mksh/shf.c

/*	$OpenBSD: shf.c,v 1.16 2013/04/19 17:36:09 millert Exp $	*/

/*-
 * Copyright (c) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2011,
 *		 2012, 2013, 2015, 2016, 2017
 *	mirabilos <m@mirbsd.org>
 * Copyright (c) 2015
 *	Daniel Richard G. <skunk@iSKUNK.ORG>
 *
 * Provided that these terms and disclaimer and all copyright notices
 * are retained or reproduced in an accompanying document, permission
 * is granted to deal in this work without restriction, including un-
 * limited rights to use, publicly perform, distribute, sell, modify,
 * merge, give away, or sublicence.
 *
 * This work is provided "AS IS" and WITHOUT WARRANTY of any kind, to
 * the utmost extent permitted by applicable law, neither express nor
 * implied; without malicious intent or gross negligence. In no event
 * may a licensor, author or contributor be held liable for indirect,
 * direct, other damage, loss, or other issues arising in any way out
 * of dealing in the work, even if advised of the possibility of such
 * damage or existence of a defect, except proven that it results out
 * of said person's immediate fault when using the work as intended.
 *-
 * Use %zX instead of %p and floating point isn't supported at all.
 */

#include "sh.h"

__RCSID("$MirOS: src/bin/mksh/shf.c,v 1.95 2017/05/05 22:45:58 tg Exp $");

/* flags to shf_emptybuf() */
#define EB_READSW	0x01	/* about to switch to reading */
#define EB_GROW		0x02	/* grow buffer if necessary (STRING+DYNAMIC) */

/*
 * Replacement stdio routines. Stdio is too flakey on too many machines
 * to be useful when you have multiple processes using the same underlying
 * file descriptors.
 */

static int shf_fillbuf(struct shf *);
static int shf_emptybuf(struct shf *, int);

/*
 * Open a file. First three args are for open(), last arg is flags for
 * this package. Returns NULL if file could not be opened, or if a dup
 * fails.
 */
struct shf *
shf_open(const char *name, int oflags, int mode, int sflags)
{
	struct shf *shf;
	ssize_t bsize =
	    /* at most 512 */
	    sflags & SHF_UNBUF ? (sflags & SHF_RD ? 1 : 0) : SHF_BSIZE;
	int fd, eno;

	/* Done before open so if alloca fails, fd won't be lost. */
	shf = alloc(sizeof(struct shf) + bsize, ATEMP);
	shf->areap = ATEMP;
	shf->buf = (unsigned char *)&shf[1];
	shf->bsize = bsize;
	shf->flags = SHF_ALLOCS;
	/* Rest filled in by reopen. */

	fd = binopen3(name, oflags, mode);
	if (fd < 0) {
		eno = errno;
		afree(shf, shf->areap);
		errno = eno;
		return (NULL);
	}
	if ((sflags & SHF_MAPHI) && fd < FDBASE) {
		int nfd;

		nfd = fcntl(fd, F_DUPFD, FDBASE);
		eno = errno;
		close(fd);
		if (nfd < 0) {
			afree(shf, shf->areap);
			errno = eno;
			return (NULL);
		}
		fd = nfd;
	}
	sflags &= ~SHF_ACCMODE;
	sflags |= (oflags & O_ACCMODE) == O_RDONLY ? SHF_RD :
	    ((oflags & O_ACCMODE) == O_WRONLY ? SHF_WR : SHF_RDWR);

	return (shf_reopen(fd, sflags, shf));
}

/* helper function for shf_fdopen and shf_reopen */
static void
shf_open_hlp(int fd, int *sflagsp, const char *where)
{
	int sflags = *sflagsp;

	/* use fcntl() to figure out correct read/write flags */
	if (sflags & SHF_GETFL) {
		int flags = fcntl(fd, F_GETFL, 0);

		if (flags < 0)
			/* will get an error on first read/write */
			sflags |= SHF_RDWR;
		else {
			switch (flags & O_ACCMODE) {
			case O_RDONLY:
				sflags |= SHF_RD;
				break;
			case O_WRONLY:
				sflags |= SHF_WR;
				break;
			case O_RDWR:
				sflags |= SHF_RDWR;
				break;
			}
		}
		*sflagsp = sflags;
	}

	if (!(sflags & (SHF_RD | SHF_WR)))
		internal_errorf(Tf_sD_s, where, "missing read/write");
}

/* Set up the shf structure for a file descriptor. Doesn't fail. */
struct shf *
shf_fdopen(int fd, int sflags, struct shf *shf)
{
	ssize_t bsize =
	    /* at most 512 */
	    sflags & SHF_UNBUF ? (sflags & SHF_RD ? 1 : 0) : SHF_BSIZE;

	shf_open_hlp(fd, &sflags, "shf_fdopen");
	if (shf) {
		if (bsize) {
			shf->buf = alloc(bsize, ATEMP);
			sflags |= SHF_ALLOCB;
		} else
			shf->buf = NULL;
	} else {
		shf = alloc(sizeof(struct shf) + bsize, ATEMP);
		shf->buf = (unsigned char *)&shf[1];
		sflags |= SHF_ALLOCS;
	}
	shf->areap = ATEMP;
	shf->fd = fd;
	shf->rp = shf->wp = shf->buf;
	shf->rnleft = 0;
	shf->rbsize = bsize;
	shf->wnleft = 0; /* force call to shf_emptybuf() */
	shf->wbsize = sflags & SHF_UNBUF ? 0 : bsize;
	shf->flags = sflags;
	shf->errnosv = 0;
	shf->bsize = bsize;
	if (sflags & SHF_CLEXEC)
		fcntl(fd, F_SETFD, FD_CLOEXEC);
	return (shf);
}

