/*
** k2ocr.c       k2pdfopt OCR functions
**
** Copyright (C) 2018  http://willus.com
**
** This program is free software: you can redistribute it and/or modify
** it under the terms of the GNU Affero General Public License as
** published by the Free Software Foundation, either version 3 of the
** License, or (at your option) any later version.
**
** This program is distributed in the hope that it will be useful,
** but WITHOUT ANY WARRANTY; without even the implied warranty of
** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
** GNU Affero General Public License for more details.
**
** You should have received a copy of the GNU Affero General Public License
** along with this program.  If not, see <http://www.gnu.org/licenses/>.
**
*/

#ifdef WIN32
#include <windows.h>
#endif
#include "k2pdfopt.h"
#include <pthread.h>

#ifdef HAVE_OCR_LIB
static int k2ocr_gocr_inited=0;
typedef struct
    {
    WILLUSBITMAP *bmp;
    int type,c1,r1,c2,r2,rowbase,dpi;
    double lcheight;
    int    target_dpi;      /* Desired OCR bitmap DPI */
    pthread_mutex_t mutex;
    int done;
    OCRWORDS ocrwords;
    } OCRRESULT;
typedef struct
    {
    OCRRESULT *ocrresult;
    int thindex;
    int n;
    int na;
    } OCRRESULTS;
static int maxthreads=0;
static double ocr_cpu_time_secs=0.;
#if (defined(HAVE_TESSERACT_LIB))
static void **ocrtess_api;
static void *otinit(void *data);
static int k2ocr_tess_inited=0;
static int k2ocr_tess_status=0;
typedef struct
    {
    K2PDFOPT_SETTINGS *k2settings;
    int index;
    int ni;
    char initstr[256];
    } OCRTESSINITINFO;
#endif
static void k2ocr_ocrwords_fill_in(MASTERINFO *masterinfo,OCRWORDS *words,BMPREGION *region,
                                   K2PDFOPT_SETTINGS *k2settings);
/*
static void *ocr_process_word_bitmaps_no_mutex(void *data);
*/
static void ocrresults_make_room(OCRRESULTS *ocrresults);
static void *ocr_process_word_bitmaps_mutex(void *data);
static void ocr_proc_bitmap(void *api,OCRRESULT *ocrresult);
static void ocrword_fillin_mupdf_info(OCRWORD *word,BMPREGION *region);
#endif /* HAVE_OCR_LIB */

/* Functions to support extracting text from PDF using MuPDF lib */
#ifdef HAVE_MUPDF_LIB
static void k2ocr_ocrwords_get_from_ocrlayer(MASTERINFO *masterinfo,OCRWORDS *words,
                                             BMPREGION *region,K2PDFOPT_SETTINGS *k2settings);
static int ocrword_map_to_bitmap(OCRWORD *word,MASTERINFO *masterinfo,BMPREGION *region,
                                 K2PDFOPT_SETTINGS *k2settings,int *index,int *i2);
static int wrectmap_srcword_inside(WRECTMAP *wrectmap,OCRWORD *word,BMPREGION *region,int *i2);
static void wtextchars_group_by_words(WTEXTCHARS *wtcs,OCRWORDS *words,
                                      K2PDFOPT_SETTINGS *k2settings);
static void wtextchars_add_one_row(WTEXTCHARS *wtcs,int i0,int i1,OCRWORDS *words);
static int wtextchars_ligature_pattern(WTEXTCHARS *wtcs,int index);
#endif


void k2ocr_init(K2PDFOPT_SETTINGS *k2settings)

    {
#ifdef HAVE_OCR_LIB
    static char *funcname="k2ocr_init";

    if (maxthreads==0)
        {
        if (k2settings->nthreads<0)
            maxthreads=wsys_num_cpus()*abs(k2settings->nthreads)/100;
        else
            maxthreads=k2settings->nthreads;
        if (maxthreads<1)
            maxthreads=1;
        }
    if (!k2settings->dst_ocr)
        return;
    /* v2.15--don't need to do this--it's in k2pdfopt_settings_init. */
    /*
    if (!k2ocr_inited)
        {
        ocrwords_init(&k2settings->dst_ocrwords);
        k2ocr_inited=1;
        }
    */
#if (!defined(HAVE_TESSERACT_LIB) && defined(HAVE_GOCR_LIB))
    if (k2settings->dst_ocr=='t')
        {
        k2printf(TTEXT_WARN "\a** Tesseract not compiled into this version.  Using GOCR. **"
                TTEXT_NORMAL "\n\n");
        k2settings->dst_ocr='g';
        }
#endif
#if (defined(HAVE_TESSERACT_LIB) && !defined(HAVE_GOCR_LIB))
    if (k2settings->dst_ocr=='g')
        {
        k2printf(TTEXT_WARN "\a** GOCR not compiled into this version.  Using Tesseract. **"
                TTEXT_NORMAL "\n\n");
        k2settings->dst_ocr='t';
        }
#endif
#ifdef HAVE_TESSERACT_LIB
#ifdef HAVE_GOCR_LIB
    if (k2settings->dst_ocr=='t')
        {
#endif
        /* v2.15 fix--specific variable for Tesseract init status */
        if (!k2ocr_tess_inited)
            {
            int i,j,ni;
            pthread_t *thread;
            OCRTESSINITINFO *otii;
            char *istr;

