_IGN) {
   2567 	(void) signal(SIGINT, JobPassSig);
   2568     }
   2569     if (signal(SIGHUP, SIG_IGN) != SIG_IGN) {
   2570 	(void) signal(SIGHUP, JobPassSig);
   2571     }
   2572     if (signal(SIGQUIT, SIG_IGN) != SIG_IGN) {
   2573 	(void) signal(SIGQUIT, JobPassSig);
   2574     }
   2575     if (signal(SIGTERM, SIG_IGN) != SIG_IGN) {
   2576 	(void) signal(SIGTERM, JobPassSig);
   2577     }
   2578     /*
   2579      * Install a NOOP  SIGCHLD handler so we are woken up if we're blocked.
   2580      */
   2581     signal(SIGCHLD, JobIgnoreSig);
   2582 
   2583     /*
   2584      * There are additional signals that need to be caught and passed if
   2585      * either the export system wants to be told directly of signals or if
   2586      * we're giving each job its own process group (since then it won't get
   2587      * signals from the terminal driver as we own the terminal)
   2588      */
   2589 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
   2590     if (signal(SIGTSTP, SIG_IGN) != SIG_IGN) {
   2591 	(void) signal(SIGTSTP, JobPassSig);
   2592     }
   2593     if (signal(SIGTTOU, SIG_IGN) != SIG_IGN) {
   2594 	(void) signal(SIGTTOU, JobPassSig);
   2595     }
   2596     if (signal(SIGTTIN, SIG_IGN) != SIG_IGN) {
   2597 	(void) signal(SIGTTIN, JobPassSig);
   2598     }
   2599     if (signal(SIGWINCH, SIG_IGN) != SIG_IGN) {
   2600 	(void) signal(SIGWINCH, JobPassSig);
   2601     }
   2602     if (signal(SIGCONT, SIG_IGN) != SIG_IGN) {
   2603 	(void) signal(SIGCONT, JobContinueSig);
   2604     }
   2605 #endif
   2606 
   2607     begin = Targ_FindNode(".BEGIN", TARG_NOCREATE);
   2608 
   2609     if (begin != NILGNODE) {
   2610 	JobStart(begin, JOB_SPECIAL, (Job *)0);
   2611 	while (nJobs) {
   2612 	    Job_CatchOutput();
   2613 #ifndef RMT_WILL_WATCH
   2614 	    Job_CatchChildren(!usePipes);
   2615 #endif /* RMT_WILL_WATCH */
   2616 	}
   2617     }
   2618     postCommands = Targ_FindNode(".END", TARG_CREATE);
   2619 }
   2620 
   2621 /*-
   2622  *-----------------------------------------------------------------------
   2623  * Job_Empty --
   2624  *	See if the job table is empty.  Because the local concurrency may
   2625  *	be set to 0, it is possible for the job table to become empty,
   2626  *	while the list of stoppedJobs remains non-empty. In such a case,
   2627  *	we want to restart as many jobs as we can.
   2628  *
   2629  * Results:
   2630  *	TRUE if it is. FALSE if it ain't.
   2631  *
   2632  * Side Effects:
   2633  *	None.
   2634  *
   2635  * -----------------------------------------------------------------------
   2636  */
   2637 Boolean
   2638 Job_Empty()
   2639 {
   2640     if (nJobs == 0) {
   2641 	if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
   2642 	    /*
   2643 	     * The job table is obviously not full if it has no jobs in
   2644 	     * it...Try and restart the stopped jobs.
   2645 	     */
   2646 	    JobRestartJobs();
   2647 	    return(FALSE);
   2648 	} else {
   2649 	    return(TRUE);
   2650 	}
   2651     } else {
   2652 	return(FALSE);
   2653     }
   2654 }
   2655 
   2656 /*-
   2657  *-----------------------------------------------------------------------
   2658  * JobMatchShell --
   2659  *	Find a matching shell in 'shells' given its final component.
   2660  *
   2661  * Results:
   2662  *	A pointer to the Shell structure.
   2663  *
   2664  * Side Effects:
   2665  *	None.