/* Set up an existing shf (and buffer) to use the given fd */
struct shf *
shf_reopen(int fd, int sflags, struct shf *shf)
{
	ssize_t bsize =
	    /* at most 512 */
	    sflags & SHF_UNBUF ? (sflags & SHF_RD ? 1 : 0) : SHF_BSIZE;

	shf_open_hlp(fd, &sflags, "shf_reopen");
	if (!shf || !shf->buf || shf->bsize < bsize)
		internal_errorf(Tf_sD_s, "shf_reopen", Tbad_bsize);

	/* assumes shf->buf and shf->bsize already set up */
	shf->fd = fd;
	shf->rp = shf->wp = shf->buf;
	shf->rnleft = 0;
	shf->rbsize = bsize;
	shf->wnleft = 0; /* force call to shf_emptybuf() */
	shf->wbsize = sflags & SHF_UNBUF ? 0 : bsize;
	shf->flags = (shf->flags & (SHF_ALLOCS | SHF_ALLOCB)) | sflags;
	shf->errnosv = 0;
	if (sflags & SHF_CLEXEC)
		fcntl(fd, F_SETFD, FD_CLOEXEC);
	return (shf);
}

/*
 * Open a string for reading or writing. If reading, bsize is the number
 * of bytes that can be read. If writing, bsize is the maximum number of
 * bytes that can be written. If shf is not NULL, it is filled in and
 * returned, if it is NULL, shf is allocated. If writing and buf is NULL
 * and SHF_DYNAMIC is set, the buffer is allocated (if bsize > 0, it is
 * used for the initial size). Doesn't fail.
 * When writing, a byte is reserved for a trailing NUL - see shf_sclose().
 */
struct shf *
shf_sopen(char *buf, ssize_t bsize, int sflags, struct shf *shf)
{
	/* can't have a read+write string */
	if (!(!(sflags & SHF_RD) ^ !(sflags & SHF_WR)))
		internal_errorf(Tf_flags, "shf_sopen",
		    (unsigned int)sflags);

	if (!shf) {
		shf = alloc(sizeof(struct shf), ATEMP);
		sflags |= SHF_ALLOCS;
	}
	shf->areap = ATEMP;
	if (!buf && (sflags & SHF_WR) && (sflags & SHF_DYNAMIC)) {
		if (bsize <= 0)
			bsize = 64;
		sflags |= SHF_ALLOCB;
		buf = alloc(bsize, shf->areap);
	}
	shf->fd = -1;
	shf->buf = shf->rp = shf->wp = (unsigned char *)buf;
	shf->rnleft = bsize;
	shf->rbsize = bsize;
	shf->wnleft = bsize - 1;	/* space for a '\0' */
	shf->wbsize = bsize;
	shf->flags = sflags | SHF_STRING;
	shf->errnosv = 0;
	shf->bsize = bsize;

	return (shf);
}

/* Flush and close file descriptor, free the shf structure */
int
shf_close(struct shf *shf)
{
	int ret = 0;

	if (shf->fd >= 0) {
		ret = shf_flush(shf);
		if (close(shf->fd) < 0)
			ret = -1;
	}
	if (shf->flags & SHF_ALLOCS)
		afree(shf, shf->areap);
	else if (shf->flags & SHF_ALLOCB)
		afree(shf->buf, shf->areap);

	return (ret);
}

/* Flush and close file descriptor, don't free file structure */
int
shf_fdclose(struct shf *shf)
{
	int ret = 0;

	if (shf->fd >= 0) {
		ret = shf_flush(shf);
		if (close(shf->fd) < 0)
			ret = -1;
		shf->rnleft = 0;
		shf->rp = shf->buf;
		shf->wnleft = 0;
		shf->fd = -1;
	}

	return (ret);
}

/*
 * Close a string - if it was opened for writing, it is NUL terminated;
 * returns a pointer to the string and frees shf if it was allocated
 * (does not free string if it was allocated).
 */
char *
shf_sclose(struct shf *shf)
{
	unsigned char *s = shf->buf;

	/* NUL terminate */
	if (shf->flags & SHF_WR) {
		shf->wnleft++;
		shf_putc('\0', shf);
	}
	if (shf->flags & SHF_ALLOCS)
		afree(shf, shf->areap);
	return ((char *)s);
}

