[nmh] / uip / mhbuildsbr.c

/* mhbuildsbr.c -- routines to expand/translate MIME composition files
 *
 * This code is Copyright (c) 2002, by the authors of nmh.  See the
 * COPYRIGHT file in the root directory of the nmh distribution for
 * complete copyright information.
 */

/*
 * This code was originally part of mhn.c.  I split it into
 * a separate program (mhbuild.c) and then later split it
 * again (mhbuildsbr.c).  But the code still has some of
 * the mhn.c code in it.  This program needs additional
 * streamlining and removal of unneeded code.
 */

#include "h/mh.h"
#include "sbr/m_gmprot.h"
#include "sbr/m_getfld.h"
#include "sbr/concat.h"
#include "sbr/r1bindex.h"
#include "sbr/encode_rfc2047.h"
#include "sbr/copyip.h"
#include "sbr/cpydata.h"
#include "sbr/trimcpy.h"
#include "sbr/uprf.h"
#include "sbr/check_charset.h"
#include "sbr/getcpy.h"
#include "sbr/m_convert.h"
#include "sbr/getfolder.h"
#include "sbr/folder_read.h"
#include "sbr/folder_free.h"
#include "sbr/context_find.h"
#include "sbr/brkstring.h"
#include "sbr/pidstatus.h"
#include "sbr/path.h"
#include "sbr/error.h"
#include <fcntl.h>
#include "h/mts.h"
#include "h/tws.h"
#include "h/fmt_scan.h"
#include "h/mime.h"
#include "h/mhparse.h"
#include "h/done.h"
#include "h/utils.h"
#include "mhmisc.h"
#include "sbr/m_mktemp.h"
#include "sbr/message_id.h"
#include "sbr/mime_type.h"
#include "mhfree.h"
#include "mhshowsbr.h"

#ifdef HAVE_SYS_TIME_H
# include <sys/time.h>
#endif
#include <time.h>


extern int debugsw;

extern bool listsw;
extern bool rfc934sw;
extern bool contentidsw;

static char prefix[] = "----- =_aaaaaaaaaa";

struct attach_list {
    char *filename;
    struct attach_list *next;
};

typedef struct convert_list {
    char *type;
    char *filename;
    char *argstring;
    struct convert_list *next;
} convert_list;


/*
 * static prototypes
 */
static int init_decoded_content (CT, const char *);
static void setup_attach_content(CT, char *);
static void set_disposition (CT);
static void set_charset (CT, int);
static void expand_pseudoheaders (CT, struct multipart *, const char *,
                                  const convert_list *);
static void expand_pseudoheader (CT, CT *, struct multipart *, const char *,
                                 const char *, const char *);
static char *fgetstr (char *, int, FILE *);
static int user_content (FILE *, char *, CT *, const char *infilename);
static void set_id (CT, int);
static int compose_content (CT, int);
static int scan_content (CT, size_t);
static int build_headers (CT, int);
static int extract_headers (CT, char *, FILE **);


static unsigned char directives_stack[32];
static unsigned int directives_index;

static int
do_direct(void)
{
    return directives_stack[directives_index];
}

static void
directive_onoff(int onoff)
{
    if (directives_index >= sizeof(directives_stack) - 1) {
        fprintf(stderr, "mhbuild: #on/off overflow, continuing\n");
        return;
    }
    directives_stack[++directives_index] = onoff;
}

static void
directive_init(int onoff)
{
    directives_index = 0;
    directives_stack[0] = onoff;
}

static void
directive_pop(void)
{
    if (directives_index > 0)
        directives_index--;
    else
        fprintf(stderr, "mhbuild: #pop underflow, continuing\n");
}

/*
 * Main routine for translating composition file
 * into valid MIME message.  It translates the draft
 * into a content structure (actually a tree of content
 * structures).  This message then can be manipulated
 * in various ways, including being output via
 * output_message().
 */

CT
build_mime (char *infile, int autobuild, int dist, int directives,
            int header_encoding, size_t maxunencoded, int verbose)
{
    int compnum, state;
    char buf[NMH_BUFSIZ], name[NAMESZ];
    char *cp, *np, *vp;
    struct multipart *m;
    struct part **pp;
    CT ct;
    FILE *in;
    HF hp;
    m_getfld_state_t gstate;
    struct attach_list *attach_head = NULL, *attach_tail = NULL, *at_entry;
    convert_list *convert_head = NULL, *convert_tail = NULL, *convert;

    directive_init(directives);

    umask (~m_gmprot ());

    /* open the composition draft */
    if ((in = fopen (infile, "r")) == NULL)
        adios (infile, "unable to open for reading");

    /*
     * Allocate space for primary (outside) content
     */
    NEW0(ct);

    /*
     * Allocate structure for handling decoded content
     * for this part.  We don't really need this, but
     * allocate it to remain consistent.
     */
    init_decoded_content (ct, infile);

    /*
     * Parse some of the header fields in the composition
     * draft into the linked list of header fields for
     * the new MIME message.
     */
    gstate = m_getfld_state_init(in);
    m_getfld_track_filepos2(&gstate);
    for (compnum = 1;;) {
        int bufsz = sizeof buf;
        switch (state = m_getfld2(&gstate, name, buf, &bufsz)) {
        case FLD:
        case FLDPLUS:
            compnum++;

            /*
             * If we are running with autobuild set, then silently
             * exit if we find a MIME-Version header.  For any other MIME
             * header, return an error.
             *
             * RFC 2045, Section 9 says that any valid MIME header should
             * start with "Content-", so we will match on that rather than
             * enumerate all current MIME headers.
             *
             * Because the headers could be in any order, just check for
             * MIME-Version here; check for Content-* later.
             */

            if (strcasecmp (name, VRSN_FIELD) == 0) {
                if (autobuild) {
                    fclose(in);
                    free (ct);
                    return NULL;
                }
                die("draft shouldn't contain %s: field", name);
            }

            /* get copies of the buffers */
            np = mh_xstrdup(name);
            vp = mh_xstrdup(buf);

            /* if necessary, get rest of field */
            while (state == FLDPLUS) {
                bufsz = sizeof buf;
                state = m_getfld2(&gstate, name, buf, &bufsz);
                vp = add (buf, vp);     /* add to previous value */
            }

            /*
             * Now add the header data to the list, unless it's an attach
             * header; in that case, add it to our attach list
             */

            if (strcasecmp(ATTACH_FIELD, np) == 0  ||
                strcasecmp(ATTACH_FIELD_ALT, np) == 0) {
                struct attach_list *entry;
                char *s = vp, *e = vp + strlen(vp) - 1;
                free(np);

                /*
                 * Make sure we can find the start of this filename.
                 * If it's blank, we skip completely.  Otherwise, strip
                 * off any leading spaces and trailing newlines.
                 */

                while (isspace((unsigned char) *s))
                    s++;

                while (e > s && *e == '\n')
                    *e-- = '\0';

                if (*s == '\0') {
                    free(vp);
                    goto finish_field;
                }

                NEW(entry);
                entry->filename = mh_xstrdup(s);
                entry->next = NULL;
                free(vp);

                if (attach_tail) {
                    attach_tail->next = entry;
                    attach_tail = entry;
                } else {
                    attach_head = attach_tail = entry;
                }
            } else if (strncasecmp(MHBUILD_FILE_PSEUDOHEADER, np,
                                   LEN(MHBUILD_FILE_PSEUDOHEADER)) == 0) {
                /* E.g.,
                 * Nmh-mhbuild-file-text/calendar: /home/user/Mail/inbox/9
                 */
                char *type = np + LEN(MHBUILD_FILE_PSEUDOHEADER);
                char *filename = vp;

                /* vp should begin with a space because m_getfld2()
                   includes the space after the colon in buf. */
                while (isspace((unsigned char) *filename)) { ++filename; }
                /* Trim trailing newline and any other whitespace. */
                rtrim (filename);

                for (convert = convert_head; convert; convert = convert->next) {
                    if (strcasecmp (convert->type, type) == 0) { break; }
                }
                if (convert) {
                    if (convert->filename  &&
                        strcasecmp (convert->filename, filename)) {
                        die("Multiple %s headers with different files"
                               " not allowed", type);
                    } else {
                        convert->filename = mh_xstrdup(filename);
                    }
                } else {
                    NEW0(convert);
                    convert->filename = mh_xstrdup(filename);
                    convert->type = mh_xstrdup(type);

                    if (convert_tail) {
                        convert_tail->next = convert;
                    } else {
                        convert_head = convert;
                    }
                    convert_tail = convert;
                }

                free (vp);
                free (np);
            } else if (strncasecmp(MHBUILD_ARGS_PSEUDOHEADER, np,
                                   LEN(MHBUILD_ARGS_PSEUDOHEADER)) == 0) {
                /* E.g.,
                 * Nmh-mhbuild-args-text/calendar: -reply accept
                 */
                char *type = np + LEN(MHBUILD_ARGS_PSEUDOHEADER);
                char *argstring = vp;