            /* v2.40 -- multithreaded init */
            willus_mem_alloc_warn((void **)(&ocrtess_api),sizeof(void*)*maxthreads,funcname,10); 
            if (maxthreads>=8)
                ni=4;
            else if (maxthreads>=2)
                ni=2;
            else
                ni=1;
            willus_mem_alloc_warn((void**)&thread,sizeof(pthread_t)*ni,funcname,10);
            willus_mem_alloc_warn((void**)&otii,sizeof(OCRTESSINITINFO)*ni,funcname,10);
            if (maxthreads>1)
                k2printf("Initializing OCR for %d threads ",maxthreads);
            for (i=0;i<ni;i++)
                {
                otii[i].k2settings=k2settings;
                otii[i].ni=ni;
                otii[i].index=i;
                otii[i].initstr[0]='\0';
                pthread_create(&thread[i],NULL,otinit,&otii[i]);
                }
            for (istr=NULL,i=0;i<ni;i++)
                {
                pthread_join(thread[i],NULL);
                if (istr==NULL && otii[i].initstr[0]!='\0')
                    istr=otii[i].initstr;
                }
            for (i=j=0;i<maxthreads;i++)
                {
                if (ocrtess_api[i]==NULL)
                    {
                    if (maxthreads>1)
                        k2printf(TTEXT_WARN "x" TTEXT_NORMAL);
                    continue;
                    }
                if (i!=j)
                    ocrtess_api[j]=ocrtess_api[i];
                j++;
                }
            if (maxthreads>1)
                k2printf("\n");
            if (j>0 && istr!=NULL)
                k2printf("%s%s%s\n",TTEXT_BOLD,istr,TTEXT_NORMAL);
            if (j>0 && j<maxthreads)
                {
                k2printf(TTEXT_WARN "** Only able to initialize %d instances of Tesseract. **"
                         TTEXT_NORMAL "\n",j);
                maxthreads=j;
                }
            if (j==0)
                {
                k2printf(TTEXT_WARN "Could not find Tesseract data" TTEXT_NORMAL " (env var TESSDATA_PREFIX = %s).\nUsing GOCR v0.50.\n\n",getenv("TESSDATA_PREFIX")==NULL?"(not assigned)":getenv("TESSDATA_PREFIX"));
                k2ocr_tess_status=-1;
                maxthreads=1;
                }
            else
                k2ocr_tess_status=0;
            willus_mem_free((double **)&otii,funcname);
            willus_mem_free((double **)&thread,funcname);
            k2ocr_tess_inited=1;
            }
#ifdef HAVE_GOCR_LIB
        }
    else
#endif
#endif
#ifdef HAVE_GOCR_LIB
        {
        if (k2settings->dst_ocr=='g')
            {
            if (!k2ocr_gocr_inited)
                {
                k2printf(TTEXT_BOLD "GOCR v0.50 OCR Engine" TTEXT_NORMAL "\n\n");
                k2ocr_gocr_inited=1;
                }
            maxthreads=1;
            }
        }
#endif
#ifdef HAVE_MUPDF_LIB
    /* Could announce MuPDF virtual OCR here, but I think it will just confuse people. */
    /*
    if (k2settings->dst_ocr=='m')
        k2printf(TTEXT_BOLD "Using MuPDF \"Virtual\" OCR" TTEXT_NORMAL "\n\n");
    */
#endif
#endif /* HAVE_OCR_LIB */
    }


#ifdef HAVE_TESSERACT_LIB
void ocrtess_debug_info(char **buf0,int use_ansi)

    {
    static char *funcname="ocrtess_debug_info";
    char *p,*buf;
    char langdef[16];
    char datapath[MAXFILENAMELEN];

    ocrtess_datapath(datapath,NULL,MAXFILENAMELEN-1);
    willus_mem_alloc_warn((void **)buf0,4096,funcname,10);
    buf=(*buf0);
    p=getenv("TESSDATA_PREFIX");
    strcpy(buf,"TESSDATA_PREFIX environment variable:  ");
    if (use_ansi && p!=NULL)
        strcat(buf,ANSI_MAGENTA);
    strcat(buf,p==NULL?"(not set)":p);
    if (use_ansi && p!=NULL)
        strcat(buf,ANSI_NORMAL);
    strcat(buf,"\nTesseract data folder:  ");
    if (use_ansi)
        strcat(buf,ANSI_MAGENTA);
    strcat(buf,datapath);
    if (use_ansi)
        strcat(buf,ANSI_NORMAL);
    strcat(buf,"\n\n");
    strcat(buf,"Contents of ");
    if (use_ansi)
        strcat(buf,ANSI_MAGENTA);
    strcat(buf,datapath);
    if (use_ansi)
        strcat(buf,ANSI_NORMAL);
    strcat(buf,":\n");
    ocrtess_lang_default(NULL,NULL,0,langdef,16,buf,3000,use_ansi);
    strcat(buf,"* - LSTM = \"Long Short-Term Memory\" training data.\n"
               "    LSTM is the latest, most accurate OCR method used by Tesseract v4.x.\n"
               "    TESS = Tesseract v3.x compatible (can be used by v4.x).\n");
    }

    
static void *otinit(void *data)

    {
    OCRTESSINITINFO *otii;
    char *lang;
    char initstr[256];
    int ntries,i,status;

    otii=(OCRTESSINITINFO*)data;
    lang=otii->k2settings->dst_ocr_lang;
    for (i=otii->index;i<maxthreads;i+=otii->ni)
        ocrtess_api[i]=NULL;
    for (ntries=0;ntries<5;ntries++)
        {
        int done=1;

        for (i=otii->index;i<maxthreads;i+=otii->ni)
            {
            if (ocrtess_api[i]!=NULL)
                continue;
            ocrtess_api[i]=ocrtess_init(NULL,NULL,0,
                                        lang[0]=='\0'?NULL:lang,NULL,initstr,255,&status);
            if (ocrtess_api[i]==NULL)
                done=0;
            else
                {
                if (otii->initstr[0]=='\0')
                    strcpy(otii->initstr,initstr);
                if (maxthreads>1)
                    k2printf(".");
                }
            }
        if (done)
            break;
        }
    pthread_exit(NULL);
    return(NULL);
    }
#endif

/*
** v2.15--call close/free functions even if k2settings->dst_ocr not set--may have
**        been called from previous GUI conversion.
*/
void k2ocr_end(K2PDFOPT_SETTINGS *k2settings)