   2666  *
   2667  *-----------------------------------------------------------------------
   2668  */
   2669 static Shell *
   2670 JobMatchShell(name)
   2671     char	  *name;      /* Final component of shell path */
   2672 {
   2673     register Shell *sh;	      /* Pointer into shells table */
   2674     Shell	   *match;    /* Longest-matching shell */
   2675     register char *cp1,
   2676 		  *cp2;
   2677     char	  *eoname;
   2678 
   2679     eoname = name + strlen(name);
   2680 
   2681     match = NULL;
   2682 
   2683     for (sh = shells; sh->name != NULL; sh++) {
   2684 	for (cp1 = eoname - strlen(sh->name), cp2 = sh->name;
   2685 	     *cp1 != '\0' && *cp1 == *cp2;
   2686 	     cp1++, cp2++) {
   2687 		 continue;
   2688 	}
   2689 	if (*cp1 != *cp2) {
   2690 	    continue;
   2691 	} else if (match == NULL || strlen(match->name) < strlen(sh->name)) {
   2692 	   match = sh;
   2693 	}
   2694     }
   2695     return(match == NULL ? sh : match);
   2696 }
   2697 
   2698 /*-
   2699  *-----------------------------------------------------------------------
   2700  * Job_ParseShell --
   2701  *	Parse a shell specification and set up commandShell, shellPath
   2702  *	and shellName appropriately.
   2703  *
   2704  * Results:
   2705  *	FAILURE if the specification was incorrect.
   2706  *
   2707  * Side Effects:
   2708  *	commandShell points to a Shell structure (either predefined or
   2709  *	created from the shell spec), shellPath is the full path of the
   2710  *	shell described by commandShell, while shellName is just the
   2711  *	final component of shellPath.
   2712  *
   2713  * Notes:
   2714  *	A shell specification consists of a .SHELL target, with dependency
   2715  *	operator, followed by a series of blank-separated words. Double
   2716  *	quotes can be used to use blanks in words. A backslash escapes
   2717  *	anything (most notably a double-quote and a space) and
   2718  *	provides the functionality it does in C. Each word consists of
   2719  *	keyword and value separated by an equal sign. There should be no
   2720  *	unnecessary spaces in the word. The keywords are as follows:
   2721  *	    name  	    Name of shell.
   2722  *	    path  	    Location of shell. Overrides "name" if given
   2723  *	    quiet 	    Command to turn off echoing.
   2724  *	    echo  	    Command to turn echoing on
   2725  *	    filter	    Result of turning off echoing that shouldn't be
   2726  *	    	  	    printed.
   2727  *	    echoFlag	    Flag to turn echoing on at the start
   2728  *	    errFlag	    Flag to turn error checking on at the start
   2729  *	    hasErrCtl	    True if shell has error checking control
   2730  *	    check 	    Command to turn on error checking if hasErrCtl
   2731  *	    	  	    is TRUE or template of command to echo a command
   2732  *	    	  	    for which error checking is off if hasErrCtl is
   2733  *	    	  	    FALSE.
   2734  *	    ignore	    Command to turn off error checking if hasErrCtl
   2735  *	    	  	    is TRUE or template of command to execute a
   2736  *	    	  	    command so as to ignore any errors it returns if
   2737  *	    	  	    hasErrCtl is FALSE.
   2738  *
   2739  *-----------------------------------------------------------------------
   2740  */
   2741 ReturnStatus
   2742 Job_ParseShell(line)
   2743     char	  *line;  /* The shell spec */
   2744 {
   2745     char    	  **words;
   2746     int	    	  wordCount;
   2747     register char **argv;
   2748     register int  argc;
   2749     char    	  *path;
   2750     Shell   	  newShell;
   2751     Boolean 	  fullSpec = FALSE;
   2752 
   2753     while (isspace((unsigned char)*line)) {
   2754 	line++;
   2755     }
   2756 
   2757     if (shellArgv)
   2758 	free(shellArgv);
   2759 
   2760     words = brk_string(line, &wordCount, TRUE, &shellArgv);
   2761 
   2762     memset((Address)&newShell, 0, sizeof(newShell));
   2763 
   2764     /*
   2765      * Parse the specification by keyword
   2766      */