/*
 * Un-read what has been read but not examined, or write what has been
 * buffered. Returns 0 for success, -1 for (write) error.
 */
int
shf_flush(struct shf *shf)
{
	int rv = 0;

	if (shf->flags & SHF_STRING)
		rv = (shf->flags & SHF_WR) ? -1 : 0;
	else if (shf->fd < 0)
		internal_errorf(Tf_sD_s, "shf_flush", "no fd");
	else if (shf->flags & SHF_ERROR) {
		errno = shf->errnosv;
		rv = -1;
	} else if (shf->flags & SHF_READING) {
		shf->flags &= ~(SHF_EOF | SHF_READING);
		if (shf->rnleft > 0) {
			if (lseek(shf->fd, (off_t)-shf->rnleft,
			    SEEK_CUR) == -1) {
				shf->flags |= SHF_ERROR;
				shf->errnosv = errno;
				rv = -1;
			}
			shf->rnleft = 0;
			shf->rp = shf->buf;
		}
	} else if (shf->flags & SHF_WRITING)
		rv = shf_emptybuf(shf, 0);

	return (rv);
}

/*
 * Write out any buffered data. If currently reading, flushes the read
 * buffer. Returns 0 for success, -1 for (write) error.
 */
static int
shf_emptybuf(struct shf *shf, int flags)
{
	int ret = 0;

	if (!(shf->flags & SHF_STRING) && shf->fd < 0)
		internal_errorf(Tf_sD_s, "shf_emptybuf", "no fd");

	if (shf->flags & SHF_ERROR) {
		errno = shf->errnosv;
		return (-1);
	}

	if (shf->flags & SHF_READING) {
		if (flags & EB_READSW)
			/* doesn't happen */
			return (0);
		ret = shf_flush(shf);
		shf->flags &= ~SHF_READING;
	}
	if (shf->flags & SHF_STRING) {
		unsigned char *nbuf;

		/*
		 * Note that we assume SHF_ALLOCS is not set if
		 * SHF_ALLOCB is set... (changing the shf pointer could
		 * cause problems)
		 */
		if (!(flags & EB_GROW) || !(shf->flags & SHF_DYNAMIC) ||
		    !(shf->flags & SHF_ALLOCB))
			return (-1);
		/* allocate more space for buffer */
		nbuf = aresize2(shf->buf, 2, shf->wbsize, shf->areap);
		shf->rp = nbuf + (shf->rp - shf->buf);
		shf->wp = nbuf + (shf->wp - shf->buf);
		shf->rbsize += shf->wbsize;
		shf->wnleft += shf->wbsize;
		shf->wbsize <<= 1;
		shf->buf = nbuf;
	} else {
		if (shf->flags & SHF_WRITING) {
			ssize_t n, ntowrite = shf->wp - shf->buf;
			unsigned char *buf = shf->buf;

			while (ntowrite > 0) {
				n = write(shf->fd, buf, ntowrite);
				if (n < 0) {
					if (errno == EINTR &&
					    !(shf->flags & SHF_INTERRUPT))
						continue;
					shf->flags |= SHF_ERROR;
					shf->errnosv = errno;
					shf->wnleft = 0;
					if (buf != shf->buf) {
						/*
						 * allow a second flush
						 * to work
						 */
						memmove(shf->buf, buf,
						    ntowrite);
						shf->wp = shf->buf + ntowrite;
					}
					return (-1);
				}
				buf += n;
				ntowrite -= n;
			}
			if (flags & EB_READSW) {
				shf->wp = shf->buf;
				shf->wnleft = 0;
				shf->flags &= ~SHF_WRITING;
				return (0);
			}
		}
		shf->wp = shf->buf;
		shf->wnleft = shf->wbsize;
	}
	shf->flags |= SHF_WRITING;

	return (ret);
}

/* Fill up a read buffer. Returns -1 for a read error, 0 otherwise. */
static int
shf_fillbuf(struct shf *shf)
{
	ssize_t n;

	if (shf->flags & SHF_STRING)
		return (0);

	if (shf->fd < 0)
		internal_errorf(Tf_sD_s, "shf_fillbuf", "no fd");

	if (shf->flags & (SHF_EOF | SHF_ERROR)) {
		if (shf->flags & SHF_ERROR)
			errno = shf->errnosv;
		return (-1);
	}

	if ((shf->flags & SHF_WRITING) && shf_emptybuf(shf, EB_READSW) == -1)
		return (-1);

	shf->flags |= SHF_READING;

	shf->rp = shf->buf;
	while (/* CONSTCOND */ 1) {
		n = blocking_read(shf->fd, (char *)shf->buf, shf->rbsize);
		if (n < 0 && errno == EINTR && !(shf->flags & SHF_INTERRUPT))
			continue;
		break;
	}
	if (n < 0) {
		shf->flags |= SHF_ERROR;
		shf->errnosv = errno;
		shf->rnleft = 0;
		shf->rp = shf->buf;
		return (-1);
	}
	if ((shf->rnleft = n) == 0)
		shf->flags |= SHF_EOF;
	return (0);
}

/*
 * Read a buffer from shf. Returns the number of bytes read into buf, if
 * no bytes were read, returns 0 if end of file was seen, -1 if a read
 * error occurred.
 */
ssize_t
shf_read(char *buf, ssize_t bsize, struct shf *shf)
{
	ssize_t ncopy, orig_bsize = bsize;

	if (!(shf->flags & SHF_RD))
		internal_errorf(Tf_flags, Tshf_read,
		    (unsigned int)shf->flags);

	if (bsize <= 0)
		internal_errorf(Tf_szs, Tshf_read, bsize, Tbsize);

	while (bsize > 0) {
		if (shf->rnleft == 0 &&
		    (shf_fillbuf(shf) == -1 || shf->rnleft == 0))
			break;
		ncopy = shf->rnleft;
		if (ncopy > bsize)
			ncopy = bsize;
		memcpy(buf, shf->rp, ncopy);
		buf += ncopy;
		bsize -= ncopy;
		shf->rp += ncopy;
		shf->rnleft -= ncopy;
	}
	/* Note: fread(3S) returns 0 for errors - this doesn't */
	return (orig_bsize == bsize ? (shf_error(shf) ? -1 : 0) :
	    orig_bsize - bsize);
}