    {
#ifdef HAVE_OCR_LIB
#ifdef HAVE_TESSERACT_LIB
    static char *funcname="k2ocr_end";
    if (k2ocr_tess_inited)
        {
        int i;

        for (i=maxthreads-1;i>=0;i--)
            {
            ocrtess_end(ocrtess_api[i]);
            ocrtess_api[i]=NULL;
            }
        willus_mem_free((double **)&ocrtess_api,funcname);
        }
#endif
    ocrwords_free(&k2settings->dst_ocrwords);
#endif /* HAVE_OCR_LIB */
    }


#ifdef HAVE_OCR_LIB
/*
** Find words in src bitmap and put into words structure.
** If ocrcolumns > maxcols, looks for multiple columns
**
** wrectmaps has information that maps the bitmap pixel space to the original
** source page space so that the location of the words can be used to extract
** the text from a native PDF document if possible (dst_ocr=='m').
**
*/
void k2ocr_ocrwords_fill_in_ex(MASTERINFO *masterinfo,OCRWORDS *words,BMPREGION *region,
                               K2PDFOPT_SETTINGS *k2settings)

    {
    PAGEREGIONS *pageregions,_pageregions;
    int i,maxlevels;

#if (WILLUSDEBUGX & 32)
{
static int ifr=0;
char filename[256];
sprintf(filename,"ocrwords_fill_in_%04d.png",++ifr);
bmpregion_write(region,filename);
printf("k2ocr_ocrwords_fill_in_ex #%d\n",ifr);
}
#endif

    if (k2settings->ocr_max_columns<0 || k2settings->ocr_max_columns <= k2settings->max_columns
           || (k2settings->ocr_detection_type=='p' && k2settings->dst_ocr=='t'))
        {
#if (WILLUSDEBUGX & 32)
printf("Call #1. k2ocr_ocrwords_fill_in\n");
#endif
        k2ocr_ocrwords_fill_in(masterinfo,words,region,k2settings);
        return;
        }
    /* Parse region into columns for OCR */
    pageregions=&_pageregions;
    pageregions_init(pageregions);
    if (k2settings->ocr_max_columns==2 || k2settings->max_columns>1)
        maxlevels = 2;
    else
        maxlevels = 3;
    pageregions_find_columns(pageregions,region,k2settings,masterinfo,maxlevels);
    for (i=0;i<pageregions->n;i++)
        {
        /*
        ** If using built-in source-file OCR layer, don't need to scan bitmap
        ** for text rows.
        */
#if (WILLUSDEBUGX & 32)
printf("\nwrectmaps->n=%d, dst_ocr='%c'\n",region->wrectmaps->n,k2settings->dst_ocr);
#endif
        if (k2settings->dst_ocr!='m' || region->wrectmaps->n!=1)
            bmpregion_find_textrows(&pageregions->pageregion[i].bmpregion,k2settings,0,1,0);
        pageregions->pageregion[i].bmpregion.wrectmaps = region->wrectmaps;
#if (WILLUSDEBUGX & 32)
printf("Call #2. k2ocr_ocrwords_fill_in, pr = %d of %d\n",i+1,pageregions->n);
#endif
        k2ocr_ocrwords_fill_in(masterinfo,words,&pageregions->pageregion[i].bmpregion,k2settings);
        }
    pageregions_free(pageregions);
    }


/*
** Find words in src bitmap and put into words structure.
** v2.40:  Multithreaded using pthreads.
*/
static void k2ocr_ocrwords_fill_in(MASTERINFO *masterinfo,OCRWORDS *words,BMPREGION *region,
                                   K2PDFOPT_SETTINGS *k2settings)

    {
    int i,type;
    OCRRESULTS _ocrresults,*ocrresults;
    OCRRESULTS *ocrr;
    pthread_t *thread;
    static char *funcname="k2ocr_ocrwords_fill_in";

#ifdef HAVE_TESSERACT_LIB
#ifdef HAVE_GOCR_LIB
    if (k2settings->dst_ocr=='t' && !k2ocr_tess_status)
#else
    if (!k2ocr_tess_status)
#endif
        type='t';
#ifdef HAVE_GOCR_LIB
    else if (k2settings->dst_ocr=='g' || k2settings->dst_ocr=='t')
#endif
#endif
#ifdef HAVE_GOCR_LIB
        type='g';
#endif
#ifdef HAVE_MUPDF_LIB
#if (defined(HAVE_TESSERACT_LIB) || defined(HAVE_GOCR_LIB))
    if (k2settings->dst_ocr=='m')
#endif
        type='m';
#endif

/*
k2printf("@ocrwords_fill_in (%d x %d)...tr=%d\n",region->bmp->width,region->bmp->height,region->textrows.n);
if (region->textrows.n==0)
{
bmpregion_write(region,"out.png");
exit(10);
}
*/
#if (WILLUSDEBUGX & 32)
k2printf("@ocrwords_fill_in... type='%c'\n",type);
#endif
/*
{
char filename[MAXFILENAMELEN];
count++;
sprintf(filename,"out%03d.png",count);
bmp_write(src,filename,stdout,100);
}
*/

    if (region->bmp->width<=0 || region->bmp->height<=0)
        return;

#if (WILLUSDEBUGX & 32)
k2printf("OCR REGION:  @%3d,%3d = %3d x %3d\n",
region->c1, region->r1, region->c2-region->c1+1, region->r2-region->r1+1);
k2printf("  %d row%s of text, dst_ocr='%c'\n",region->textrows.n,region->textrows.n==1?"":"s",k2settings->dst_ocr);
#endif
#if (defined(HAVE_MUPDF_LIB))
    if (k2settings->dst_ocr=='m')
        {
        k2ocr_ocrwords_get_from_ocrlayer(masterinfo,words,region,k2settings);
        return;
        }
#endif
    willus_mem_alloc_warn((void**)&ocrr,sizeof(OCRRESULTS)*maxthreads,funcname,10);
    willus_mem_alloc_warn((void**)&thread,sizeof(pthread_t)*maxthreads,funcname,10);
    ocrresults=&_ocrresults;
    ocrresults->n=0;
    ocrresults->na=16;
    willus_mem_alloc_warn((void **)&ocrresults->ocrresult,sizeof(OCRRESULT)*ocrresults->na,funcname,10);

    if (k2settings->dst_ocr=='t' && 
          (k2settings->ocr_detection_type=='p' || k2settings->ocr_detection_type=='c'))
        {
        OCRRESULT *ocrresult;