   2767     for (path = NULL, argc = wordCount - 1, argv = words;
   2768 	argc != 0;
   2769 	argc--, argv++) {
   2770 	    if (strncmp(*argv, "path=", 5) == 0) {
   2771 		path = &argv[0][5];
   2772 	    } else if (strncmp(*argv, "name=", 5) == 0) {
   2773 		newShell.name = &argv[0][5];
   2774 	    } else {
   2775 		if (strncmp(*argv, "quiet=", 6) == 0) {
   2776 		    newShell.echoOff = &argv[0][6];
   2777 		} else if (strncmp(*argv, "echo=", 5) == 0) {
   2778 		    newShell.echoOn = &argv[0][5];
   2779 		} else if (strncmp(*argv, "filter=", 7) == 0) {
   2780 		    newShell.noPrint = &argv[0][7];
   2781 		    newShell.noPLen = strlen(newShell.noPrint);
   2782 		} else if (strncmp(*argv, "echoFlag=", 9) == 0) {
   2783 		    newShell.echo = &argv[0][9];
   2784 		} else if (strncmp(*argv, "errFlag=", 8) == 0) {
   2785 		    newShell.exit = &argv[0][8];
   2786 		} else if (strncmp(*argv, "hasErrCtl=", 10) == 0) {
   2787 		    char c = argv[0][10];
   2788 		    newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
   2789 					   (c != 'T') && (c != 't'));
   2790 		} else if (strncmp(*argv, "check=", 6) == 0) {
   2791 		    newShell.errCheck = &argv[0][6];
   2792 		} else if (strncmp(*argv, "ignore=", 7) == 0) {
   2793 		    newShell.ignErr = &argv[0][7];
   2794 		} else {
   2795 		    Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"",
   2796 				*argv);
   2797 		    free(words);
   2798 		    return(FAILURE);
   2799 		}
   2800 		fullSpec = TRUE;
   2801 	    }
   2802     }
   2803 
   2804     if (path == NULL) {
   2805 	/*
   2806 	 * If no path was given, the user wants one of the pre-defined shells,
   2807 	 * yes? So we find the one s/he wants with the help of JobMatchShell
   2808 	 * and set things up the right way. shellPath will be set up by
   2809 	 * Job_Init.
   2810 	 */
   2811 	if (newShell.name == NULL) {
   2812 	    Parse_Error(PARSE_FATAL, "Neither path nor name specified");
   2813 	    return(FAILURE);
   2814 	} else {
   2815 	    commandShell = JobMatchShell(newShell.name);
   2816 	    shellName = newShell.name;
   2817 	}
   2818     } else {
   2819 	/*
   2820 	 * The user provided a path. If s/he gave nothing else (fullSpec is
   2821 	 * FALSE), try and find a matching shell in the ones we know of.
   2822 	 * Else we just take the specification at its word and copy it
   2823 	 * to a new location. In either case, we need to record the
   2824 	 * path the user gave for the shell.
   2825 	 */
   2826 	shellPath = path;
   2827 	path = strrchr(path, '/');
   2828 	if (path == NULL) {
   2829 	    path = shellPath;
   2830 	} else {
   2831 	    path += 1;
   2832 	}
   2833 	if (newShell.name != NULL) {
   2834 	    shellName = newShell.name;
   2835 	} else {
   2836 	    shellName = path;
   2837 	}
   2838 	if (!fullSpec) {
   2839 	    commandShell = JobMatchShell(shellName);
   2840 	} else {
   2841 	    commandShell = (Shell *) emalloc(sizeof(Shell));
   2842 	    *commandShell = newShell;
   2843 	}
   2844     }
   2845 
   2846     if (commandShell->echoOn && commandShell->echoOff) {
   2847 	commandShell->hasEchoCtl = TRUE;
   2848     }
   2849 
   2850     if (!commandShell->hasErrCtl) {
   2851 	if (commandShell->errCheck == NULL) {
   2852 	    commandShell->errCheck = "";
   2853 	}
   2854 	if (commandShell->ignErr == NULL) {
   2855 	    commandShell->ignErr = "%s\n";
   2856 	}
   2857     }
   2858 
   2859     /*
   2860      * Do not free up the words themselves, since they might be in use by the
   2861      * shell specification.
   2862      */
   2863     free(words);
   2864     return SUCCESS;
   2865 }
   2866 
   2867 /*-
   2868  *-----------------------------------------------------------------------
   2869  * JobInterrupt --
   2870  *	Handle the receipt of an interrupt.
   2871  *
   2872  * Results:
   2873  *	None
   2874  *
   2875  * Side Effects:
   2876  *	All children are killed. Another job will be started if the
   2877  *	.INTERRUPT target was given.