                /* vp should begin with a space because m_getfld2()
                   includes the space after the colon in buf. */
                while (isspace((unsigned char) *argstring)) { ++argstring; }
                /* Trim trailing newline and any other whitespace. */
                rtrim (argstring);

                for (convert = convert_head; convert; convert = convert->next) {
                    if (strcasecmp (convert->type, type) == 0) { break; }
                }
                if (convert) {
                    if (convert->argstring  &&
                        strcasecmp (convert->argstring, argstring)) {
                        die("Multiple %s headers with different "
                               "argstrings not allowed", type);
                    } else {
                        convert->argstring = mh_xstrdup(argstring);
                    }
                } else {
                    NEW0(convert);
                    convert->type = mh_xstrdup(type);
                    convert->argstring = mh_xstrdup(argstring);

                    if (convert_tail) {
                        convert_tail->next = convert;
                    } else {
                        convert_head = convert;
                    }
                    convert_tail = convert;
                }

                free (vp);
                free (np);
            } else {
                add_header (ct, np, vp);
            }

finish_field:
            /* if this wasn't the last header field, then continue */
            continue;

        case BODY:
            fseek (in, (long) (-strlen (buf)), SEEK_CUR);
            break;
        case FILEEOF:
            break;

        case LENERR:
        case FMTERR:
            die("message format error in component #%d", compnum);

        default:
            die("getfld() returned %d", state);
        }
        break;
    }
    m_getfld_state_destroy (&gstate);

    /*
     * If we see any Content-* headers at this point, it is an error.
     */

    for (hp = ct->c_first_hf; hp != NULL; hp = hp->next) {
        if (uprf (hp->name, "Content-")) {
            die ("draft shouldn't contain %s: field", hp->name);
        }
    }

    if (header_encoding != CE_8BIT) {
        /*
         * Iterate through the list of headers and call the function to MIME-ify
         * them if required.
         */

        for (hp = ct->c_first_hf; hp != NULL; hp = hp->next) {
            if (encode_rfc2047(hp->name, &hp->value, header_encoding, NULL)) {
                die("Unable to encode header \"%s\"", hp->name);
            }
        }
    }

    /*
     * Now add the MIME-Version header field
     * to the list of header fields.
     */

    if (! dist) {
        np = mh_xstrdup(VRSN_FIELD);
        vp = concat (" ", VRSN_VALUE, "\n", NULL);
        add_header (ct, np, vp);
    }

    /*
     * We initially assume we will find multiple contents in the
     * draft.  So create a multipart/mixed content to hold everything.
     * We can remove this later, if it is not needed.
     */
    if (get_ctinfo ("multipart/mixed", ct, 0) == NOTOK)
        done (1);
    ct->c_type = CT_MULTIPART;
    ct->c_subtype = MULTI_MIXED;

    NEW0(m);
    ct->c_ctparams = (void *) m;
    pp = &m->mp_parts;

    /*
     * read and parse the composition file
     * and the directives it contains.
     */
    while (fgetstr (buf, sizeof(buf) - 1, in)) {
        struct part *part;
        CT p;

        if (user_content (in, buf, &p, infile) == DONE) {
            inform("ignoring spurious #end, continuing...");
            continue;
        }
        if (!p)
            continue;

        NEW0(part);
        *pp = part;
        pp = &part->mp_next;
        part->mp_part = p;
    }

    /*
     * Add any Attach headers to the list of MIME parts at the end of the
     * message.
     */

    for (at_entry = attach_head; at_entry; ) {
        struct attach_list *at_prev = at_entry;
        struct part *part;
        CT p;

        if (access(at_entry->filename, R_OK) != 0) {
            adios("reading", "Unable to open %s for", at_entry->filename);
        }

        NEW0(p);
        init_decoded_content(p, infile);

        /*
         * Initialize our content structure based on the filename,
         * and fill in all of the relevant fields.  Also place MIME
         * parameters in the attributes array.
         */

        setup_attach_content(p, at_entry->filename);

        NEW0(part);
        *pp = part;
        pp = &part->mp_next;
        part->mp_part = p;

        at_entry = at_entry->next;
        free(at_prev->filename);
        free(at_prev);
    }

    /*
     * Handle the mhbuild pseudoheaders, which deal with specific
     * content types.
     */
    if (convert_head) {
        CT *ctp;
        convert_list *next;

        set_done(freects_done);

        /* In case there are multiple calls that land here, prevent leak. */
        for (ctp = cts; ctp && *ctp; ++ctp) { free_content (*ctp); }
        free (cts);

        /* Extract the type part (as a CT) from filename. */
        cts = mh_xcalloc(2, sizeof *cts);
        if (! (cts[0] = parse_mime (convert_head->filename))) {
            die("failed to parse %s", convert_head->filename);
        }

        expand_pseudoheaders (cts[0], m, infile, convert_head);

        /* Free the convert list. */
        for (convert = convert_head; convert; convert = next) {
            next = convert->next;
            free (convert->type);
            free (convert->filename);
            free (convert->argstring);
            free (convert);
        }
        convert_head = NULL;
    }

    /*
     * To allow for empty message bodies, if we've found NO content at all
     * yet cook up an empty text/plain part.
     */

    if (!m->mp_parts) {
        CT p;
        struct part *part;
        struct text *t;

        NEW0(p);
        init_decoded_content(p, infile);

        if (get_ctinfo ("text/plain", p, 0) == NOTOK)
            done (1);

        p->c_type = CT_TEXT;
        p->c_subtype = TEXT_PLAIN;
        p->c_encoding = CE_7BIT;
        /*
         * Sigh.  ce_file contains the "decoded" contents of this part.
         * So this seems like the best option available since we're going
         * to call scan_content() on this.
         */
        p->c_cefile.ce_file = mh_xstrdup("/dev/null");
        p->c_begin = ftell(in);
        p->c_end = ftell(in);

        NEW0(t);
        t->tx_charset = CHARSET_SPECIFIED;
        p->c_ctparams = t;

        NEW0(part);
        *pp = part;
        part->mp_part = p;
    }

    /*
     * close the composition draft since
     * it's not needed any longer.
     */
    fclose (in);

    /*
     * If only one content was found, then remove and
     * free the outer multipart content.
     */
    if (!m->mp_parts->mp_next) {
        CT p;

        p = m->mp_parts->mp_part;
        m->mp_parts->mp_part = NULL;

        /* move header fields */
        p->c_first_hf = ct->c_first_hf;
        p->c_last_hf = ct->c_last_hf;
        ct->c_first_hf = NULL;
        ct->c_last_hf = NULL;

        free_content (ct);
        ct = p;
    } else {
        set_id (ct, 1);
    }

    /*
     * Fill out, or expand directives.  Parse and execute
     * commands specified by profile composition strings.
     */
    compose_content (ct, verbose);

    if ((cp = strchr(prefix, 'a')) == NULL)
        die("internal error(4)");

    /*
     * If using EAI, force 8-bit charset.
     */
    if (header_encoding == CE_8BIT) {
        set_charset (ct, 1);
    }

    /*
     * Scan the contents.  Choose a transfer encoding, and
     * check if prefix for multipart boundary clashes with
     * any of the contents.
     */
    while (scan_content (ct, maxunencoded) == NOTOK) {
        if (*cp < 'z') {
            (*cp)++;
        } else {
            if (*++cp == 0)
                die("giving up trying to find a unique delimiter string");
            (*cp)++;
        }
    }

    /* Build the rest of the header field structures */
    if (! dist)
        build_headers (ct, header_encoding);

    return ct;
}


/*
 * Set up structures for placing unencoded
 * content when building parts.
 */

static int
init_decoded_content (CT ct, const char *filename)
{
    ct->c_ceopenfnx  = open7Bit;        /* since unencoded */
    ct->c_ceclosefnx = close_encoding;
    ct->c_cesizefnx  = NULL;            /* since unencoded */
    ct->c_encoding = CE_7BIT;           /* Seems like a reasonable default */
    ct->c_file = mh_xstrdup(FENDNULL(filename));

    return OK;
}


static char *
fgetstr (char *s, int n, FILE *stream)
{
    char *cp, *ep;

    ep = s + n;
    while(1) {
        for (cp = s; cp < ep;) {
            int len;

            if (!fgets (cp, n, stream))
                return cp == s ? NULL : s; /* "\\\nEOF" ignored. */

            if (! do_direct()  ||  (cp == s && *cp != '#'))
                return s; /* Plaintext line. */

            len = strlen(cp);
            if (len <= 1)
                break; /* Can't contain "\\\n". */
            cp += len - 1; /* Just before NUL. */
            if (*cp-- != '\n' || *cp != '\\')
                break;
            *cp = '\0'; /* Erase the trailing "\\\n". */
            n -= (len - 2);
        }

        if (strcmp(s, "#on\n") == 0) {
            directive_onoff(1);
        } else if (strcmp(s, "#off\n") == 0) {
            directive_onoff(0);
        } else if (strcmp(s, "#pop\n") == 0) {
            directive_pop();
        } else {
            return s;
        }
    }
}