        /* New in v2.50:  Use Tesseract to parse entire region for OCR */
        ocrresults_make_room(ocrresults);
        ocrresult=&ocrresults->ocrresult[ocrresults->n];
        ocrresult->bmp=region->bmp8;
        ocrresult->dpi=region->dpi;
        ocrresult->c1=region->c1;
        ocrresult->r1=region->r1;
        ocrresult->c2=region->c2;
        ocrresult->r2=region->r2;
        pthread_mutex_init(&ocrresult->mutex,NULL);
        ocrresult->done=0;
        ocrwords_init(&ocrresult->ocrwords);
        ocrresult->rowbase=-1;
        ocrresult->lcheight=-1;
        ocrresult->target_dpi=k2settings->ocr_dpi;
        ocrresult->type=type;
        ocrresults->n++;
        }
    else /* Use k2pdfopt engine to parse row by row */
        {
        /*
        ** Go text row by text row--store bitmap for each word into ocrresults
        ** structure which will then be processed with multiple parallel threads.
        ** (v2.40)
        */
        for (i=0;i<region->textrows.n;i++)
            {
            BMPREGION _newregion,*newregion;
            int j,rowbase,r1,r2;
            double lcheight;
            TEXTWORDS *textwords;

            newregion=&_newregion;
            bmpregion_init(newregion);
            bmpregion_copy(newregion,region,0);
            r1=newregion->r1=region->textrows.textrow[i].r1;
            r2=newregion->r2=region->textrows.textrow[i].r2;
#if (WILLUSDEBUGX & 32)
printf("  Row %2d:  @%3d,%3d = %3d x %3d\n",i+1,
region->textrows.textrow[i].c1,
region->textrows.textrow[i].r1,
region->textrows.textrow[i].c2-region->textrows.textrow[i].c1+1,
region->textrows.textrow[i].r2-region->textrows.textrow[i].r1+1);
#endif
            newregion->bbox.type = REGION_TYPE_TEXTLINE;
            rowbase=region->textrows.textrow[i].rowbase;
            lcheight=region->textrows.textrow[i].lcheight;
            /* Sanity check on rowbase, lcheight */
            if ((double)(rowbase-r1)/(r2-r1) < .5)
                rowbase = r1+(r2-r1)*0.7;
            if (lcheight/(r2-r1) < .33)
                lcheight = 0.33*(r2-r1);
            if (k2settings->dst_ocr=='t' && k2settings->ocr_detection_type=='l')
                {
                OCRRESULT *ocrresult;

                /* New in v2.50:  Use Tesseract to parse entire line of text */
                /* Don't OCR if line height exceeds spec */
                if ((double)(region->textrows.textrow[i].r2-region->textrows.textrow[i].r1+1)
                              / region->dpi > k2settings->ocr_max_height_inches)
                    continue;
                ocrresults_make_room(ocrresults);
                ocrresult=&ocrresults->ocrresult[ocrresults->n];
                ocrresult->bmp=region->bmp8;
                ocrresult->dpi=region->dpi;
                ocrresult->c1=region->textrows.textrow[i].c1;
                ocrresult->r1=region->textrows.textrow[i].r1;
                ocrresult->c2=region->textrows.textrow[i].c2;
                ocrresult->r2=region->textrows.textrow[i].r2;
                pthread_mutex_init(&ocrresult->mutex,NULL);
                ocrresult->done=0;
                ocrwords_init(&ocrresult->ocrwords);
                ocrresult->rowbase=rowbase;
                ocrresult->lcheight=lcheight;
                ocrresult->target_dpi=k2settings->ocr_dpi;
                ocrresult->type=type;
                ocrresults->n++;
                continue;
                }

            /* Use k2pdfopt engine to break text row into words (also establishes lcheight) */
            bmpregion_one_row_find_textwords(newregion,k2settings,0);
            textwords = &newregion->textrows;
            /* Add each word */
            for (j=0;j<textwords->n;j++)
                {
                OCRRESULT *ocrresult;

                /* Don't OCR if "word" height exceeds spec */
                if ((double)(textwords->textrow[j].r2-textwords->textrow[j].r1+1)/region->dpi
                         > k2settings->ocr_max_height_inches)
                    continue;
                ocrresults_make_room(ocrresults);
                ocrresult=&ocrresults->ocrresult[ocrresults->n];
                ocrresult->bmp=region->bmp8;
                ocrresult->dpi=region->dpi;
                ocrresult->c1=textwords->textrow[j].c1;
                ocrresult->r1=textwords->textrow[j].r1;
                ocrresult->c2=textwords->textrow[j].c2;
                ocrresult->r2=textwords->textrow[j].r2;
                pthread_mutex_init(&ocrresult->mutex,NULL);
                ocrresult->done=0;
                ocrwords_init(&ocrresult->ocrwords);
                ocrresult->rowbase=rowbase;
                ocrresult->lcheight=lcheight;
                ocrresult->target_dpi=k2settings->ocr_dpi;
                ocrresult->type=type;
                ocrresults->n++;
                } /* text word loop */
            bmpregion_free(newregion);
            } /* text row loop */
        } /* If detection type == 'p' */