   2878  *-----------------------------------------------------------------------
   2879  */
   2880 static void
   2881 JobInterrupt(runINTERRUPT, signo)
   2882     int	    runINTERRUPT;   	/* Non-zero if commands for the .INTERRUPT
   2883 				 * target should be executed */
   2884     int	    signo;		/* signal received */
   2885 {
   2886     LstNode 	  ln;		/* element in job table */
   2887     Job           *job;	    	/* job descriptor in that element */
   2888     GNode         *interrupt;	/* the node describing the .INTERRUPT target */
   2889 
   2890     aborting = ABORT_INTERRUPT;
   2891 
   2892    (void) Lst_Open(jobs);
   2893     while ((ln = Lst_Next(jobs)) != NILLNODE) {
   2894 	job = (Job *) Lst_Datum(ln);
   2895 
   2896 	if (!Targ_Precious(job->node)) {
   2897 	    char  	*file = (job->node->path == NULL ?
   2898 				 job->node->name :
   2899 				 job->node->path);
   2900 	    if (!noExecute && eunlink(file) != -1) {
   2901 		Error("*** %s removed", file);
   2902 	    }
   2903 	}
   2904 #ifdef RMT_WANTS_SIGNALS
   2905 	if (job->flags & JOB_REMOTE) {
   2906 	    /*
   2907 	     * If job is remote, let the Rmt module do the killing.
   2908 	     */
   2909 	    if (!Rmt_Signal(job, signo)) {
   2910 		/*
   2911 		 * If couldn't kill the thing, finish it out now with an
   2912 		 * error code, since no exit report will come in likely.
   2913 		 */
   2914 		int status;
   2915 
   2916 		status.w_status = 0;
   2917 		status.w_retcode = 1;
   2918 		JobFinish(job, &status);
   2919 	    }
   2920 	} else if (job->pid) {
   2921 	    KILL(job->pid, signo);
   2922 	}
   2923 #else
   2924 	if (job->pid) {
   2925 	    if (DEBUG(JOB)) {
   2926 		(void) fprintf(stdout,
   2927 			       "JobInterrupt passing signal to child %d.\n",
   2928 			       job->pid);
   2929 		(void) fflush(stdout);
   2930 	    }
   2931 	    KILL(job->pid, signo);
   2932 	}
   2933 #endif /* RMT_WANTS_SIGNALS */
   2934     }
   2935 
   2936 #ifdef REMOTE
   2937    (void)Lst_Open(stoppedJobs);
   2938     while ((ln = Lst_Next(stoppedJobs)) != NILLNODE) {
   2939 	job = (Job *) Lst_Datum(ln);
   2940 
   2941 	if (job->flags & JOB_RESTART) {
   2942 	    if (DEBUG(JOB)) {
   2943 		(void) fprintf(stdout, "%s%s",
   2944 			       "JobInterrupt skipping job on stopped queue",
   2945 			       "-- it was waiting to be restarted.\n");
   2946 		(void) fflush(stdout);
   2947 	    }
   2948 	    continue;
   2949 	}
   2950 	if (!Targ_Precious(job->node)) {
   2951 	    char  	*file = (job->node->path == NULL ?
   2952 				 job->node->name :
   2953 				 job->node->path);
   2954 	    if (eunlink(file) == 0) {
   2955 		Error("*** %s removed", file);
   2956 	    }
   2957 	}
   2958 	/*
   2959 	 * Resume the thing so it will take the signal.
   2960 	 */
   2961 	if (DEBUG(JOB)) {
   2962 	    (void) fprintf(stdout,
   2963 			   "JobInterrupt passing CONT to stopped child %d.\n",
   2964 			   job->pid);
   2965 	    (void) fflush(stdout);
   2966 	}
   2967 	KILL(job->pid, SIGCONT);
   2968 #ifdef RMT_WANTS_SIGNALS
   2969 	if (job->flags & JOB_REMOTE) {
   2970 	    /*
   2971 	     * If job is remote, let the Rmt module do the killing.
   2972 	     */
   2973 	    if (!Rmt_Signal(job, SIGINT)) {
   2974 		/*
   2975 		 * If couldn't kill the thing, finish it out now with an
   2976 		 * error code, since no exit report will come in likely.