/*
 * Parse the composition draft for text and directives.
 * Do initial setup of Content structure.
 */

static int
user_content (FILE *in, char *buf, CT *ctp, const char *infilename)
{
    int extrnal, vrsn;
    char *cp, **ap;
    char buffer[NMH_BUFSIZ];
    struct multipart *m;
    struct part **pp;
    struct stat st;
    struct str2init *s2i;
    CI ci;
    CT ct;
    CE ce;

    if (buf[0] == '\n' || (do_direct() && strcmp (buf, "#\n") == 0)) {
        *ctp = NULL;
        return OK;
    }

    /* allocate basic Content structure */
    NEW0(ct);
    *ctp = ct;

    /* allocate basic structure for handling decoded content */
    init_decoded_content (ct, infilename);
    ce = &ct->c_cefile;

    ci = &ct->c_ctinfo;
    set_id (ct, 0);

    /*
     * Handle inline text.  Check if line
     * is one of the following forms:
     *
     * 1) doesn't begin with '#'        (implicit directive)
     * 2) begins with "##"              (implicit directive)
     * 3) begins with "#<"
     */
    if (!do_direct() || buf[0] != '#' || buf[1] == '#' || buf[1] == '<') {
        int headers;
        bool inlineD;
        long pos;
        char content[BUFSIZ];
        FILE *out;
        char *cp;

        if ((cp = m_mktemp2(NULL, invo_name, NULL, &out)) == NULL) {
            adios("mhbuildsbr", "unable to create temporary file in %s",
                  get_temp_dir());
        }

        /* use a temp file to collect the plain text lines */
        ce->ce_file = mh_xstrdup(cp);
        ce->ce_unlink = 1;

        if (do_direct() && (buf[0] == '#' && buf[1] == '<')) {
            strncpy (content, buf + 2, sizeof(content));
            inlineD = true;
            goto rock_and_roll;
        }
        inlineD = false;

        /* the directive is implicit */
        strncpy (content, "text/plain", sizeof(content));
        headers = 0;
        strncpy (buffer, (!do_direct() || buf[0] != '#') ? buf : buf + 1, sizeof(buffer));
        for (;;) {
            int i;

            if (headers >= 0 && do_direct() && uprf (buffer, DESCR_FIELD)
                && buffer[i = LEN(DESCR_FIELD)] == ':') {
                headers = 1;

again_descr:
                ct->c_descr = add (buffer + i + 1, ct->c_descr);
                if (!fgetstr (buffer, sizeof(buffer) - 1, in))
                    die("end-of-file after %s: field in plaintext", DESCR_FIELD);
                switch (buffer[0]) {
                case ' ':
                case '\t':
                    i = -1;
                    goto again_descr;

                case '#':
                    die("#-directive after %s: field in plaintext", DESCR_FIELD);
                    /* NOTREACHED */

                default:
                    break;
                }
            }

            if (headers >= 0 && do_direct() && uprf (buffer, DISPO_FIELD)
                && buffer[i = LEN(DISPO_FIELD)] == ':') {
                headers = 1;

again_dispo:
                ct->c_dispo = add (buffer + i + 1, ct->c_dispo);
                if (!fgetstr (buffer, sizeof(buffer) - 1, in))
                    die("end-of-file after %s: field in plaintext", DISPO_FIELD);
                switch (buffer[0]) {
                case ' ':
                case '\t':
                    i = -1;
                    goto again_dispo;

                case '#':
                    die("#-directive after %s: field in plaintext", DISPO_FIELD);
                    /* NOTREACHED */

                default:
                    break;
                }
            }

            if (headers != 1 || buffer[0] != '\n')
                fputs (buffer, out);

rock_and_roll:
            headers = -1;
            pos = ftell (in);
            if ((cp = fgetstr (buffer, sizeof(buffer) - 1, in)) == NULL)
                break;
            if (do_direct() && buffer[0] == '#') {
                char *bp;

                if (buffer[1] != '#')
                    break;
                for (cp = (bp = buffer) + 1; *cp; cp++)
                    *bp++ = *cp;
                *bp = '\0';
            }
        }

        if (listsw)
            ct->c_end = ftell (out);
        fclose (out);

        /* parse content type */
        if (get_ctinfo (content, ct, inlineD) == NOTOK)
            done (1);

        for (s2i = str2cts; s2i->si_key; s2i++)
            if (!strcasecmp (ci->ci_type, s2i->si_key))
                break;
        if (!s2i->si_key && !uprf (ci->ci_type, "X-"))
            s2i++;

        /*
         * check type specified (possibly implicitly)
         */
        switch (ct->c_type = s2i->si_val) {
        case CT_MESSAGE:
            if (!strcasecmp (ci->ci_subtype, "rfc822")) {
                ct->c_encoding = CE_7BIT;
                goto call_init;
            }
            /* FALLTHRU */
        case CT_MULTIPART:
            die("it doesn't make sense to define an in-line %s content",
                   ct->c_type == CT_MESSAGE ? "message" : "multipart");
            /* NOTREACHED */

        default:
call_init:
            if ((ct->c_ctinitfnx = s2i->si_init))
                (*ct->c_ctinitfnx) (ct);
            break;
        }

        if (cp)
            fseek (in, pos, SEEK_SET);
        return OK;
    }

    /*
     * If we've reached this point, the next line
     * must be some type of explicit directive.
     */

    /* check if directive is external-type */
    extrnal = (buf[1] == '@');

    /* parse directive */
    if (get_ctinfo (buf + (extrnal ? 2 : 1), ct, 1) == NOTOK)
        done (1);

    /* check directive against the list of MIME types */
    for (s2i = str2cts; s2i->si_key; s2i++)
        if (!strcasecmp (ci->ci_type, s2i->si_key))
            break;

    /*
     * Check if the directive specified a valid type.
     * This will happen if it was one of the following forms:
     *
     *    #type/subtype  (or)
     *    #@type/subtype
     */
    if (s2i->si_key) {
        if (!ci->ci_subtype)
            die("missing subtype in \"#%s\"", ci->ci_type);

        switch (ct->c_type = s2i->si_val) {
        case CT_MULTIPART:
            die("use \"#begin ... #end\" instead of \"#%s/%s\"",
                   ci->ci_type, ci->ci_subtype);
            /* NOTREACHED */

        case CT_MESSAGE:
            if (!strcasecmp (ci->ci_subtype, "partial"))
                die("sorry, \"#%s/%s\" isn't supported",
                       ci->ci_type, ci->ci_subtype);
            if (!strcasecmp (ci->ci_subtype, "external-body"))
                die("use \"#@type/subtype ... [] ...\" instead of \"#%s/%s\"",
                       ci->ci_type, ci->ci_subtype);
use_forw:
            die(                   "use \"#forw [+folder] [msgs]\" instead of \"#%s/%s\"",
                   ci->ci_type, ci->ci_subtype);
            /* NOTREACHED */

        default:
            if ((ct->c_ctinitfnx = s2i->si_init))
                (*ct->c_ctinitfnx) (ct);
            break;
        }

        /*
         * #@type/subtype (external types directive)
         */
        if (extrnal) {
            struct exbody *e;
            CT p;

            if (!ci->ci_magic)
                die("need external information for \"#@%s/%s\"",
                       ci->ci_type, ci->ci_subtype);
            p = ct;

            snprintf (buffer, sizeof(buffer), "message/external-body; %s", ci->ci_magic);
            free (ci->ci_magic);
            ci->ci_magic = NULL;

            /*
             * Since we are using the current Content structure to
             * hold information about the type of the external
             * reference, we need to create another Content structure
             * for the message/external-body to wrap it in.
             */
            NEW0(ct);
            init_decoded_content(ct, infilename);
            *ctp = ct;
            if (get_ctinfo (buffer, ct, 0) == NOTOK)
                done (1);
            ct->c_type = CT_MESSAGE;
            ct->c_subtype = MESSAGE_EXTERNAL;