#if (WILLUSDEBUGX & 32)
{
char filename[256];
int i,j;
WILLUSBITMAP *bmp,_bmp;
static int iregion=0;

bmp=&_bmp;
bmp_init(bmp);
for (j=0;j<ocrresults->n;j++)
{
OCRRESULT *ocrresult;

ocrresult=&ocrresults->ocrresult[j];
bmp->width=ocrresult->c2-ocrresult->c1+1;
bmp->height=ocrresult->r2-ocrresult->r1+1;
bmp->bpp=8;
bmp_alloc(bmp);
for (i=0;i<256;i++)
bmp->red[i]=bmp->green[i]=bmp->blue[i]=i;
for (i=0;i<bmp->height;i++)
{
unsigned char *s,*d;
s=bmp_rowptr_from_top(ocrresult->bmp,ocrresult->r1+i)+ocrresult->c1;
d=bmp_rowptr_from_top(bmp,i);
memcpy(d,s,bmp->width);
}
sprintf(filename,"ocr_region_%04d.png",iregion+1);
bmp_write(bmp,filename,stdout,100);
bmp_free(bmp);
iregion++;
}
}
#endif


    /*
    ** Process the OCR regions with multiple threads.
    ** If more than 4 threads, don't do mutex blocking--just have each thread
    ** process 1/maxthreads of the words.
    ** (v2.40)
    */
    {
    clock_t start,stop;
    start=clock();
    for (i=0;i<maxthreads;i++)
        {
        ocrr[i]=(*ocrresults);
        ocrr[i].thindex=i;
        pthread_create(&thread[i],NULL,ocr_process_word_bitmaps_mutex,&ocrr[i]);
        }
    for (i=0;i<maxthreads;i++)
        pthread_join(thread[i],NULL);
    stop=clock();
    ocr_cpu_time_secs += (double)stop/CLOCKS_PER_SEC-(double)start/CLOCKS_PER_SEC;
    /* printf("TOT TIME = %.2f seconds.\n",ocr_cpu_time_secs); */
    }


    /*
    ** Done with multi-threading section.  Store UTF-8 text from OCR results.
    */
    for (i=0;i<ocrresults->n;i++)
        {
        OCRRESULT *ocrresult;
        int j;

        ocrresult=&ocrresults->ocrresult[i];
/*
#if (WILLUSDEBUGX & 32)
printf("ocrresults->ocrresult[%d]=%p, word=%p\n",i,ocrresult,ocrresult->ocrwords.word);
#endif
*/
        for (j=0;j<ocrresult->ocrwords.n;j++)
/*
        if (ocrresult->word[0]!='\0' && ocrresult->done==1)
*/
            {
            OCRWORD word;

/*
            ocrword_init(&word);
            word.c=ocrresult->c1;
*/
            /* Use same rowbase for whole row */
/*
            word.r=ocrresult->rowbase;
            word.maxheight=ocrresult->rowbase-ocrresult->r1;
            word.w=ocrresult->c2-ocrresult->c1+1;
            word.h=ocrresult->r2-ocrresult->r1+1;
            word.lcheight=ocrresult->lcheight;
            word.rot=0;
            word.text=ocrresult->word;
*/
/*
#if (WILLUSDEBUGX & 32)
{
OCRWORD *word0;
word0=&ocrresult->ocrwords.word[j];
printf("%d.%d. '%s', rowbase=%d, w=%d, h=%d, mh=%g\n",i+1,j+1,word0->text,word0->r,word0->w,word0->h,word0->maxheight);
}
#endif
*/
            word=ocrresult->ocrwords.word[j];
            word.c += ocrresult->c1;
            word.r += ocrresult->r1;
            ocrword_fillin_mupdf_info(&word,region);
#if (WILLUSDEBUGX & 32)
if (word.text[0]!='\0')
{
char filename[MAXFILENAMELEN];
FILE *f;
static int iregion=0;
sprintf(filename,"words_r%04d.txt",iregion+1);
iregion++;
f=fopen(filename,j==0?"wb":"ab");
fprintf(f,"%s%s\n",j==0?"":" ",word.text);
fclose(f);
k2printf("%d.%d. '%s', rowbase=%d, w=%d, h=%d, mh=%g\n",i+1,j+1,word.text,word.r,word.w,word.h,word.maxheight);
}
else
k2printf("%d.%d. (OCR failed)\n",i+1,j+1);
#endif
            ocrwords_add_word(words,&word);
            }
        } /* Loop ocrresults->n */
/*
    for (i=ocrresults->n-1;i>=0;i--)
        willus_mem_free((double **)&ocrresults->ocrresult[i].word,funcname);
*/
    for (i=ocrresults->n-1;i>=0;i--)
        ocrwords_free(&ocrresults->ocrresult[i].ocrwords);
    willus_mem_free((double**)&ocrresults->ocrresult,funcname);
    willus_mem_free((double**)&thread,funcname);
    willus_mem_free((double**)&ocrr,funcname);
#if (WILLUSDEBUGX & 32)
printf("Done k2ocr_ocrwords_fill_in()\n");
#endif
    }


static void ocrresults_make_room(OCRRESULTS *ocrresults)

    {
    static char *funcname="ocrresults_make_room";

    if (ocrresults->n>=ocrresults->na)
        {
        willus_mem_realloc_robust_warn((void **)&ocrresults->ocrresult,
                                       ocrresults->na*2*sizeof(OCRRESULT),
                                       ocrresults->na*sizeof(OCRRESULT),funcname,10);
        ocrresults->na*=2;
        }
    }


double k2ocr_cpu_time_secs(void)

    {
    return(ocr_cpu_time_secs);
    }


void k2ocr_cpu_time_reset(void)

    {
    ocr_cpu_time_secs=0.;
    }


int k2ocr_max_threads(void)

    {
    return(maxthreads);
    }


/*
** In tests I did, the ...no_mutex function is never faster
** than the ..._mutex() function.
*/
#if 0
static void *ocr_process_word_bitmaps_no_mutex(void *data)

    {
    OCRRESULTS *ocrresults;
    int i;

    ocrresults=(OCRRESULTS *)data;
    for (i=ocrresults->thindex;i<ocrresults->n;i+=maxthreads)
        {
        OCRRESULT *ocrresult;

        ocrresult=&ocrresults->ocrresult[i];
        ocrresult->done=1;
        ocr_proc_bitmap(ocrtess_api[ocrresults->thindex],ocrresult);
        }
    pthread_exit(NULL);
    return(NULL);
    }
#endif


static void *ocr_process_word_bitmaps_mutex(void *data)