   2977 		 */
   2978 		int status;
   2979 		status.w_status = 0;
   2980 		status.w_retcode = 1;
   2981 		JobFinish(job, &status);
   2982 	    }
   2983 	} else if (job->pid) {
   2984 	    if (DEBUG(JOB)) {
   2985 		(void) fprintf(stdout,
   2986 		       "JobInterrupt passing interrupt to stopped child %d.\n",
   2987 			       job->pid);
   2988 		(void) fflush(stdout);
   2989 	    }
   2990 	    KILL(job->pid, SIGINT);
   2991 	}
   2992 #endif /* RMT_WANTS_SIGNALS */
   2993     }
   2994 #endif
   2995     Lst_Close(stoppedJobs);
   2996 
   2997     if (runINTERRUPT && !touchFlag) {
   2998 	interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE);
   2999 	if (interrupt != NILGNODE) {
   3000 	    ignoreErrors = FALSE;
   3001 
   3002 	    JobStart(interrupt, JOB_IGNDOTS, (Job *)0);
   3003 	    while (nJobs) {
   3004 		Job_CatchOutput();
   3005 #ifndef RMT_WILL_WATCH
   3006 		Job_CatchChildren(!usePipes);
   3007 #endif /* RMT_WILL_WATCH */
   3008 	    }
   3009 	}
   3010     }
   3011     Trace_Log(MAKEINTR, 0);
   3012     exit(signo);
   3013 }
   3014 
   3015 /*
   3016  *-----------------------------------------------------------------------
   3017  * Job_Finish --
   3018  *	Do final processing such as the running of the commands
   3019  *	attached to the .END target.
   3020  *
   3021  * Results:
   3022  *	Number of errors reported.
   3023  *
   3024  * Side Effects:
   3025  *	None.
   3026  *-----------------------------------------------------------------------
   3027  */
   3028 int
   3029 Job_Finish()
   3030 {
   3031     if (postCommands != NILGNODE && !Lst_IsEmpty(postCommands->commands)) {
   3032 	if (errors) {
   3033 	    Error("Errors reported so .END ignored");
   3034 	} else {
   3035 	    JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL);
   3036 
   3037 	    while (nJobs) {
   3038 		Job_CatchOutput();
   3039 #ifndef RMT_WILL_WATCH
   3040 		Job_CatchChildren(!usePipes);
   3041 #endif /* RMT_WILL_WATCH */
   3042 	    }
   3043 	}
   3044     }
   3045     Job_TokenFlush();
   3046     return(errors);
   3047 }
   3048 
   3049 /*-
   3050  *-----------------------------------------------------------------------
   3051  * Job_End --
   3052  *	Cleanup any memory used by the jobs module
   3053  *
   3054  * Results:
   3055  *	None.
   3056  *
   3057  * Side Effects:
   3058  *	Memory is freed
   3059  *-----------------------------------------------------------------------
   3060  */
   3061 void
   3062 Job_End()
   3063 {
   3064 #ifdef CLEANUP
   3065     if (shellArgv)
   3066 	free(shellArgv);
   3067 #endif
   3068 }
   3069 
   3070 /*-
   3071  *-----------------------------------------------------------------------
   3072  * Job_Wait --
   3073  *	Waits for all running jobs to finish and returns. Sets 'aborting'
   3074  *	to ABORT_WAIT to prevent other jobs from starting.
   3075  *
   3076  * Results:
   3077  *	None.
   3078  *
   3079  * Side Effects:
   3080  *	Currently running jobs finish.
   3081  *
   3082  *-----------------------------------------------------------------------
   3083  */
   3084 void
   3085 Job_Wait()
   3086 {
   3087     aborting = ABORT_WAIT;
   3088     while (nJobs != 0) {
   3089 	Job_CatchOutput();
   3090 #ifndef RMT_WILL_WATCH
   3091 	Job_CatchChildren(!usePipes);
   3092 #endif /* RMT_WILL_WATCH */
   3093     }
   3094     Job_TokenFlush();
   3095     aborting = 0;
   3096 }
   3097 
   3098 /*-
   3099  *-----------------------------------------------------------------------
   3100  * Job_AbortAll --
   3101  *	Abort all currently running jobs without handling output or anything.
   3102  *	This function is to be called only in the event of a major
   3103  *	error. Most definitely NOT to be called from JobInterrupt.