            NEW0(e);
            ct->c_ctparams = (void *) e;

            e->eb_parent = ct;
            e->eb_content = p;
            p->c_ctexbody = e;

            if (params_external (ct, 1) == NOTOK)
                done (1);

            return OK;
        }

        /* Handle [file] argument */
        if (ci->ci_magic) {
            /* check if specifies command to execute */
            if (*ci->ci_magic == '|' || *ci->ci_magic == '!') {
                for (cp = ci->ci_magic + 1; isspace ((unsigned char) *cp); cp++)
                    continue;
                if (!*cp)
                    die("empty pipe command for #%s directive", ci->ci_type);
                cp = mh_xstrdup(cp);
                free (ci->ci_magic);
                ci->ci_magic = cp;
            } else {
                /* record filename of decoded contents */
                ce->ce_file = ci->ci_magic;
                if (access (ce->ce_file, R_OK) == NOTOK)
                    adios ("reading", "unable to access %s for", ce->ce_file);
                if (listsw && stat (ce->ce_file, &st) != NOTOK)
                    ct->c_end = (long) st.st_size;
                ci->ci_magic = NULL;
            }
            return OK;
        }

        /*
         * No [file] argument, so check profile for
         * method to compose content.
         */
        cp = context_find_by_type ("compose", ci->ci_type, ci->ci_subtype);
        if (cp == NULL) {
            content_error (NULL, ct, "don't know how to compose content");
            done (1);
        }
        ci->ci_magic = mh_xstrdup(cp);
        return OK;
    }

    if (extrnal)
        die("external definition not allowed for \"#%s\"", ci->ci_type);

    /*
     * Message directive
     * #forw [+folder] [msgs]
     */
    if (!strcasecmp (ci->ci_type, "forw")) {
        int msgnum;
        char *folder, *arguments[MAXARGS];
        struct msgs *mp;

        if (ci->ci_magic) {
            ap = brkstring (ci->ci_magic, " ", "\n");
            copyip (ap, arguments, MAXARGS);
        } else {
            arguments[0] = "cur";
            arguments[1] = NULL;
        }
        folder = NULL;

        /* search the arguments for a folder name */
        for (ap = arguments; *ap; ap++) {
            cp = *ap;
            if (*cp == '+' || *cp == '@') {
                if (folder)
                    die("only one folder per #forw directive");
                folder = pluspath (cp);
            }
        }

        /* else, use the current folder */
        if (!folder)
            folder = mh_xstrdup(getfolder(1));

        if (!(mp = folder_read (folder, 0)))
            die("unable to read folder %s", folder);
        for (ap = arguments; *ap; ap++) {
            cp = *ap;
            if (*cp != '+' && *cp != '@')
                if (!m_convert (mp, cp))
                    done (1);
        }
        free (folder);
        free_ctinfo (ct);

        /*
         * If there is more than one message to include, make this
         * a content of type "multipart/digest" and insert each message
         * as a subpart.  If there is only one message, then make this
         * a content of type "message/rfc822".
         */
        if (mp->numsel > 1) {
            /* we are forwarding multiple messages */
            if (get_ctinfo ("multipart/digest", ct, 0) == NOTOK)
                done (1);
            ct->c_type = CT_MULTIPART;
            ct->c_subtype = MULTI_DIGEST;

            NEW0(m);
            ct->c_ctparams = (void *) m;
            pp = &m->mp_parts;

            for (msgnum = mp->lowsel; msgnum <= mp->hghsel; msgnum++) {
                if (is_selected(mp, msgnum)) {
                    struct part *part;
                    CT p;
                    CE pe;

                    NEW0(p);
                    init_decoded_content (p, infilename);
                    pe = &p->c_cefile;
                    if (get_ctinfo ("message/rfc822", p, 0) == NOTOK)
                        done (1);
                    p->c_type = CT_MESSAGE;
                    p->c_subtype = MESSAGE_RFC822;

                    snprintf (buffer, sizeof(buffer), "%s/%d", mp->foldpath, msgnum);
                    pe->ce_file = mh_xstrdup(buffer);
                    if (listsw && stat (pe->ce_file, &st) != NOTOK)
                        p->c_end = (long) st.st_size;

                    NEW0(part);
                    *pp = part;
                    pp = &part->mp_next;
                    part->mp_part = p;
                }
            }
        } else {
            /* we are forwarding one message */
            if (get_ctinfo ("message/rfc822", ct, 0) == NOTOK)
                done (1);
            ct->c_type = CT_MESSAGE;
            ct->c_subtype = MESSAGE_RFC822;

            msgnum = mp->lowsel;
            snprintf (buffer, sizeof(buffer), "%s/%d", mp->foldpath, msgnum);
            ce->ce_file = mh_xstrdup(buffer);
            if (listsw && stat (ce->ce_file, &st) != NOTOK)
                ct->c_end = (long) st.st_size;
        }

        folder_free (mp);       /* free folder/message structure */
        return OK;
    }

    /*
     * #end
     */
    if (!strcasecmp (ci->ci_type, "end")) {
        free_content (ct);
        *ctp = NULL;
        return DONE;
    }

    /*
     * #begin [ alternative | parallel ]
     */
    if (!strcasecmp (ci->ci_type, "begin")) {
        if (!ci->ci_magic) {
            vrsn = MULTI_MIXED;
            cp = SubMultiPart[vrsn - 1].kv_key;
        } else if (!strcasecmp (ci->ci_magic, "alternative")) {
            vrsn = MULTI_ALTERNATE;
            cp = SubMultiPart[vrsn - 1].kv_key;
        } else if (!strcasecmp (ci->ci_magic, "parallel")) {
            vrsn = MULTI_PARALLEL;
            cp = SubMultiPart[vrsn - 1].kv_key;
        } else if (uprf (ci->ci_magic, "digest")) {
            goto use_forw;
        } else {
            vrsn = MULTI_UNKNOWN;
            cp = ci->ci_magic;
        }

        free_ctinfo (ct);
        snprintf (buffer, sizeof(buffer), "multipart/%s", cp);
        if (get_ctinfo (buffer, ct, 0) == NOTOK)
            done (1);
        ct->c_type = CT_MULTIPART;
        ct->c_subtype = vrsn;

        NEW0(m);
        ct->c_ctparams = (void *) m;

        pp = &m->mp_parts;
        while (fgetstr (buffer, sizeof(buffer) - 1, in)) {
            struct part *part;
            CT p;

            if (user_content (in, buffer, &p, infilename) == DONE) {
                if (!m->mp_parts)
                    die("empty \"#begin ... #end\" sequence");
                return OK;
            }
            if (!p)
                continue;

            NEW0(part);
            *pp = part;
            pp = &part->mp_next;
            part->mp_part = p;
        }
        inform("premature end-of-file, missing #end, continuing...");
        return OK;
    }

    /*
     * Unknown directive
     */
    die("unknown directive \"#%s\"", ci->ci_type);
    return NOTOK;       /* NOT REACHED */
}


static void
set_id (CT ct, int top)
{
    char contentid[BUFSIZ];
    static int partno;
    static time_t clock = 0;
    static char *msgfmt;

    if (clock == 0) {
        time (&clock);
        snprintf (contentid, sizeof(contentid), "%s\n", message_id (clock, 1));
        partno = 0;
        msgfmt = mh_xstrdup(contentid);
    }
    snprintf (contentid, sizeof(contentid), msgfmt, top ? 0 : ++partno);
    ct->c_id = mh_xstrdup(contentid);
}


/*
 * Fill out, or expand the various contents in the composition
 * draft.  Read-in any necessary files.  Parse and execute any
 * commands specified by profile composition strings.
 */

static int
compose_content (CT ct, int verbose)
{
    CE ce = &ct->c_cefile;

    switch (ct->c_type) {
    case CT_MULTIPART:
    {
        int partnum;
        char *pp;
        char partnam[BUFSIZ];
        struct multipart *m = (struct multipart *) ct->c_ctparams;
        struct part *part;

        if (ct->c_partno) {
            snprintf (partnam, sizeof(partnam), "%s.", ct->c_partno);
            pp = partnam + strlen (partnam);
        } else {
            pp = partnam;
        }

        /* first, we call compose_content on all the subparts */
        for (part = m->mp_parts, partnum = 1; part; part = part->mp_next, partnum++) {
            CT p = part->mp_part;

            sprintf (pp, "%d", partnum);
            p->c_partno = mh_xstrdup(partnam);
            if (compose_content (p, verbose) == NOTOK)
                return NOTOK;
        }