/*
 * Read up to a newline or -1. The newline is put in buf; buf is always
 * NUL terminated. Returns NULL on read error or if nothing was read
 * before end of file, returns a pointer to the NUL byte in buf
 * otherwise.
 */
char *
shf_getse(char *buf, ssize_t bsize, struct shf *shf)
{
	unsigned char *end;
	ssize_t ncopy;
	char *orig_buf = buf;

	if (!(shf->flags & SHF_RD))
		internal_errorf(Tf_flags, "shf_getse",
		    (unsigned int)shf->flags);

	if (bsize <= 0)
		return (NULL);

	/* save room for NUL */
	--bsize;
	do {
		if (shf->rnleft == 0) {
			if (shf_fillbuf(shf) == -1)
				return (NULL);
			if (shf->rnleft == 0) {
				*buf = '\0';
				return (buf == orig_buf ? NULL : buf);
			}
		}
		end = (unsigned char *)memchr((char *)shf->rp, '\n',
		    shf->rnleft);
		ncopy = end ? end - shf->rp + 1 : shf->rnleft;
		if (ncopy > bsize)
			ncopy = bsize;
		memcpy(buf, (char *) shf->rp, ncopy);
		shf->rp += ncopy;
		shf->rnleft -= ncopy;
		buf += ncopy;
		bsize -= ncopy;
#ifdef MKSH_WITH_TEXTMODE
		if (end && buf > orig_buf + 1 && buf[-2] == '\r') {
			buf--;
			bsize++;
			buf[-1] = '\n';
		}
#endif
	} while (!end && bsize);
#ifdef MKSH_WITH_TEXTMODE
	if (!bsize && buf[-1] == '\r') {
		int c = shf_getc(shf);
		if (c == '\n')
			buf[-1] = '\n';
		else if (c != -1)
			shf_ungetc(c, shf);
	}
#endif
	*buf = '\0';
	return (buf);
}

/* Returns the char read. Returns -1 for error and end of file. */
int
shf_getchar(struct shf *shf)
{
	if (!(shf->flags & SHF_RD))
		internal_errorf(Tf_flags, "shf_getchar",
		    (unsigned int)shf->flags);

	if (shf->rnleft == 0 && (shf_fillbuf(shf) == -1 || shf->rnleft == 0))
		return (-1);
	--shf->rnleft;
	return (*shf->rp++);
}

/*
 * Put a character back in the input stream. Returns the character if
 * successful, -1 if there is no room.
 */
int
shf_ungetc(int c, struct shf *shf)
{
	if (!(shf->flags & SHF_RD))
		internal_errorf(Tf_flags, "shf_ungetc",
		    (unsigned int)shf->flags);

	if ((shf->flags & SHF_ERROR) || c == -1 ||
	    (shf->rp == shf->buf && shf->rnleft))
		return (-1);

	if ((shf->flags & SHF_WRITING) && shf_emptybuf(shf, EB_READSW) == -1)
		return (-1);

	if (shf->rp == shf->buf)
		shf->rp = shf->buf + shf->rbsize;
	if (shf->flags & SHF_STRING) {
		/*
		 * Can unget what was read, but not something different;
		 * we don't want to modify a string.
		 */
		if ((int)(shf->rp[-1]) != c)
			return (-1);
		shf->flags &= ~SHF_EOF;
		shf->rp--;
		shf->rnleft++;
		return (c);
	}
	shf->flags &= ~SHF_EOF;
	*--(shf->rp) = c;
	shf->rnleft++;
	return (c);
}

/*
 * Write a character. Returns the character if successful, -1 if the
 * char could not be written.
 */
int
shf_putchar(int c, struct shf *shf)
{
	if (!(shf->flags & SHF_WR))
		internal_errorf(Tf_flags, "shf_putchar",
		    (unsigned int)shf->flags);

	if (c == -1)
		return (-1);

	if (shf->flags & SHF_UNBUF) {
		unsigned char cc = (unsigned char)c;
		ssize_t n;

		if (shf->fd < 0)
			internal_errorf(Tf_sD_s, "shf_putchar", "no fd");
		if (shf->flags & SHF_ERROR) {
			errno = shf->errnosv;
			return (-1);
		}
		while ((n = write(shf->fd, &cc, 1)) != 1)
			if (n < 0) {
				if (errno == EINTR &&
				    !(shf->flags & SHF_INTERRUPT))
					continue;
				shf->flags |= SHF_ERROR;
				shf->errnosv = errno;
				return (-1);
			}
	} else {
		/* Flush deals with strings and sticky errors */
		if (shf->wnleft == 0 && shf_emptybuf(shf, EB_GROW) == -1)
			return (-1);
		shf->wnleft--;
		*shf->wp++ = c;
	}

	return (c);
}

/*
 * Write a string. Returns the length of the string if successful, -1
 * if the string could not be written.
 */
ssize_t
shf_puts(const char *s, struct shf *shf)
{
	if (!s)
		return (-1);

	return (shf_write(s, strlen(s), shf));
}

/* Write a buffer. Returns nbytes if successful, -1 if there is an error. */
ssize_t
shf_write(const char *buf, ssize_t nbytes, struct shf *shf)
{
	ssize_t n, ncopy, orig_nbytes = nbytes;

	if (!(shf->flags & SHF_WR))
		internal_errorf(Tf_flags, Tshf_write,
		    (unsigned int)shf->flags);

	if (nbytes < 0)
		internal_errorf(Tf_szs, Tshf_write, nbytes, Tbytes);