   3104  *
   3105  * Results:
   3106  *	None
   3107  *
   3108  * Side Effects:
   3109  *	All children are killed, not just the firstborn
   3110  *-----------------------------------------------------------------------
   3111  */
   3112 void
   3113 Job_AbortAll()
   3114 {
   3115     LstNode           	ln;	/* element in job table */
   3116     Job            	*job;	/* the job descriptor in that element */
   3117     int     	  	foo;
   3118 
   3119     aborting = ABORT_ERROR;
   3120 
   3121     if (nJobs) {
   3122 
   3123 	(void) Lst_Open(jobs);
   3124 	while ((ln = Lst_Next(jobs)) != NILLNODE) {
   3125 	    job = (Job *) Lst_Datum(ln);
   3126 
   3127 	    /*
   3128 	     * kill the child process with increasingly drastic signals to make
   3129 	     * darn sure it's dead.
   3130 	     */
   3131 #ifdef RMT_WANTS_SIGNALS
   3132 	    if (job->flags & JOB_REMOTE) {
   3133 		Rmt_Signal(job, SIGINT);
   3134 		Rmt_Signal(job, SIGKILL);
   3135 	    } else {
   3136 		KILL(job->pid, SIGINT);
   3137 		KILL(job->pid, SIGKILL);
   3138 	    }
   3139 #else
   3140 	    KILL(job->pid, SIGINT);
   3141 	    KILL(job->pid, SIGKILL);
   3142 #endif /* RMT_WANTS_SIGNALS */
   3143 	}
   3144     }
   3145 
   3146     /*
   3147      * Catch as many children as want to report in at first, then give up
   3148      */
   3149     while (waitpid((pid_t) -1, &foo, WNOHANG) > 0)
   3150 	continue;
   3151 }
   3152 
   3153 #ifdef REMOTE
   3154 /*-
   3155  *-----------------------------------------------------------------------
   3156  * JobFlagForMigration --
   3157  *	Handle the eviction of a child. Called from RmtStatusChange.
   3158  *	Flags the child as remigratable and then suspends it.
   3159  *
   3160  * Results:
   3161  *	none.
   3162  *
   3163  * Side Effects:
   3164  *	The job descriptor is flagged for remigration.
   3165  *
   3166  *-----------------------------------------------------------------------
   3167  */
   3168 void
   3169 JobFlagForMigration(hostID)
   3170     int 	  hostID;    	/* ID of host we used, for matching children. */
   3171 {
   3172     register Job  *job;	    	/* job descriptor for dead child */
   3173     LstNode       jnode;    	/* list element for finding job */
   3174 
   3175     if (DEBUG(JOB)) {
   3176 	(void) fprintf(stdout, "JobFlagForMigration(%d) called.\n", hostID);
   3177 	(void) fflush(stdout);
   3178     }
   3179     jnode = Lst_Find(jobs, (ClientData)hostID, JobCmpRmtID);
   3180 
   3181     if (jnode == NILLNODE) {
   3182 	jnode = Lst_Find(stoppedJobs, (ClientData)hostID, JobCmpRmtID);
   3183 		if (jnode == NILLNODE) {
   3184 		    if (DEBUG(JOB)) {
   3185 			Error("Evicting host(%d) not in table", hostID);
   3186 		    }
   3187 		    return;
   3188 		}
   3189     }
   3190     job = (Job *) Lst_Datum(jnode);
   3191 
   3192     if (DEBUG(JOB)) {
   3193 	(void) fprintf(stdout,
   3194 		       "JobFlagForMigration(%d) found job '%s'.\n", hostID,
   3195 		       job->node->name);
   3196 	(void) fflush(stdout);
   3197     }
   3198 
   3199     KILL(job->pid, SIGSTOP);
   3200 
   3201     job->flags |= JOB_REMIGRATE;
   3202 }
   3203 
   3204 #endif
   3205 
   3206 /*-
   3208  *-----------------------------------------------------------------------
   3209  * JobRestartJobs --
   3210  *	Tries to restart stopped jobs if there are slots available.
   3211  *	Note that this tries to restart them regardless of pending errors.
   3212  *	It's not good to leave stopped jobs lying around!
   3213  *
   3214  * Results:
   3215  *	None.
   3216  *
   3217  * Side Effects:
   3218  *	Resumes(and possibly migrates) jobs.