        /*
         * If the -rfc934mode switch is given, then check all
         * the subparts of a multipart/digest.  If they are all
         * message/rfc822, then mark this content and all
         * subparts with the rfc934 compatibility mode flag.
         */
        if (rfc934sw && ct->c_subtype == MULTI_DIGEST) {
            bool is934 = true;

            for (part = m->mp_parts; part; part = part->mp_next) {
                CT p = part->mp_part;

                if (p->c_subtype != MESSAGE_RFC822) {
                    is934 = false;
                    break;
                }
            }
            ct->c_rfc934 = is934;
            for (part = m->mp_parts; part; part = part->mp_next) {
                CT p = part->mp_part;

                if ((p->c_rfc934 = is934))
                    p->c_end++;
            }
        }

        if (listsw) {
            ct->c_end = (partnum = strlen (prefix) + 2) + 2;
            if (ct->c_rfc934)
                ct->c_end++;

            for (part = m->mp_parts; part; part = part->mp_next)
                ct->c_end += part->mp_part->c_end + partnum;
        }
    }
    break;

    case CT_MESSAGE:
        /* Nothing to do for type message */
        break;

    /*
     * Discrete types (text/application/audio/image/video)
     */
    default:
        if (!ce->ce_file) {
            pid_t child_id;
            bool xstdout;
            int len, buflen;
            char *bp, *cp;
            char *vec[4], buffer[BUFSIZ];
            FILE *out;
            CI ci = &ct->c_ctinfo;
            char *tfile = NULL;

            if (!(cp = ci->ci_magic))
                die("internal error(5)");

            if ((tfile = m_mktemp2(NULL, invo_name, NULL, NULL)) == NULL) {
                adios("mhbuildsbr", "unable to create temporary file in %s",
                      get_temp_dir());
            }
            ce->ce_file = mh_xstrdup(tfile);
            ce->ce_unlink = 1;

            xstdout = false;

            /* Get buffer ready to go */
            bp = buffer;
            bp[0] = '\0';
            buflen = sizeof(buffer);

            /*
             * Parse composition string into buffer
             */
            for ( ; *cp; cp++) {
                if (*cp == '%') {
                    switch (*++cp) {
                    case 'a':
                    {
                        /* insert parameters from directive */
                        char *s = "";
                        PM pm;

                        for (pm = ci->ci_first_pm; pm; pm = pm->pm_next) {
                            snprintf (bp, buflen, "%s%s=\"%s\"", s,
                                    pm->pm_name, get_param_value(pm, '?'));
                            len = strlen (bp);
                            bp += len;
                            buflen -= len;
                            s = " ";
                        }
                    }
                    break;

                    case 'F':
                        /* %f, and stdout is not-redirected */
                        xstdout = true;
                        /* FALLTHRU */

                    case 'f':
                        /*
                         * insert temporary filename where
                         * content should be written
                         */
                        snprintf (bp, buflen, "%s", ce->ce_file);
                        break;

                    case 's':
                        /* insert content subtype */
                        strncpy (bp, ci->ci_subtype, buflen);
                        break;

                    case '%':
                        /* insert character % */
                        goto raw;

                    default:
                        *bp++ = *--cp;
                        *bp = '\0';
                        buflen--;
                        continue;
                    }
                    len = strlen (bp);
                    bp += len;
                    buflen -= len;
                } else {
raw:
                *bp++ = *cp;
                *bp = '\0';
                buflen--;
                }
            }

            if (verbose)
                printf ("composing content %s/%s from command\n\t%s\n",
                        ci->ci_type, ci->ci_subtype, buffer);

            fflush (stdout);    /* not sure if need for -noverbose */

            vec[0] = "/bin/sh";
            vec[1] = "-c";
            vec[2] = buffer;
            vec[3] = NULL;

            if ((out = fopen (ce->ce_file, "w")) == NULL)
                adios (ce->ce_file, "unable to open for writing");

            child_id = fork();
            switch (child_id) {
            case NOTOK:
                adios ("fork", "unable to fork");
                /* NOTREACHED */

            case OK:
                if (!xstdout)
                    dup2 (fileno (out), 1);
                close (fileno (out));
                execvp ("/bin/sh", vec);
                fprintf (stderr, "unable to exec ");
                perror ("/bin/sh");
                _exit(1);
                /* NOTREACHED */

            default:
                fclose (out);
                if (pidXwait(child_id, NULL))
                    done (1);
                break;
            }
        }

        /* Check size of file */
        if (listsw && ct->c_end == 0L) {
            struct stat st;

            if (stat (ce->ce_file, &st) != NOTOK)
                ct->c_end = (long) st.st_size;
        }
        break;
    }

    return OK;
}


/*
 * Scan the content.
 *
 *    1) choose a transfer encoding.
 *    2) check for clashes with multipart boundary string.
 *    3) for text content, figure out which character set is being used.
 *
 * If there is a clash with one of the contents and the multipart boundary,
 * this function will exit with NOTOK.  This will cause the scanning process
 * to be repeated with a different multipart boundary.  It is possible
 * (although highly unlikely) that this scan will be repeated multiple times.
 */

static int
scan_content (CT ct, size_t maxunencoded)
{
    int prefix_len;
    bool check8bit = false, contains8bit = false;  /* check if contains 8bit data */
    bool checknul = false, containsnul = false;  /* check if contains NULs */
    bool checklinelen = false, linelen = false;  /* check for long lines */
    bool checkllinelen = false; /* check for extra-long lines */
    bool checkboundary = false, boundaryclash = false; /* check if clashes with multipart boundary   */
    bool checklinespace = false, linespace = false;  /* check if any line ends with space          */
    char *cp = NULL;
    char *bufp = NULL;
    size_t buflen;
    ssize_t gotlen;
    struct text *t = NULL;
    FILE *in = NULL;
    CE ce = &ct->c_cefile;

    /*
     * handle multipart by scanning all subparts
     * and then checking their encoding.
     */
    if (ct->c_type == CT_MULTIPART) {
        struct multipart *m = (struct multipart *) ct->c_ctparams;
        struct part *part;

        /* initially mark the domain of enclosing multipart as 7bit */
        ct->c_encoding = CE_7BIT;

        for (part = m->mp_parts; part; part = part->mp_next) {
            CT p = part->mp_part;

            if (scan_content (p, maxunencoded) == NOTOK)        /* choose encoding for subpart */
                return NOTOK;

            /* if necessary, enlarge encoding for enclosing multipart */
            if (p->c_encoding == CE_BINARY)
                ct->c_encoding = CE_BINARY;
            if (p->c_encoding == CE_8BIT && ct->c_encoding != CE_BINARY)
                ct->c_encoding = CE_8BIT;
        }

        return OK;
    }

    /*
     * Decide what to check while scanning this content.  Note that
     * for text content we always check for 8bit characters if the
     * charset is unspecified, because that controls whether or not the
     * character set is us-ascii or retrieved from the locale.  And
     * we check even if the charset is specified, to allow setting
     * the proper Content-Transfer-Encoding.
     */

    if (ct->c_type == CT_TEXT) {
        t = (struct text *) ct->c_ctparams;
        if (t->tx_charset == CHARSET_UNSPECIFIED) {
            checknul = true;
        }
        check8bit = true;
    }

    switch (ct->c_reqencoding) {
    case CE_8BIT:
        checkllinelen = true;
        checkboundary = true;
        break;
    case CE_QUOTED:
        checkboundary = true;
        break;
    case CE_BASE64:
        break;
    case CE_UNKNOWN:
        /* Use the default rules based on content-type */
        switch (ct->c_type) {
        case CT_TEXT:
            checkboundary = true;
            checklinelen = true;
            if (ct->c_subtype == TEXT_PLAIN) {
                checklinespace = false;
            } else {
                checklinespace = true;
            }
        break;

        case CT_APPLICATION:
            check8bit = true;
            checknul = true;
            checklinelen = true;
            checklinespace = true;
            checkboundary = true;
        break;

        case CT_MESSAGE:
            checklinelen = false;
            checklinespace = false;

            /* don't check anything for message/external */
            if (ct->c_subtype == MESSAGE_EXTERNAL) {
                checkboundary = false;
                check8bit = false;
            } else {
                checkboundary = true;
                check8bit = true;
            }
            break;

        case CT_AUDIO:
        case CT_IMAGE:
        case CT_VIDEO:
            /*
             * Don't check anything for these types,
             * since we are forcing use of base64, unless
             * the content-type was specified by a mhbuild directive.
             */
            check8bit = false;
            checklinelen = false;
            checklinespace = false;
            checkboundary = false;
            break;
        }
    }

    /*
     * Scan the unencoded content
     */
    if (check8bit || checklinelen || checklinespace || checkboundary ||
        checkllinelen || checknul) {
        if ((in = fopen (ce->ce_file, "r")) == NULL)
            adios (ce->ce_file, "unable to open for reading");
        prefix_len = strlen (prefix);

        while ((gotlen = getline(&bufp, &buflen, in)) != -1) {
            /*
             * Check for 8bit and NUL data.
             */
            for (cp = bufp; (check8bit || checknul) &&
                                        cp < bufp + gotlen; cp++) {
                if (!isascii ((unsigned char) *cp)) {
                    contains8bit = true;
                    check8bit = false;  /* no need to keep checking */
                }
                if (!*cp) {
                    containsnul = true;
                    checknul = false;   /* no need to keep checking */
                }
            }