	/* Don't buffer if buffer is empty and we're writting a large amount. */
	if ((ncopy = shf->wnleft) &&
	    (shf->wp != shf->buf || nbytes < shf->wnleft)) {
		if (ncopy > nbytes)
			ncopy = nbytes;
		memcpy(shf->wp, buf, ncopy);
		nbytes -= ncopy;
		buf += ncopy;
		shf->wp += ncopy;
		shf->wnleft -= ncopy;
	}
	if (nbytes > 0) {
		if (shf->flags & SHF_STRING) {
			/* resize buffer until there's enough space left */
			while (nbytes > shf->wnleft)
				if (shf_emptybuf(shf, EB_GROW) == -1)
					return (-1);
			/* then write everything into the buffer */
		} else {
			/* flush deals with sticky errors */
			if (shf_emptybuf(shf, EB_GROW) == -1)
				return (-1);
			/* write chunks larger than window size directly */
			if (nbytes > shf->wbsize) {
				ncopy = nbytes;
				if (shf->wbsize)
					ncopy -= nbytes % shf->wbsize;
				nbytes -= ncopy;
				while (ncopy > 0) {
					n = write(shf->fd, buf, ncopy);
					if (n < 0) {
						if (errno == EINTR &&
						    !(shf->flags & SHF_INTERRUPT))
							continue;
						shf->flags |= SHF_ERROR;
						shf->errnosv = errno;
						shf->wnleft = 0;
						/*
						 * Note: fwrite(3) returns 0
						 * for errors - this doesn't
						 */
						return (-1);
					}
					buf += n;
					ncopy -= n;
				}
			}
			/* ... and buffer the rest */
		}
		if (nbytes > 0) {
			/* write remaining bytes to buffer */
			memcpy(shf->wp, buf, nbytes);
			shf->wp += nbytes;
			shf->wnleft -= nbytes;
		}
	}

	return (orig_nbytes);
}

ssize_t
shf_fprintf(struct shf *shf, const char *fmt, ...)
{
	va_list args;
	ssize_t n;

	va_start(args, fmt);
	n = shf_vfprintf(shf, fmt, args);
	va_end(args);

	return (n);
}

ssize_t
shf_snprintf(char *buf, ssize_t bsize, const char *fmt, ...)
{
	struct shf shf;
	va_list args;
	ssize_t n;

	if (!buf || bsize <= 0)
		internal_errorf("shf_snprintf: buf %zX, bsize %zd",
		    (size_t)buf, bsize);

	shf_sopen(buf, bsize, SHF_WR, &shf);
	va_start(args, fmt);
	n = shf_vfprintf(&shf, fmt, args);
	va_end(args);
	/* NUL terminates */
	shf_sclose(&shf);
	return (n);
}

char *
shf_smprintf(const char *fmt, ...)
{
	struct shf shf;
	va_list args;

	shf_sopen(NULL, 0, SHF_WR|SHF_DYNAMIC, &shf);
	va_start(args, fmt);
	shf_vfprintf(&shf, fmt, args);
	va_end(args);
	/* NUL terminates */
	return (shf_sclose(&shf));
}

#define	FL_HASH		0x001	/* '#' seen */
#define FL_PLUS		0x002	/* '+' seen */
#define FL_RIGHT	0x004	/* '-' seen */
#define FL_BLANK	0x008	/* ' ' seen */
#define FL_SHORT	0x010	/* 'h' seen */
#define FL_LONG		0x020	/* 'l' seen */
#define FL_ZERO		0x040	/* '0' seen */
#define FL_DOT		0x080	/* '.' seen */
#define FL_UPPER	0x100	/* format character was uppercase */
#define FL_NUMBER	0x200	/* a number was formated %[douxefg] */
#define FL_SIZET	0x400	/* 'z' seen */
#define FM_SIZES	0x430	/* h/l/z mask */

ssize_t
shf_vfprintf(struct shf *shf, const char *fmt, va_list args)
{
	const char *s;
	char c, *cp;
	int tmp = 0, flags;
	size_t field, precision, len;
	unsigned long lnum;
	/* %#o produces the longest output */
	char numbuf[(8 * sizeof(long) + 2) / 3 + 1 + /* NUL */ 1];
	/* this stuff for dealing with the buffer */
	ssize_t nwritten = 0;