    {
    OCRRESULTS *ocrresults;
    int i;

    ocrresults=(OCRRESULTS *)data;
    for (i=0;i<ocrresults->n;i++)
        {
        OCRRESULT *ocrresult;

        ocrresult=&ocrresults->ocrresult[i];
        pthread_mutex_lock(&ocrresult->mutex);
        if (ocrresult->done==1)
            {
            pthread_mutex_unlock(&ocrresult->mutex);
            continue;
            }
        ocrresult->done=1;
        pthread_mutex_unlock(&ocrresult->mutex);
        ocr_proc_bitmap(ocrtess_api[ocrresults->thindex],ocrresult);
        }
    pthread_exit(NULL);
    return(NULL);
    }


/*
** v2.50:  New call type that gets multiple words into OCRWORDS structure
*/
static void ocr_proc_bitmap(void *api,OCRRESULT *ocrresult)

    {
    /*
    char wordbuf[256];
    */

    switch (ocrresult->type)
        {
#ifdef HAVE_TESSERACT_LIB
        case 't':
            {
            double downsample;

            if (ocrresult->lcheight > 0. && ocrresult->target_dpi < 0
                      && ocrresult->lcheight > -ocrresult->target_dpi)
                downsample = (double)-ocrresult->target_dpi / ocrresult->lcheight;
            else if (ocrresult->dpi > 0 && ocrresult->target_dpi > 0
                      && ocrresult->dpi > ocrresult->target_dpi)
                downsample = (double)ocrresult->target_dpi / ocrresult->dpi;
            else
                downsample = 1.;
            ocrtess_ocrwords_from_bmp8(api,&ocrresult->ocrwords,ocrresult->bmp,
                              ocrresult->c1,ocrresult->r1,
                              ocrresult->c2,ocrresult->r2,ocrresult->dpi,
                              -1,downsample,NULL);
/*
#if (WILLUSDEBUGX & 32)
{
int i;
printf("      OCR TESS:  (%d,%d)-(%d,%d) -- got %d words.\n",
ocrresult->c1,ocrresult->r1,ocrresult->c2,ocrresult->r2,ocrresult->ocrwords.n);
printf("                 ocrresult=%p, ocrresult->ocrwords.word=%p\n",ocrresult,ocrresult->ocrwords.word);
for (i=0;i<ocrresult->ocrwords.n;i++)
{
OCRWORD *word;
word=&ocrresult->ocrwords.word[i];
printf("        %2d. '%s' (%d,%d) w=%d, h=%d\n",i+1,word->text,word->c,word->r,word->w,word->h);
}
}
#endif
*/
            break;
            }
#endif
#ifdef HAVE_GOCR_LIB
        case 'g':
            gocr_ocrwords_from_bmp8(&ocrresult->ocrwords,ocrresult->bmp,
                                       ocrresult->c1,ocrresult->r1,
                                       ocrresult->c2,ocrresult->r2,0,1);
            break;
#endif
        default:
            /*
            wordbuf[0]='m';
            wordbuf[1]='\0';
            */
            break;
        }

    /*
    willus_mem_alloc_warn((void **)&ocrresult->word,strlen(wordbuf)+1,"ocr_proc_bitmap",10);
    strcpy(ocrresult->word,wordbuf);
    */
    }
    

static void ocrword_fillin_mupdf_info(OCRWORD *word,BMPREGION *region)

    {
    int xp,yp; /* Top left of word on passed bitmap */
    int xc,yc; /* Center of word */
    int i;
    double pix2ptsw,pix2ptsh;
    WRECTMAP *wrmap;
    WRECTMAPS *wrectmaps;

    wrectmaps=region->wrectmaps;
#if (WILLUSDEBUGX & 0x400)
printf("Fillin: wrectmaps->n=%d\n",wrectmaps==NULL ? 0 : wrectmaps->n);
#endif
    if (wrectmaps==NULL || wrectmaps->n==0)
        {
        word->x0=word->c;
        word->y0=word->r-word->maxheight;
        word->w0=word->w;
        word->h0=word->h;
        word->rot0_deg=region->rotdeg;
        word->pageno=0;
        return;
        }
    xp=word->c;
    yp=word->r-word->maxheight;
    xc=xp+word->w/2;
    yc=yp+word->h/2;
#if (WILLUSDEBUGX & 0x400)
printf("    word xc,yc = (%d,%d)\n",xc,yc);
#endif
    for (i=0;i<wrectmaps->n;i++)
        {
#if (WILLUSDEBUGX & 0x400)
printf("    wrectmap[%d]= (%g,%g) , %g x %g\n",i,
wrectmaps->wrectmap[i].coords[1].x,
wrectmaps->wrectmap[i].coords[1].y,
wrectmaps->wrectmap[i].coords[2].x,
wrectmaps->wrectmap[i].coords[2].y);
#endif
        if (wrectmap_inside(&wrectmaps->wrectmap[i],xc,yc))
            break;
        }
    if (i>=wrectmaps->n)
        {
#if (WILLUSDEBUGX & 0x400)
printf("Not inside.\n");
#endif
        i=0;
        }
    wrmap=&wrectmaps->wrectmap[i];
    word->pageno = wrmap->srcpageno;
    word->rot0_deg = wrmap->srcrot;
    pix2ptsw = 72./wrmap->srcdpiw;
    pix2ptsh = 72./wrmap->srcdpih;
    word->x0 = (wrmap->coords[0].x + (xp-wrmap->coords[1].x))*pix2ptsw;
    word->y0 = (wrmap->coords[0].y + (yp-wrmap->coords[1].y))*pix2ptsh;
    word->w0 = word->w*pix2ptsw;
    word->h0 = word->h*pix2ptsh;
#if (WILLUSDEBUGX & 0x400)
printf("Word: (%5.1f,%5.1f) = %5.1f x %5.1f (page %2d)\n",word->x0,word->y0,word->w0,word->h0,word->pageno);
#endif
    }
#endif /* HAVE_OCR_LIB */