            /*
             * Check line length.
             */
            if (checklinelen && ((size_t)gotlen > maxunencoded + 1)) {
                linelen = true;
                checklinelen = false;   /* no need to keep checking */
            }

            /*
             * RFC 5322 specifies that a message cannot contain a line
             * greater than 998 characters (excluding the CRLF).  If we
             * get one of those lines and linelen is NOT set, then abort.
             */

            if (checkllinelen && !linelen &&
                                        (gotlen > MAXLONGLINE + 1)) {
                die("Line in content exceeds maximum line limit (%d)",
                      MAXLONGLINE);
            }

            /*
             * Check if line ends with a space.
             */
            if (checklinespace && (cp = bufp + gotlen - 2) > bufp &&
                        isspace ((unsigned char) *cp)) {
                linespace = true;
                checklinespace = false; /* no need to keep checking */
            }

            /*
             * Check if content contains a line that clashes
             * with our standard boundary for multipart messages.
             */
            if (checkboundary && bufp[0] == '-' && bufp[1] == '-') {
                for (cp = bufp + gotlen - 1; cp >= bufp; cp--)
                    if (!isspace ((unsigned char) *cp))
                        break;
                *++cp = '\0';
                if (!strncmp(bufp + 2, prefix, prefix_len) &&
                            isdigit((unsigned char) bufp[2 + prefix_len])) {
                    boundaryclash = true;
                    checkboundary = false;      /* no need to keep checking */
                }
            }
        }
        fclose (in);
        free(bufp);
    }

    /*
     * If the content is text and didn't specify a character set,
     * we need to figure out which one was used.
     */
    set_charset (ct, contains8bit);

    /*
     * Decide which transfer encoding to use.
     */

    if (ct->c_reqencoding != CE_UNKNOWN)
        ct->c_encoding = ct->c_reqencoding;
    else {
        int wants_q_p = (containsnul || linelen || linespace);

        switch (ct->c_type) {
        case CT_TEXT:
            if (wants_q_p)
                 ct->c_encoding = CE_QUOTED;
            else if (contains8bit)
                 ct->c_encoding = CE_8BIT;
            else
                 ct->c_encoding = CE_7BIT;

            break;

        case CT_APPLICATION:
            /* For application type, use base64, except when postscript */
            if (wants_q_p || contains8bit) {
                if (ct->c_subtype == APPLICATION_POSTSCRIPT)
                    ct->c_encoding = CE_QUOTED;  /* historical */
                else
                    ct->c_encoding = CE_BASE64;
            } else {
                ct->c_encoding = CE_7BIT;
            }
            break;

        case CT_MESSAGE:
            ct->c_encoding = contains8bit ? CE_8BIT : CE_7BIT;
            break;

        case CT_AUDIO:
        case CT_IMAGE:
        case CT_VIDEO:
            /* For audio, image, and video contents, just use base64 */
            ct->c_encoding = CE_BASE64;
            break;
        }
    }

    return boundaryclash ? NOTOK : OK;
}


/*
 * Scan the content structures, and build header
 * fields that will need to be output into the
 * message.
 */

static int
build_headers (CT ct, int header_encoding)
{
    int cc, mailbody, extbody, len;
    char *np, *vp, buffer[BUFSIZ];
    CI ci = &ct->c_ctinfo;

    /*
     * If message is type multipart, then add the multipart
     * boundary to the list of attribute/value pairs.
     */
    if (ct->c_type == CT_MULTIPART) {
        static int level = 0;   /* store nesting level */

        snprintf (buffer, sizeof(buffer), "%s%d", prefix, level++);
        add_param(&ci->ci_first_pm, &ci->ci_last_pm, "boundary", buffer, 0);
    }

    /*
     * Skip the output of Content-Type, parameters, content
     * description and disposition, and Content-ID if the
     * content is of type "message" and the rfc934 compatibility
     * flag is set (which means we are inside multipart/digest
     * and the switch -rfc934mode was given).
     */
    if (ct->c_type == CT_MESSAGE && ct->c_rfc934)
        goto skip_headers;

    /*
     * output the content type and subtype
     */
    np = mh_xstrdup(TYPE_FIELD);
    vp = concat (" ", ci->ci_type, "/", ci->ci_subtype, NULL);

    /* keep track of length of line */
    len = LEN(TYPE_FIELD) + strlen (ci->ci_type)
                + strlen (ci->ci_subtype) + 3;

    extbody = ct->c_type == CT_MESSAGE && ct->c_subtype == MESSAGE_EXTERNAL;
    mailbody = extbody && ((struct exbody *) ct->c_ctparams)->eb_body;

    /*
     * Append the attribute/value pairs to
     * the end of the Content-Type line.
     */

    if (ci->ci_first_pm) {
        char *s = output_params(len, ci->ci_first_pm, &len, mailbody);

        if (!s)
            die("Internal error: failed outputting Content-Type "
                "parameters");

        vp = add (s, vp);
        free(s);
    }

    /*
     * Append any RFC-822 comment to the end of
     * the Content-Type line.
     */
    if (ci->ci_comment) {
        snprintf (buffer, sizeof(buffer), "(%s)", ci->ci_comment);
        if (len + 1 + (cc = 2 + strlen (ci->ci_comment)) >= CPERLIN) {
            vp = add ("\n\t", vp);
            len = 8;
        } else {
            vp = add (" ", vp);
            len++;
        }
        vp = add (buffer, vp);
        len += cc;
    }
    vp = add ("\n", vp);
    add_header (ct, np, vp);

    /*
     * output the Content-ID, unless disabled by -nocontentid.  Note that
     * RFC 2045 always requires a Content-ID header for message/external-body
     * entities.
     */
    if ((contentidsw || ct->c_ctexbody) && ct->c_id) {
        np = mh_xstrdup(ID_FIELD);
        vp = concat (" ", ct->c_id, NULL);
        add_header (ct, np, vp);
    }
    /*
     * output the Content-Description
     */
    if (ct->c_descr) {
        np = mh_xstrdup(DESCR_FIELD);
        vp = concat (" ", ct->c_descr, NULL);
        if (header_encoding != CE_8BIT) {
            if (encode_rfc2047(DESCR_FIELD, &vp, header_encoding, NULL)) {
                die("Unable to encode %s header", DESCR_FIELD);
            }
        }
        add_header (ct, np, vp);
    }

    /*
     * output the Content-Disposition.  If it's NULL but c_dispo_type is
     * set, then we need to build it.
     */
    if (ct->c_dispo) {
        np = mh_xstrdup(DISPO_FIELD);
        vp = concat (" ", ct->c_dispo, NULL);
        add_header (ct, np, vp);
    } else if (ct->c_dispo_type) {
        vp = concat (" ", ct->c_dispo_type, NULL);
        len = LEN(DISPO_FIELD) + strlen(vp) + 1;
        np = output_params(len, ct->c_dispo_first, NULL, 0);
        vp = add(np, vp);
        vp = add("\n", vp);
        free(np);
        add_header (ct, mh_xstrdup(DISPO_FIELD), vp);
    }

skip_headers:
    /*
     * If this is the internal content structure for a
     * "message/external", then we are done with the
     * headers (since it has no body).
     */
    if (ct->c_ctexbody)
        return OK;

    /*
     * output the Content-Transfer-Encoding
     * If using EAI and message body is 7-bit, force 8-bit C-T-E.
     */
    if (header_encoding == CE_8BIT  &&  ct->c_encoding == CE_7BIT) {
        ct->c_encoding = CE_8BIT;
    }

    switch (ct->c_encoding) {
    case CE_7BIT:
        /* Nothing to output */
        break;

    case CE_8BIT:
        np = mh_xstrdup(ENCODING_FIELD);
        vp = concat (" ", "8bit", "\n", NULL);
        add_header (ct, np, vp);
        break;

    case CE_QUOTED:
        if (ct->c_type == CT_MESSAGE || ct->c_type == CT_MULTIPART)
            die("internal error, invalid encoding");

        np = mh_xstrdup(ENCODING_FIELD);
        vp = concat (" ", "quoted-printable", "\n", NULL);
        add_header (ct, np, vp);
        break;

    case CE_BASE64:
        if (ct->c_type == CT_MESSAGE || ct->c_type == CT_MULTIPART)
            die("internal error, invalid encoding");

        np = mh_xstrdup(ENCODING_FIELD);
        vp = concat (" ", "base64", "\n", NULL);
        add_header (ct, np, vp);
        break;

    case CE_BINARY:
        if (ct->c_type == CT_MESSAGE)
            die("internal error, invalid encoding");

        np = mh_xstrdup(ENCODING_FIELD);
        vp = concat (" ", "binary", "\n", NULL);
        add_header (ct, np, vp);
        break;

    default:
        die("unknown transfer encoding in content");
        break;
    }