#define VA(type) va_arg(args, type)

	if (!fmt)
		return (0);

	while ((c = *fmt++)) {
		if (c != '%') {
			shf_putc(c, shf);
			nwritten++;
			continue;
		}
		/*
		 * This will accept flags/fields in any order - not just
		 * the order specified in printf(3), but this is the way
		 * _doprnt() seems to work (on BSD and SYSV). The only
		 * restriction is that the format character must come
		 * last :-).
		 */
		flags = 0;
		field = precision = 0;
		while ((c = *fmt++)) {
			switch (c) {
			case '#':
				flags |= FL_HASH;
				continue;

			case '+':
				flags |= FL_PLUS;
				continue;

			case '-':
				flags |= FL_RIGHT;
				continue;

			case ' ':
				flags |= FL_BLANK;
				continue;

			case '0':
				if (!(flags & FL_DOT))
					flags |= FL_ZERO;
				continue;

			case '.':
				flags |= FL_DOT;
				precision = 0;
				continue;

			case '*':
				tmp = VA(int);
				if (tmp < 0) {
					if (flags & FL_DOT)
						precision = 0;
					else {
						field = (unsigned int)-tmp;
						flags |= FL_RIGHT;
					}
				} else if (flags & FL_DOT)
					precision = (unsigned int)tmp;
				else
					field = (unsigned int)tmp;
				continue;

			case 'l':
				flags &= ~FM_SIZES;
				flags |= FL_LONG;
				continue;

			case 'h':
				flags &= ~FM_SIZES;
				flags |= FL_SHORT;
				continue;

			case 'z':
				flags &= ~FM_SIZES;
				flags |= FL_SIZET;
				continue;
			}
			if (ctype(c, C_DIGIT)) {
				bool overflowed = false;

				tmp = ksh_numdig(c);
				while (ctype((c = *fmt++), C_DIGIT))
					if (notok2mul(2147483647, tmp, 10))
						overflowed = true;
					else
						tmp = tmp * 10 + ksh_numdig(c);
				--fmt;
				if (overflowed)
					tmp = 0;
				if (flags & FL_DOT)
					precision = (unsigned int)tmp;
				else
					field = (unsigned int)tmp;
				continue;
			}
			break;
		}

		if (!c)
			/* nasty format */
			break;

		if (ctype(c, C_UPPER)) {
			flags |= FL_UPPER;
			c = ksh_tolower(c);
		}

		switch (c) {
		case 'd':
		case 'i':
			if (flags & FL_SIZET)
				lnum = (long)VA(ssize_t);
			else if (flags & FL_LONG)
				lnum = VA(long);
			else if (flags & FL_SHORT)
				lnum = (long)(short)VA(int);
			else
				lnum = (long)VA(int);
			goto integral;

		case 'o':
		case 'u':
		case 'x':
			if (flags & FL_SIZET)
				lnum = VA(size_t);
			else if (flags & FL_LONG)
				lnum = VA(unsigned long);
			else if (flags & FL_SHORT)
				lnum = (unsigned long)(unsigned short)VA(int);
			else
				lnum = (unsigned long)VA(unsigned int);

 integral:
			flags |= FL_NUMBER;
			cp = numbuf + sizeof(numbuf);
			*--cp = '\0';

			switch (c) {
			case 'd':
			case 'i':
				if (0 > (long)lnum) {
					lnum = -(long)lnum;
					tmp = 1;
				} else
					tmp = 0;
				/* FALLTHROUGH */
			case 'u':
				do {
					*--cp = digits_lc[lnum % 10];
					lnum /= 10;
				} while (lnum);

				if (c != 'u') {
					if (tmp)
						*--cp = '-';
					else if (flags & FL_PLUS)
						*--cp = '+';
					else if (flags & FL_BLANK)
						*--cp = ' ';
				}
				break;

			case 'o':
				do {
					*--cp = digits_lc[lnum & 0x7];
					lnum >>= 3;
				} while (lnum);

				if ((flags & FL_HASH) && *cp != '0')
					*--cp = '0';
				break;

			case 'x': {
				const char *digits = (flags & FL_UPPER) ?
				    digits_uc : digits_lc;
				do {
					*--cp = digits[lnum & 0xF];
					lnum >>= 4;
				} while (lnum);

				if (flags & FL_HASH) {
					*--cp = (flags & FL_UPPER) ? 'X' : 'x';
					*--cp = '0';
				}
			    }
			}
			len = numbuf + sizeof(numbuf) - 1 - (s = cp);
			if (flags & FL_DOT) {
				if (precision > len) {
					field = precision;
					flags |= FL_ZERO;
				} else
					/* no loss */
					precision = len;
			}
			break;

		case 's':
			if ((s = VA(const char *)) == NULL)
				s = "(null)";
			else if (flags & FL_HASH) {
				print_value_quoted(shf, s);
				continue;
			}
			len = utf_mbswidth(s);
			break;

		case 'c':
			flags &= ~FL_DOT;
			c = (char)(VA(int));
			/* FALLTHROUGH */

		case '%':
		default:
			numbuf[0] = c;
			numbuf[1] = 0;
			s = numbuf;
			len = 1;
			break;
		}

		/*
		 * At this point s should point to a string that is to be
		 * formatted, and len should be the length of the string.
		 */
		if (!(flags & FL_DOT) || len < precision)
			precision = len;
		if (field > precision) {
			field -= precision;
			if (!(flags & FL_RIGHT)) {
				/* skip past sign or 0x when padding with 0 */
				if ((flags & FL_ZERO) && (flags & FL_NUMBER)) {
					if (ctype(*s, C_SPC | C_PLUS | C_MINUS)) {
						shf_putc(*s, shf);
						s++;
						precision--;
						nwritten++;
					} else if (*s == '0') {
						shf_putc(*s, shf);
						s++;
						nwritten++;
						if (--precision &&
						    ksh_eq(*s, 'X', 'x')) {
							shf_putc(*s, shf);
							s++;
							precision--;
							nwritten++;
						}
					}
					c = '0';
				} else
					c = flags & FL_ZERO ? '0' : ' ';
				nwritten += field;
				while (field--)
					shf_putc(c, shf);
				field = 0;
			} else
				c = ' ';
		} else
			field = 0;