/*
** In a contiguous rectangular region that is mapped to the PDF source file,
** find rows of text assuming a single column of text.
*/
#if (defined(HAVE_MUPDF_LIB))
static void k2ocr_ocrwords_get_from_ocrlayer(MASTERINFO *masterinfo,OCRWORDS *dwords,
                                             BMPREGION *region,K2PDFOPT_SETTINGS *k2settings)

    {
    static WTEXTCHARS *wtcs=NULL;
    static WTEXTCHARS _wtcs;
    static OCRWORDS *words;
    static OCRWORDS _words;
    static int pageno=-1;
    static char pdffile[512];
    int i;

#if (WILLUSDEBUGX & 0x10000)
printf("@k2ocr_ocrwords_get_from_ocrlayer.\n");
#endif
    if (wtcs==NULL)
        {
        wtcs=&_wtcs;
        wtextchars_init(wtcs);
        words=&_words;
        ocrwords_init(words);
        pdffile[0]='\0';
        }
    if (pageno!=masterinfo->pageinfo.srcpage || strcmp(pdffile,masterinfo->srcfilename))
        {
        ocrwords_free(words);
        wtextchars_clear(wtcs);
        wtextchars_fill_from_page(wtcs,masterinfo->srcfilename,masterinfo->pageinfo.srcpage,"");
#if (WILLUSDEBUGX & 0x10000)
{
int i;
for (i=0;i<wtcs->n;i++)
printf("Char[%2d]:  ucs='%c'(%d), xp=%7.3f, yp=%7.3f, x1=%7.3f, x2=%7.3f\n",i,wtcs->wtextchar[i].ucs,wtcs->wtextchar[i].ucs,wtcs->wtextchar[i].xp,wtcs->wtextchar[i].yp,wtcs->wtextchar[i].x1,wtcs->wtextchar[i].x2);
}
#endif
        wtextchars_rotate_clockwise(wtcs,360-(int)masterinfo->pageinfo.srcpage_rot_deg);
        wtextchars_group_by_words(wtcs,words,k2settings);
#if (WILLUSDEBUGX & 0x10000)
{
int i;
for (i=0;i<words->n;i++)
printf("Word[%4d]: (%6.2f,%6.2f) %6.2f x %6.2f '%s'\n",i,words->word[i].x0,words->word[i].y0,words->word[i].w0,words->word[i].h0,words->word[i].text);
}
#endif
        pageno=masterinfo->pageinfo.srcpage;
        strncpy(pdffile,masterinfo->srcfilename,511);
        pdffile[511]='\0';
        }
#if (WILLUSDEBUGX & 0x10000)
{
FILE *f;
int y0;
static int count=0;
WRECTMAPS *wrectmaps;
wrectmaps=region->wrectmaps;
printf("\n");
if (wrectmaps!=NULL)
{
printf("    Got %d word regions.\n",wrectmaps->n);
printf("    region = (%d,%d) - (%d,%d) pixels\n",region->c1,region->r1,region->c2,region->r2);
for (i=0;i<wrectmaps->n;i++)
{
printf("    wrectmap[%d]= srcorg=(%g,%g); dstorg=(%g,%g) , %g x %g\n",i,
wrectmaps->wrectmap[i].coords[0].x,
wrectmaps->wrectmap[i].coords[0].y,
wrectmaps->wrectmap[i].coords[1].x,
wrectmaps->wrectmap[i].coords[1].y,
wrectmaps->wrectmap[i].coords[2].x,
wrectmaps->wrectmap[i].coords[2].y);
printf("        srcpageno=%d\n",wrectmaps->wrectmap[i].srcpageno);
printf("        srcwidth=%d\n",wrectmaps->wrectmap[i].srcwidth);
printf("        srcheight=%d\n",wrectmaps->wrectmap[i].srcheight);
printf("        srcdpiw=%g\n",wrectmaps->wrectmap[i].srcdpiw);
printf("        srcdpih=%g\n",wrectmaps->wrectmap[i].srcdpih);
printf("        srcrot=%d\n",wrectmaps->wrectmap[i].srcrot);
}
}
y0=region->r2;
f=fopen("words.ep","w");
fprintf(f,"; srcdpi=%d, dstdpi=%d\n",k2settings->src_dpi,k2settings->dst_dpi);
fprintf(f,"/sm off\n/sc on\n/sd off\n");
fprintf(f,"/sa l \"Region (pixels)\" 2\n");
fprintf(f,"%d %d\n",region->c1,y0-region->r1);
fprintf(f,"%d %d\n",region->c2,y0-region->r1);
fprintf(f,"%d %d\n",region->c2,y0-region->r2);
fprintf(f,"%d %d\n",region->c1,y0-region->r2);
fprintf(f,"%d %d\n",region->c1,y0-region->r1);
fprintf(f,"//nc\n");
#endif

    /* Map word PDF positions (in source file) to destination bitmap pixels */
    /* I'm not sure this will work entirely correctly with right-to-left text */
#if (WILLUSDEBUGX & 0x10000)
printf("words->n=%d\n",words->n);
#endif
    for (i=0;i<words->n;i++)
        {
        int index,i2;

#if (WILLUSDEBUGX & 0x10000)
printf("**Word[%4d] = '%s'!\n",i,words->word[i].text);
#endif
        for (i2=-1,index=0;1;index++)
            {
            OCRWORD _word,*word;
            int n,dn;