    /*
     * Additional content specific header processing
     */
    switch (ct->c_type) {
    case CT_MULTIPART:
    {
        struct multipart *m;
        struct part *part;

        m = (struct multipart *) ct->c_ctparams;
        for (part = m->mp_parts; part; part = part->mp_next) {
            CT p;

            p = part->mp_part;
            build_headers (p, header_encoding);
        }
    }
        break;

    case CT_MESSAGE:
        if (ct->c_subtype == MESSAGE_EXTERNAL) {
            struct exbody *e;

            e = (struct exbody *) ct->c_ctparams;
            build_headers (e->eb_content, header_encoding);
        }
        break;

    default:
        /* Nothing to do */
        break;
    }

    return OK;
}


/*
 * Set things up for the content structure for file "filename" that
 * we want to attach
 */

static void
setup_attach_content(CT ct, char *filename)
{
    char *type, *simplename = r1bindex(filename, '/');
    struct str2init *s2i;
    PM pm;

    if (! (type = mime_type(filename))) {
        die("Unable to determine MIME type of \"%s\"", filename);
    }

    /*
     * Parse the Content-Type.  get_ctinfo() parses MIME parameters, but
     * since we're just feeding it a MIME type we have to add those ourselves.
     * Map that to a valid content-type label and call any initialization
     * function.
     */

    if (get_ctinfo(type, ct, 0) == NOTOK)
        done(1);

    free(type);

    for (s2i = str2cts; s2i->si_key; s2i++)
        if (strcasecmp(ct->c_ctinfo.ci_type, s2i->si_key) == 0)
            break;
    if (!s2i->si_key && !uprf(ct->c_ctinfo.ci_type, "X-"))
        s2i++;

    /*
     * Make sure the type isn't incompatible with what we can handle
     */

    switch (ct->c_type = s2i->si_val) {
    case CT_MULTIPART:
        die("multipart types must be specified by mhbuild directives");
        /* NOTREACHED */

    case CT_MESSAGE:
        if (strcasecmp(ct->c_ctinfo.ci_subtype, "partial") == 0)
            die("Sorry, %s/%s isn't supported", ct->c_ctinfo.ci_type,
                ct->c_ctinfo.ci_subtype);
        if (strcasecmp(ct->c_ctinfo.ci_subtype, "external-body") == 0)
            die("external-body messages must be specified "
                "by mhbuild directives");
        /* FALLTHRU */

    default:
        /*
         * This sets the subtype, if it's significant
         */
        if ((ct->c_ctinitfnx = s2i->si_init))
            (*ct->c_ctinitfnx)(ct);
        break;
    }

    /*
     * Feed in a few attributes; specifically, the name attribute, the
     * content-description, and the content-disposition.
     */

    for (pm = ct->c_ctinfo.ci_first_pm; pm; pm = pm->pm_next) {
        if (strcasecmp(pm->pm_name, "name") == 0) {
            free(pm->pm_value);
            pm->pm_value = mh_xstrdup(simplename);
            break;
        }
    }

    if (pm == NULL)
        add_param(&ct->c_ctinfo.ci_first_pm, &ct->c_ctinfo.ci_last_pm,
                  "name", simplename, 0);

    ct->c_descr = mh_xstrdup(simplename);
    ct->c_descr = add("\n", ct->c_descr);
    ct->c_cefile.ce_file = mh_xstrdup(filename);

    set_disposition (ct);

    add_param(&ct->c_dispo_first, &ct->c_dispo_last, "filename", simplename, 0);
}

/*
 * If disposition type hasn't already been set in ct:
 * Look for mhbuild-disposition-<type>/<subtype> entry
 * that specifies Content-Disposition type.  Only
 * 'attachment' and 'inline' are allowed.  Default to
 * 'attachment'.
 */
void
set_disposition (CT ct)
{
    if (ct->c_dispo_type == NULL) {
        char *cp = context_find_by_type ("disposition", ct->c_ctinfo.ci_type,
                                         ct->c_ctinfo.ci_subtype);

        if (cp  &&  strcasecmp (cp, "attachment")  &&
            strcasecmp (cp, "inline")) {
            inform("configuration problem: %s-disposition-%s%s%s specifies "
                "'%s' but only 'attachment' and 'inline' are allowed, "
                "continuing...", invo_name,
                ct->c_ctinfo.ci_type,
                ct->c_ctinfo.ci_subtype ? "/" : "",
                FENDNULL(ct->c_ctinfo.ci_subtype),
                cp);
        }

        if (!cp)
            cp = "attachment";
        ct->c_dispo_type = mh_xstrdup(cp);
    }
}

/*
 * Set text content charset if it was unspecified.  contains8bit
 * selections:
 * 0: content does not contain 8-bit characters
 * 1: content contains 8-bit characters
 * -1: ignore content and use user's locale to determine charset
 */
void
set_charset (CT ct, int contains8bit)
{
    if (ct->c_type == CT_TEXT) {
        struct text *t;

        if (ct->c_ctparams == NULL) {
            NEW0(t);
            ct->c_ctparams = t;
            t->tx_charset = CHARSET_UNSPECIFIED;
        } else {
            t = (struct text *) ct->c_ctparams;
        }

        if (t->tx_charset == CHARSET_UNSPECIFIED) {
            CI ci = &ct->c_ctinfo;
            char *eightbitcharset = write_charset_8bit();
            char *charset = contains8bit ? eightbitcharset : "us-ascii";

            if (contains8bit == 1  &&
                strcasecmp (eightbitcharset, "US-ASCII") == 0) {
                die("Text content contains 8 bit characters, but "
                       "character set is US-ASCII");
            }

            add_param (&ci->ci_first_pm, &ci->ci_last_pm, "charset", charset,
                       0);

            t->tx_charset = CHARSET_SPECIFIED;
        }
    }
}


/*
 * Look at all of the replied-to message parts and expand any that
 * are matched by a pseudoheader.  Except don't descend into
 * message parts.
 */
void
expand_pseudoheaders (CT ct, struct multipart *m, const char *infile,
                      const convert_list *convert_head)
{
    /* text_plain_ct is used to concatenate all of the text/plain
       replies into one part, instead of having each one in a separate
       part. */
    CT text_plain_ct = NULL;

    switch (ct->c_type) {
    case CT_MULTIPART: {
        struct multipart *mp = (struct multipart *) ct->c_ctparams;
        struct part *part;

        if (ct->c_subtype == MULTI_ALTERNATE) {
            bool matched = false;

            /* The parts are in descending priority order (defined by
               RFC 2046 Sec. 5.1.4) because they were reversed by
               parse_mime ().  So, stop looking for matches with
               immediate subparts after the first match of an
               alternative. */
            for (part = mp->mp_parts; ! matched && part; part = part->mp_next) {
                char *type_subtype =
                    concat (part->mp_part->c_ctinfo.ci_type, "/",
                            part->mp_part->c_ctinfo.ci_subtype, NULL);

                if (part->mp_part->c_type == CT_MULTIPART) {
                    expand_pseudoheaders (part->mp_part, m, infile,
                                          convert_head);
                } else {
                    const convert_list *c;

                    for (c = convert_head; c; c = c->next) {
                        if (! strcasecmp (type_subtype, c->type)) {
                            expand_pseudoheader (part->mp_part, &text_plain_ct,
                                                 m, infile,
                                                 c->type, c->argstring);
                            matched = true;
                            break;
                        }
                    }
                }
                free (type_subtype);
            }
        } else {
            for (part = mp->mp_parts; part; part = part->mp_next) {
                expand_pseudoheaders (part->mp_part, m, infile, convert_head);
            }
        }
        break;
    }

    default: {
        char *type_subtype =
            concat (ct->c_ctinfo.ci_type, "/", ct->c_ctinfo.ci_subtype,
                    NULL);
        const convert_list *c;

        for (c = convert_head; c; c = c->next) {
            if (! strcasecmp (type_subtype, c->type)) {
                expand_pseudoheader (ct, &text_plain_ct, m, infile, c->type,
                                     c->argstring);
                break;
            }
        }
        free (type_subtype);
        break;
    }
    }
}