   3219  *
   3220  *-----------------------------------------------------------------------
   3221  */
   3222 static void
   3223 JobRestartJobs()
   3224 {
   3225     while (!Lst_IsEmpty(stoppedJobs)) {
   3226 	if (DEBUG(JOB)) {
   3227 	    (void) fprintf(stdout, "Restarting a stopped job.\n");
   3228 	    (void) fflush(stdout);
   3229 	}
   3230 	JobRestart((Job *)Lst_DeQueue(stoppedJobs));
   3231     }
   3232 }
   3233 
   3234 #ifndef RMT_WILL_WATCH
   3235 #ifndef USE_SELECT
   3236 static void
   3237 watchfd(job)
   3238     Job *job;
   3239 {
   3240     int i;
   3241     if (job->inPollfd != NULL)
   3242 	Punt("Watching watched job");
   3243     if (fds == NULL) {
   3244 	maxfds = JBSTART;
   3245 	fds = emalloc(sizeof(struct pollfd) * maxfds);
   3246 	jobfds = emalloc(sizeof(Job **) * maxfds);
   3247 
   3248 	fds[0].fd = job_pipe[0];
   3249 	fds[0].events = POLLIN;
   3250 	jobfds[0] = &tokenWaitJob;
   3251 	tokenWaitJob.inPollfd = &fds[0];
   3252 	nfds++;
   3253     } else if (nfds == maxfds) {
   3254 	maxfds *= JBFACTOR;
   3255 	fds = erealloc(fds, sizeof(struct pollfd) * maxfds);
   3256 	jobfds = erealloc(jobfds, sizeof(Job **) * maxfds);
   3257 	for (i = 0; i < nfds; i++)
   3258 	    jobfds[i]->inPollfd = &fds[i];
   3259     }
   3260 
   3261     fds[nfds].fd = job->inPipe;
   3262     fds[nfds].events = POLLIN;
   3263     jobfds[nfds] = job;
   3264     job->inPollfd = &fds[nfds];
   3265     nfds++;
   3266 }
   3267 
   3268 static void
   3269 clearfd(job)
   3270     Job *job;
   3271 {
   3272     int i;
   3273     if (job->inPollfd == NULL)
   3274 	Punt("Unwatching unwatched job");
   3275     i = job->inPollfd - fds;
   3276     nfds--;
   3277     /*
   3278      * Move last job in table into hole made by dead job.
   3279      */
   3280     if (nfds != i) {
   3281 	fds[i] = fds[nfds];
   3282 	jobfds[i] = jobfds[nfds];
   3283 	jobfds[i]->inPollfd = &fds[i];
   3284     }
   3285     job->inPollfd = NULL;
   3286 }
   3287 
   3288 static int
   3289 readyfd(job)
   3290     Job *job;
   3291 {
   3292     if (job->inPollfd == NULL)
   3293 	Punt("Polling unwatched job");
   3294     return (job->inPollfd->revents & POLLIN) != 0;
   3295 }
   3296 #endif
   3297 #endif
   3298 
   3299 /*-
   3300  *-----------------------------------------------------------------------
   3301  * JobTokenAdd --
   3302  *	Put a token into the job pipe so that some make process can start
   3303  *	another job.
   3304  *
   3305  * Side Effects:
   3306  *	Allows more build jobs to be spawned somewhere.
   3307  *
   3308  *-----------------------------------------------------------------------
   3309  */
   3310 
   3311 static void
   3312 JobTokenAdd()
   3313 {
   3314 
   3315     if (DEBUG(JOB))
   3316 	printf("deposit token\n");
   3317     write(job_pipe[1], "+", 1);
   3318 }
   3319 
   3320 /*-
   3321  *-----------------------------------------------------------------------
   3322  * Job_ServerStartTokenAdd --
   3323  *	Prep the job token pipe in the root make process.
   3324  *
   3325  *-----------------------------------------------------------------------
   3326  */
   3327 
   3328 void Job_ServerStart(maxproc)
   3329     int maxproc;
   3330 {
   3331     int i, flags;
   3332     char jobarg[64];
   3333 
   3334     if (pipe(job_pipe) < 0)
   3335 	Fatal ("error in pipe: %s", strerror(errno));
   3336 
   3337     /*
   3338      * We mark the input side of the pipe non-blocking; we poll(2) the
   3339      * pipe when we're waiting for a job token, but we might lose the
   3340      * race for the token when a new one becomes available, so the read
   3341      * from the pipe should not block.
   3342      */
   3343     flags = fcntl(job_pipe[0], F_GETFL, 0);
   3344     flags |= O_NONBLOCK;
   3345     fcntl(job_pipe[0], F_SETFL, flags);
   3346 
   3347     /*
   3348      * Mark job pipes as close-on-exec.
