coshell - shell coprocess support

The coshell routines support the shell as a coprocess. This coprocess may be either ksh(1) or sh(1) executing on the local host, or it may be coshell(1) with access to shells on hosts throughout the local network.

The coshell inherits the environment of the calling process. Signals sent to the calling process are passed to the coshell.

More than one coshell may be open in the current process. If the Coshell_t* argument to the cowait(), cokill(), copending(), cozombie(), or coclose() calls below is 0 then the call is applied to all open coshell.

Coshell_t* coopen(const char* shell, int flags, const char* attributes)

Returns a pointer to a new coshell. NULL is returned on error. If shell is NULL then the coshell executable is determined by doing the usual path search, in order, on the value of the environment variable COSHELL and the commands ksh and sh. flags is the inclusive-or of the following:

CO_ANY

Return a pointer to a previously opened coshell if possible, otherwise open a new coshell.

CO_DEBUG

Enable library debug tracing.

CO_IGNORE

Ignore any command errors. By default any command error that is not tested by a condtional causes the job to terminate.

CO_LOCAL

Commands are to be executed on the local host only.

CO_SHELL

The caller is sh(1): internal file descriptors are moved to 10 or above; SIGSTOP and SIGCONT handlers are not installed.

CO_SILENT

Don't trace commands. By default commands are traced using the shell -x option.

CO_NONBLOCK

Normally coexec() blocks when the job queue is full and waits until a job completes. CO_NONBLOCK causes coexec() to return NULL when the job queue is full.

attributes is a string that is interpreted by the coshell. If attributes is NULL then the value of the environment variable COATTRIBUTES is used if defined. ksh and sh ignore this string. The return value points to a structure with the following readonly elements:

int flags

The default flags.

int outstanding

The number of jobs that have not been waited for.

int running

The number of jobs still running.

int total

The total number of jobs sent to the coshell.

unsigned long user

The total user time of all completed jobs in 1/CO_QUANT second increments.

unsigned long sys

The total system time of all completed jobs in 1/CO_QUANT second increments.

int coclose(Coshell_t* sh)

Close an open coshell pointed to by sh. The coshell exit status is returned.

Cojob_t* coexec(Coshell_t* sh, const char* cmd, int flags, const char* out, const char* err, const char* att)

Sends the shell command line cmd to the open coshell pointed to by sh for execution. flags are the same as in the coopen() call, and are used to augment the default settings from coopen(). out is the standard output file name and defaults to stdout if NULL. err is the standard error file name and defaults to stderr if NULL. att, if non-NULL, contains job attributes that are appended to the attributes from coopen() before being sent to the coshell. The return value points to a structure with the following elements:

int id

A number that uniquely identifies the job within the coshell.

int status

The job exit status, valid only for job pointers returned by cowait().

int flags

The flags enabled for this job.

void* local

A user reserved pointer, initially set to NULL on return from coexec(). This is the only job field that may be modified by the user. The user defined value is preserved until after the cowait() call that returns the job pointer.

unsigned long user

The user time of this job in 1/CO_QUANT second increments, valid only for job pointers returned by cowait().

unsigned long sys

The system time of this job in 1/CO_QUANT second increments, valid only for job pointers returned by cowait().

Cojob_t* cowait(Coshell_t* sh, Cojob_t* job, int timeout)

Returns the job pointer in the coshell pointed to by sh for the job pointed to by job. If job is NULL then a pointer to any completed job is returned. cowait() blocks until the specified job(s) complete. NULL is returned on error or if all jobs have completed and errno is set to EINVAL for coshell communication errors, ECHILD if job is NULL and there are no children, and ESRCH if job is not NULL and not an active job. cozombie(sh) is the number of jobs that may be waited for without blocking. The return value status, user and sys job fields are set to their final values. The return value is valid until the next coexec(), cowait() or coclose() call. timeout is the maximum time in milliseconds that wait will block. If the wait times out then 0 is returned. A negative timeout waits until a job completes or a signal is received.

int cojobs(Coshell_t* sh)

Returns the number of outstanding jobs that are children of the caller. (Remote jobs or jobs executed by a separate daemon are not counted here.)

int copending(Coshell_t* sh)

Returns the number of pending jobs; this is the number of cowait() calls required to reap all running jobs.

int cozombie(Coshell_t* sh)

Returns the number of jobs that have completed but have not been cowait()'d for.

int cokill(Coshell_t* sh, Cojob_t* job, int sig)

The signal sig is sent to the job pointed to by job running in the coshell pointed to by sh. If job is NULL then the signal is sent to all jobs in the coshell. If both sh and job are NULL then the signal is sent to all jobs in all coshells. -1 is returned on error, 0 otherwise.

int cosync(Coshell_t* sh, const char* path, int fd, int mode)

Sync all outstanding file operations for either the file path or the file descriptor fd after its shell action has completed in sh. If path is NULL then fd is used. If fd<0 and mode>=0 then path is opened using mode. This is an unfortunate workaround for NFS and remote coshells, and is a no-op for all real file systems. It should be called after cowait() to ensure that all file system cache info has been flushed. sync(2) or fsync(2) must still be called to schedule file data to be written to disk.

char* coinit(Coshell_t* sh)

Returns the shell initialization commands for the next job. These commands represent process state changes that may have occurred since the last call to coinit(), e.g., current working directory or umask. If sh is local (a child of the calling process) coinit() may return the empty string; if sh is remote then considerably more information may be returned. This routine is used by remote coshell implementations and is not normally called from user code.

void coquote(Sfio_T* sp, const char* string, int type)

Applies shell single quoting to string and copies the result into the sfio stream sp. If type!=0 then any occurence of /hosttype/ is translated to /$HOSTTYPE/, where hosttype is the current value of the HOSTTYPE environment variable. This routine is used by remote coshell implementations and is not normally called from user code.

cosync() is a hack workaround, but we do have to work in the real world.

A bug in bsh(1) and ksh(1) implementations up to and including ksh88e causes some interrupted jobs to return 0 exit status. This should be fixed in later shell releases.

trap 0 is reserved by coexec() at the outermost scope. To use trap 0 use (...) to force a subshell.

coshell(1), ksh(1), nmake(1), sh(1), cs(3), libast(3)