/*
 * Expand a single pseudoheader.  It's for the specified type.
 */
void
expand_pseudoheader (CT ct, CT *text_plain_ct, struct multipart *m,
                     const char *infile, const char *type,
                     const char *argstring)
{
    char *reply_file;
    FILE *reply_fp = NULL;
    char *convert, *type_p, *subtype_p;
    char *convert_command;
    char *charset = NULL;
    char *cp;
    struct str2init *s2i;
    CT reply_ct;
    struct part *part;
    int status;

    type_p = getcpy (type);
    if ((subtype_p = strchr (type_p, '/'))) {
        *subtype_p++ = '\0';
        convert = context_find_by_type ("convert", type_p, subtype_p);
    } else {
        free (type_p);
        type_p = concat ("mhbuild-convert-", type, NULL);
        convert = context_find (type_p);
    }
    free (type_p);

    if (! (convert)) {
        /* No mhbuild-convert- entry in mhn.defaults or profile for type. */
        return;
    }
    /* reply_file is used to pass the output of the convert. */
    reply_file = getcpy (m_mktemp2 (NULL, invo_name, NULL, NULL));
    convert_command =
        concat (convert, " ", FENDNULL(argstring), " >", reply_file, NULL);

    /* Convert here . . . */
    ct->c_storeproc = mh_xstrdup(convert_command);
    ct->c_umask = ~m_gmprot ();

    if ((status = show_content_aux (ct, 0, convert_command, NULL, NULL)) !=
        OK) {
        inform("store of %s content failed, continuing...", type);
    }
    free (convert_command);

    /* Fill out the the new ct, reply_ct. */
    NEW0(reply_ct);
    init_decoded_content (reply_ct, infile);

    if (extract_headers (reply_ct, reply_file, &reply_fp) == NOTOK) {
        inform("failed to extract headers from convert output in %s, "
            "continuing...", reply_file);
        free(reply_file);
        return;
    }

    /* This sets reply_ct->c_ctparams, and reply_ct->c_termproc if the
       charset can't be handled natively. */
    for (s2i = str2cts; s2i->si_key; s2i++) {
        if (strcasecmp(reply_ct->c_ctinfo.ci_type, s2i->si_key) == 0) {
            break;
        }
    }

    if ((reply_ct->c_ctinitfnx = s2i->si_init)) {
        (*reply_ct->c_ctinitfnx)(reply_ct);
    }

    if ((cp = get_param (reply_ct->c_ctinfo.ci_first_pm, "charset", '?', 1))) {
        /* The reply Content-Type had the charset. */
        charset = cp;
    } else {
        set_charset (reply_ct, -1);
        charset = get_param (reply_ct->c_ctinfo.ci_first_pm, "charset", '?', 1);
    }

    /* Concatenate text/plain parts. */
    if (reply_ct->c_type == CT_TEXT  &&  reply_ct->c_subtype == TEXT_PLAIN) {
        if (! *text_plain_ct  &&  m->mp_parts  &&  m->mp_parts->mp_part  &&
            m->mp_parts->mp_part->c_type == CT_TEXT  &&
            m->mp_parts->mp_part->c_subtype == TEXT_PLAIN) {
            *text_plain_ct = m->mp_parts->mp_part;
            /* Make sure that the charset is set in the text/plain part. */
            set_charset (*text_plain_ct, -1);
        }

        if (*text_plain_ct) {
            /* Only concatenate if the charsets are identical. */
            char *text_plain_ct_charset =
                get_param ((*text_plain_ct)->c_ctinfo.ci_first_pm, "charset",
                           '?', 1);

            if (strcasecmp (text_plain_ct_charset, charset) == 0) {
                /* Append this text/plain reply to the first one.
                   If there's a problem anywhere along the way,
                   instead attach it is a separate part. */
                int text_plain_reply =
                    open ((*text_plain_ct)->c_cefile.ce_file,
                          O_WRONLY | O_APPEND);
                int addl_reply = open (reply_file, O_RDONLY);

                if (text_plain_reply != NOTOK  &&  addl_reply != NOTOK) {
                    /* Insert blank line before each addl part. */
                    /* It would be nice not to do this for the first one. */
                    if (write (text_plain_reply, "\n", 1) == 1) {
                        /* Copy the text from the new reply and
                           then free its Content struct. */
                        cpydata (addl_reply, text_plain_reply,
                                 (*text_plain_ct)->c_cefile.ce_file,
                                 reply_file);
                        if (close (text_plain_reply) == OK  &&
                            close (addl_reply) == OK) {
                            /* If appended text needed 8-bit but first text didn't,
                               propagate the 8-bit indication. */
                            if ((*text_plain_ct)->c_reqencoding == CE_7BIT  &&
                                reply_ct->c_reqencoding == CE_8BIT) {
                                (*text_plain_ct)->c_reqencoding = CE_8BIT;
                            }

                            if (reply_fp) { fclose (reply_fp); }
                            free (reply_file);
                            free_content (reply_ct);
                            return;
                        }
                    }
                }
            }
        } else {
            *text_plain_ct = reply_ct;
        }
    }

    reply_ct->c_cefile.ce_file = reply_file;
    reply_ct->c_cefile.ce_fp = reply_fp;
    reply_ct->c_cefile.ce_unlink = 1;

    /* Attach the new part to the parent multipart/mixed, "m". */
    NEW0(part);
    part->mp_part = reply_ct;
    if (m->mp_parts) {
        struct part *p;

        for (p = m->mp_parts; p && p->mp_next; p = p->mp_next) { continue; }
        p->mp_next = part;
    } else {
        m->mp_parts = part;
    }
}


/* Extract any Content-Type header from beginning of convert output. */
int
extract_headers (CT ct, char *reply_file, FILE **reply_fp)
{
    char *buffer = NULL, *cp, *end_of_header;
    bool found_header = false;
    struct stat statbuf;

    /* Read the convert reply from the file to memory. */
    if (stat (reply_file, &statbuf) == NOTOK) {
        admonish (reply_file, "failed to stat");
        goto failed_to_extract_ct;
    }

    buffer = mh_xmalloc (statbuf.st_size + 1);

    if ((*reply_fp = fopen (reply_file, "r+")) == NULL  ||
        fread (buffer, 1, (size_t) statbuf.st_size, *reply_fp) <
            (size_t) statbuf.st_size) {
        admonish (reply_file, "failed to read");
        goto failed_to_extract_ct;
    }
    buffer[statbuf.st_size] = '\0';

    /* Look for a header in the convert reply. */
    if (strncasecmp (buffer, TYPE_FIELD, LEN(TYPE_FIELD)) == 0  &&
        buffer[LEN(TYPE_FIELD)] == ':') {
        if ((end_of_header = strstr (buffer, "\r\n\r\n"))) {
            end_of_header += 2;
            found_header = true;
        } else if ((end_of_header = strstr (buffer, "\n\n"))) {
            ++end_of_header;
            found_header = true;
        }
    }

    if (found_header) {
        CT tmp_ct;
        char *tmp_file;
        FILE *tmp_f;
        size_t n, written;

        /* Truncate buffer to just the C-T. */
        *end_of_header = '\0';
        n = strlen (buffer);

        if (get_ctinfo (buffer + 14, ct, 0) != OK) {
            inform("unable to get content info for reply, continuing...");
            goto failed_to_extract_ct;
        }

        /* Hack.  Use parse_mime() to detect the type/subtype of the
           reply, which we'll use below. */
        tmp_file = getcpy (m_mktemp2 (NULL, invo_name, NULL, NULL));
        tmp_f = fopen(tmp_file, "w");
        if (!tmp_f)
            goto failed_to_extract_ct;
        written = fwrite(buffer, 1, n, tmp_f);
        fclose(tmp_f);
        if (written != n)
            goto failed_to_extract_ct;

        tmp_ct = parse_mime (tmp_file);
        if (tmp_ct) {
            /* The type and subtype were detected from the reply
               using parse_mime() above. */
            ct->c_type = tmp_ct->c_type;
            ct->c_subtype = tmp_ct->c_subtype;
            free_content (tmp_ct);
        }

        free (tmp_file);

        /* Rewrite the content without the header. */
        cp = end_of_header + 1;
        rewind (*reply_fp);

        if (fwrite (cp, 1, statbuf.st_size - (cp - buffer), *reply_fp) <
            (size_t) (statbuf.st_size - (cp - buffer))) {
            admonish (reply_file, "failed to write");
            goto failed_to_extract_ct;
        }

        if (ftruncate (fileno (*reply_fp), statbuf.st_size - (cp - buffer)) !=
            0) {
            advise (reply_file, "ftruncate");
            goto failed_to_extract_ct;
        }
    } else {
        /* No header section, assume the reply is text/plain. */
        ct->c_type = CT_TEXT;
        ct->c_subtype = TEXT_PLAIN;
        if (get_ctinfo ("text/plain", ct, 0) == NOTOK) {
            /* This never should fail, but just in case. */
            die("unable to get content info for reply");
        }
    }

    /* free_encoding() will close reply_fp, which is passed through
       ct->c_cefile.ce_fp. */
    free (buffer);
    return OK;

failed_to_extract_ct:
    if (*reply_fp) { fclose (*reply_fp); }
    free (buffer);
    return NOTOK;
}