   3349      * Note that we will clear this when executing submakes.
   3350      */
   3351     fcntl(job_pipe[0], F_SETFD, 1);
   3352     fcntl(job_pipe[1], F_SETFD, 1);
   3353 
   3354     snprintf(jobarg, sizeof(jobarg), "%d,%d", job_pipe[0], job_pipe[1]);
   3355 
   3356     Var_Append(MAKEFLAGS, "-J", VAR_GLOBAL);
   3357     Var_Append(MAKEFLAGS, jobarg, VAR_GLOBAL);
   3358 
   3359     /*
   3360      * Preload job_pipe with one token per job, save the one
   3361      * "extra" token for the primary job.
   3362      *
   3363      * XXX should clip maxJobs against PIPE_BUF -- if maxJobs is
   3364      * larger than the write buffer size of the pipe, we will
   3365      * deadlock here.
   3366      */
   3367     for (i=1; i < maxproc; i++)
   3368 	JobTokenAdd();
   3369 }
   3370 
   3371 /*
   3372  * this tracks the number of tokens currently "out" to build jobs.
   3373  */
   3374 int jobTokensRunning = 0;
   3375 int jobTokensFree = 0;
   3376 /*-
   3377  *-----------------------------------------------------------------------
   3378  * Job_TokenReturn --
   3379  *	Return a withdrawn token to the pool.
   3380  *
   3381  *-----------------------------------------------------------------------
   3382  */
   3383 
   3384 void
   3385 Job_TokenReturn()
   3386 {
   3387     jobTokensRunning--;
   3388     if (jobTokensRunning < 0)
   3389 	Punt("token botch");
   3390     if (jobTokensRunning)
   3391 	jobTokensFree++;
   3392 }
   3393 
   3394 /*-
   3395  *-----------------------------------------------------------------------
   3396  * Job_TokenWithdraw --
   3397  *	Attempt to withdraw a token from the pool.
   3398  *
   3399  * Results:
   3400  *	Returns TRUE if a token was withdrawn, and FALSE if the pool
   3401  *	is currently empty.
   3402  *
   3403  * Side Effects:
   3404  * 	If pool is empty, set wantToken so that we wake up
   3405  *	when a token is released.
   3406  *
   3407  *-----------------------------------------------------------------------
   3408  */
   3409 
   3410 
   3411 Boolean
   3412 Job_TokenWithdraw()
   3413 {
   3414     char tok;
   3415     int count;
   3416 
   3417     if (aborting)
   3418 	    return FALSE;
   3419 
   3420     if (jobTokensRunning == 0) {
   3421 	if (DEBUG(JOB))
   3422 	    printf("first one's free\n");
   3423 	jobTokensRunning++;
   3424 	wantToken = FALSE;
   3425 	return TRUE;
   3426     }
   3427     if (jobTokensFree > 0) {
   3428 	jobTokensFree--;
   3429 	jobTokensRunning++;
   3430 	wantToken = FALSE;
   3431 	return TRUE;
   3432     }
   3433     count = read(job_pipe[0], &tok, 1);
   3434     if (count == 0)
   3435 	Fatal("eof on job pipe!");
   3436     else if (count < 0) {
   3437 	if (errno != EAGAIN) {
   3438 	    Fatal("job pipe read: %s", strerror(errno));
   3439 	}
   3440 	if (DEBUG(JOB))
   3441 	    printf("blocked for token\n");
   3442 	wantToken = TRUE;
   3443 	return FALSE;
   3444     }
   3445     wantToken = FALSE;
   3446     jobTokensRunning++;
   3447     if (DEBUG(JOB))
   3448 	printf("withdrew token\n");
   3449     return TRUE;
   3450 }
   3451 
   3452 /*-
   3453  *-----------------------------------------------------------------------
   3454  * Job_TokenFlush --
   3455  *	Return free tokens to the pool.
   3456  *
   3457  *-----------------------------------------------------------------------
   3458  */
   3459 
   3460 void
   3461 Job_TokenFlush()
   3462 {
   3463     if (compatMake) return;
   3464 
   3465     while (jobTokensFree > 0) {
   3466 	JobTokenAdd();
   3467 	jobTokensFree--;
   3468     }
   3469 }
   3470 
   3471