		nwritten += precision;
		precision = utf_skipcols(s, precision, &tmp) - s;
		while (precision--)
			shf_putc(*s++, shf);

		nwritten += field;
		while (field--)
			shf_putc(c, shf);
	}

	return (shf_error(shf) ? -1 : nwritten);
}

#if defined(MKSH_SMALL) && !defined(MKSH_SMALL_BUT_FAST)
int
shf_getc(struct shf *shf)
{
	return (shf_getc_i(shf));
}

int
shf_putc(int c, struct shf *shf)
{
	return (shf_putc_i(c, shf));
}
#endif

#ifdef DEBUG
const char *
cstrerror(int errnum)
{
#undef strerror
	return (strerror(errnum));
#define strerror dontuse_strerror /* poisoned */
}
#elif !HAVE_STRERROR

#if HAVE_SYS_ERRLIST
#if !HAVE_SYS_ERRLIST_DECL
extern const int sys_nerr;
extern const char * const sys_errlist[];
#endif
#endif

const char *
cstrerror(int errnum)
{
	/* "Unknown error: " + sign + rough estimate + NUL */
	static char errbuf[15 + 1 + (8 * sizeof(int) + 2) / 3 + 1];

#if HAVE_SYS_ERRLIST
	if (errnum > 0 && errnum < sys_nerr && sys_errlist[errnum])
		return (sys_errlist[errnum]);
#endif

	switch (errnum) {
	case 0:
		return ("Undefined error: 0");
	case EPERM:
		return ("Operation not permitted");
	case ENOENT:
		return ("No such file or directory");
#ifdef ESRCH
	case ESRCH:
		return ("No such process");
#endif
#ifdef E2BIG
	case E2BIG:
		return ("Argument list too long");
#endif
	case ENOEXEC:
		return ("Exec format error");
	case EBADF:
		return ("Bad file descriptor");
#ifdef ENOMEM
	case ENOMEM:
		return ("Cannot allocate memory");
#endif
	case EACCES:
		return ("Permission denied");
	case EEXIST:
		return ("File exists");
	case ENOTDIR:
		return ("Not a directory");
#ifdef EINVAL
	case EINVAL:
		return ("Invalid argument");
#endif
#ifdef ELOOP
	case ELOOP:
		return ("Too many levels of symbolic links");
#endif
	default:
		shf_snprintf(errbuf, sizeof(errbuf),
		    "Unknown error: %d", errnum);
		return (errbuf);
	}
}
#endif

/* fast character classes */
const uint32_t tpl_ctypes[128] = {
	/* 0x00 */
	CiNUL,		CiCNTRL,	CiCNTRL,	CiCNTRL,
	CiCNTRL,	CiCNTRL,	CiCNTRL,	CiCNTRL,
	CiCNTRL,	CiTAB,		CiNL,		CiSPX,
	CiSPX,		CiCR,		CiCNTRL,	CiCNTRL,
	/* 0x10 */
	CiCNTRL,	CiCNTRL,	CiCNTRL,	CiCNTRL,
	CiCNTRL,	CiCNTRL,	CiCNTRL,	CiCNTRL,
	CiCNTRL,	CiCNTRL,	CiCNTRL,	CiCNTRL,
	CiCNTRL,	CiCNTRL,	CiCNTRL,	CiCNTRL,
	/* 0x20 */
	CiSP,		CiALIAS | CiVAR1,	CiQC,	CiHASH,
	CiSS,		CiPERCT,	CiQCL,		CiQC,
	CiQCL,		CiQCL,		CiQCX | CiVAR1,	CiPLUS,
	CiALIAS,	CiMINUS,	CiALIAS,	CiQCM,
	/* 0x30 */
	CiOCTAL,	CiOCTAL,	CiOCTAL,	CiOCTAL,
	CiOCTAL,	CiOCTAL,	CiOCTAL,	CiOCTAL,
	CiDIGIT,	CiDIGIT,	CiCOLON,	CiQCL,
	CiANGLE,	CiEQUAL,	CiANGLE,	CiQUEST,
	/* 0x40 */
	CiALIAS | CiVAR1,	CiUPPER | CiHEXLT,
	CiUPPER | CiHEXLT,	CiUPPER | CiHEXLT,
	CiUPPER | CiHEXLT,	CiUPPER | CiHEXLT,
	CiUPPER | CiHEXLT,	CiUPPER,
	CiUPPER,	CiUPPER,	CiUPPER,	CiUPPER,
	CiUPPER,	CiUPPER,	CiUPPER,	CiUPPER,
	/* 0x50 */
	CiUPPER,	CiUPPER,	CiUPPER,	CiUPPER,
	CiUPPER,	CiUPPER,	CiUPPER,	CiUPPER,
	CiUPPER,	CiUPPER,	CiUPPER,	CiQCX | CiBRACK,
	CiQCX,		CiBRACK,	CiQCM,		CiUNDER,
	/* 0x60 */
	CiGRAVE,		CiLOWER | CiHEXLT,
	CiLOWER | CiHEXLT,	CiLOWER | CiHEXLT,
	CiLOWER | CiHEXLT,	CiLOWER | CiHEXLT,
	CiLOWER | CiHEXLT,	CiLOWER,
	CiLOWER,	CiLOWER,	CiLOWER,	CiLOWER,
	CiLOWER,	CiLOWER,	CiLOWER,	CiLOWER,
	/* 0x70 */
	CiLOWER,	CiLOWER,	CiLOWER,	CiLOWER,
	CiLOWER,	CiLOWER,	CiLOWER,	CiLOWER,
	CiLOWER,	CiLOWER,	CiLOWER,	CiCURLY,
	CiQCL,		CiCURLY,	CiQCM,		CiCNTRL
};

void
set_ifs(const char *s)
{
#if defined(MKSH_EBCDIC) || defined(MKSH_FAUX_EBCDIC)
	int i = 256;

	memset(ksh_ctypes, 0, sizeof(ksh_ctypes));
	while (i--)
		if (ebcdic_map[i] < 0x80U)
			ksh_ctypes[i] = tpl_ctypes[ebcdic_map[i]];
#else
	memcpy(ksh_ctypes, tpl_ctypes, sizeof(tpl_ctypes));
	memset((char *)ksh_ctypes + sizeof(tpl_ctypes), '\0',
	    sizeof(ksh_ctypes) - sizeof(tpl_ctypes));
#endif
	ifs0 = *s;
	while (*s)
		ksh_ctypes[ord(*s++)] |= CiIFS;
}

#if defined(MKSH_EBCDIC) || defined(MKSH_FAUX_EBCDIC)
#include <locale.h>

/*
 * Many headaches with EBCDIC:
 * 1. There are numerous EBCDIC variants, and it is not feasible for us
 *    to support them all. But we can support the EBCDIC code pages that
 *    contain all (most?) of the characters in ASCII, and these
 *    usually tend to agree on the code points assigned to the ASCII
 *    subset. If you need a representative example, look at EBCDIC 1047,
 *    which is first among equals in the IBM MVS development
 *    environment: https://en.wikipedia.org/wiki/EBCDIC_1047
 *    Unfortunately, the square brackets are not consistently mapped,
 *    and for certain reasons, we need an unambiguous bijective
 *    mapping between EBCDIC and "extended ASCII".
 * 2. Character ranges that are contiguous in ASCII, like the letters
 *    in [A-Z], are broken up into segments (i.e. [A-IJ-RS-Z]), so we
 *    can't implement e.g. islower() as { return c >= 'a' && c <= 'z'; }
 *    because it will also return true for a handful of extraneous
 *    characters (like the plus-minus sign at 0x8F in EBCDIC 1047, a
 *    little after 'i'). But at least '_' is not one of these.
 * 3. The normal [0-9A-Za-z] characters are at codepoints beyond 0x80.
 *    Not only do they require all 8 bits instead of 7, if chars are
 *    signed, they will have negative integer values! Something like
 *    (c - 'A') could actually become (c + 63)! Use the ord() macro to
 *    ensure you're getting a value in [0, 255].
 * 4. '\n' is actually NL (0x15, U+0085) instead of LF (0x25, U+000A).
 *    EBCDIC has a proper newline character instead of "emulating" one
 *    with line feeds, although this is mapped to LF for our purposes.
 * 5. Note that it is possible to compile programs in ASCII mode on IBM
 *    mainframe systems, using the -qascii option to the XL C compiler.
 *    We can determine the build mode by looking at __CHARSET_LIB:
 *    0 == EBCDIC, 1 == ASCII
 */

void
ebcdic_init(void)
{
	int i = 256;
	unsigned char t;
	bool mapcache[256];

	while (i--)
		ebcdic_rtt_toascii[i] = i;
	memset(ebcdic_rtt_fromascii, 0xFF, sizeof(ebcdic_rtt_fromascii));
	setlocale(LC_ALL, "");
#ifdef MKSH_EBCDIC
	if (__etoa_l(ebcdic_rtt_toascii, 256) != 256) {
		write(2, "mksh: could not map EBCDIC to ASCII\n", 36);
		exit(255);
	}
#endif

	memset(mapcache, 0, sizeof(mapcache));
	i = 256;
	while (i--) {
		t = ebcdic_rtt_toascii[i];
		/* ensure unique round-trip capable mapping */
		if (mapcache[t]) {
			write(2, "mksh: duplicate EBCDIC to ASCII mapping\n", 40);
			exit(255);
		}
		/*
		 * since there are 256 input octets, this also ensures
		 * the other mapping direction is completely filled
		 */
		mapcache[t] = true;
		/* fill the complete round-trip map */
		ebcdic_rtt_fromascii[t] = i;
		/*
		 * Only use the converted value if it's in the range
		 * [0x00; 0x7F], which I checked; the "extended ASCII"
		 * characters can be any encoding, not just Latin1,
		 * and the C1 control characters other than NEL are
		 * hopeless, but we map EBCDIC NEL to ASCII LF so we
		 * cannot even use C1 NEL.
		 * If ever we map to Unicode, bump the table width to
		 * an unsigned int, and or the raw unconverted EBCDIC
		 * values with 0x01000000 instead.
		 */
		if (t < 0x80U)
			ebcdic_map[i] = (unsigned short)ord(t);
		else
			ebcdic_map[i] = (unsigned short)(0x100U | ord(i));
	}
	if (ebcdic_rtt_toascii[0] || ebcdic_rtt_fromascii[0] || ebcdic_map[0]) {
		write(2, "mksh: NUL not at position 0\n", 28);
		exit(255);
	}
}
#endif