            /*
            ** Make copy since the function call below can modify the value inside "word"
            */
            word=&_word;
            ocrword_copy(word,&words->word[i]); /* Allocates new memory */
#if (WILLUSDEBUGX & 0x10000)
printf("   (word = '%s')\n",word->text);
#endif
            if (i2>=0)
                {
                ocrword_truncate(word,i2+1,word->n-1);
                dn = words->word[i].n-word->n;
#if (WILLUSDEBUGX & 0x10000)
printf("truncated word (%d-%d) = '%s'!\n",i2+1,word->n-1,word->text);
#endif
                }
            else
                dn=0;
            n=word->n;
            if (!ocrword_map_to_bitmap(word,masterinfo,region,k2settings,&index,&i2))
                {
                ocrword_free(word);
                break;
                }
#if (WILLUSDEBUGX & 0x10000)
printf("i2 = %d (dn=%d, n=%d); word='%s' (len=%d)\n",i2,dn,n,word->text,word->n);
fprintf(f,"/sa m 2 2\n");
fprintf(f,"/sa l \"'%s'\" 2\n",word->text);
fprintf(f,"%d %d\n",word->c,y0-word->r);
fprintf(f,"%d %d\n",word->c+word->w,y0-word->r);
fprintf(f,"%d %d\n",word->c+word->w,y0-(word->r-word->h));
fprintf(f,"%d %d\n",word->c,y0-(word->r-word->h));
fprintf(f,"%d %d\n",word->c,y0-word->r);
fprintf(f,"//nc\n");
#endif
            ocrwords_add_word(dwords,word);
            ocrword_free(word);
            if (i2>=n-1)
                break;
#if (WILLUSDEBUGX & 0x10000)
printf("continuing\n");
#endif
            i2+=dn;
            }
        }
    /* These are static--keep them around for the next call in case it's on the same page */
    /*
    ocrwords_free(words);
    wtextchars_free(wtcs);
    */
#if (WILLUSDEBUGX & 0x10000)
fclose(f);
wfile_written_info("words.ep",stdout);
count++;
/*
if (count==5)
exit(10);
*/
}
printf("ALL DONE.\n");
#endif
    }


/*
** Determine pixel locations of OCR word on source bitmap region.
** Return 1 if any part of word is within region, 0 if not.
**
** If returns 1, then:
**
**     wrectmaps->wrectmap[(*index)] is the region containing the word
**
**     (*i2) gets the index of the "last" letter of the word which is still within
**        the wrectmaps->wrectmap[(*index)].
**
** WARNING:  Contents of "word" may be modified.
**
*/
static int ocrword_map_to_bitmap(OCRWORD *word,MASTERINFO *masterinfo,BMPREGION *region,
                                 K2PDFOPT_SETTINGS *k2settings,int *index,int *i2)

    {
    int i;
    WRECTMAPS *wrectmaps;

    wrectmaps=region->wrectmaps;
/*
#if (WILLUSDEBUGX & 0x10000)
printf("wrectmaps->n = %d\n",wrectmaps->n);
#endif
*/
    for (i=(*index);i<wrectmaps->n;i++)
        if (wrectmap_srcword_inside(&wrectmaps->wrectmap[i],word,region,i2))
            {
            word->rot0_deg=masterinfo->pageinfo.srcpage_rot_deg;
            word->pageno=masterinfo->pageinfo.srcpage;
            (*index)=i;
            return(1);
            }
    return(0);
    }


#if (WILLUSDEBUGX & 0x10000)
static void wrectmaps_echo(WRECTMAPS *wrectmaps)

    {
    int i;
    for (i=0;i<wrectmaps->n;i++)
        {
        printf("wrectmap %d of %d\n",i+1,wrectmaps->n);
        printf("    source bm top left = (%g,%g)\n",wrectmaps->wrectmap[i].coords[0].x,
                                                    wrectmaps->wrectmap[i].coords[0].y);
        printf("    wrapbmp top left = (%g,%g)\n",wrectmaps->wrectmap[i].coords[1].x,
                                                  wrectmaps->wrectmap[i].coords[1].y);
        printf("    w/h of region = (%g,%g)\n",wrectmaps->wrectmap[i].coords[2].x,
                                               wrectmaps->wrectmap[i].coords[2].y);
        }
    }
#endif


/*
** WARNING:  Contents of "word" may be modified.
**
** If 1 is returned, (*index2) gets the index to the "last" letter of the word that 
** is still within the wrectmap.
*/
static int wrectmap_srcword_inside(WRECTMAP *wrectmap,OCRWORD *word,BMPREGION *region,int *index2)

    {
    double cx_word,cy_word;
    double x0_wrect,x1_wrect,y0_wrect,y1_wrect;
    double cx_wrect,cy_wrect;

    (*index2)=-1;
    /* compute word and wrectmap in src coords (pts) */
    /*
    ** (cx_word, cy_word) = position of center of word on source page, from top left,
    **                      in points.
    */
    cx_word=word->x0+word->w0/2.;
    cy_word=word->y0+word->h0/2.;
    /*
    ** x0_wrect, y0_wrect = position of upper-left region box on source page, in points.
    ** y1_wrect, y1_wrect = position of lower-right region box on source page, in points.
    */
    /* Original formula from before 7-13-2014--seems off.
    x0_wrect=(wrectmap->coords[0].x+(region->c1-wrectmap->coords[1].x))*72./wrectmap->srcdpiw;
    x1_wrect=(wrectmap->coords[0].x+(region->c2+1-wrectmap->coords[1].x))*72./wrectmap->srcdpiw;
    y0_wrect=(wrectmap->coords[0].y+(region->r1-wrectmap->coords[1].y))*72./wrectmap->srcdpih;
    y1_wrect=(wrectmap->coords[0].y+(region->r2+1-wrectmap->coords[1].y))*72./wrectmap->srcdpih;
    */
    x0_wrect=wrectmap->coords[0].x*72./wrectmap->srcdpiw;
    x1_wrect=x0_wrect + wrectmap->coords[2].x*72./wrectmap->srcdpiw;
    y0_wrect=wrectmap->coords[0].y*72./wrectmap->srcdpih;
    y1_wrect=y0_wrect + wrectmap->coords[2].y*72./wrectmap->srcdpih;
    cx_wrect=(x0_wrect+x1_wrect)/2.;
    cy_wrect=(y0_wrect+y1_wrect)/2.;

    /* If word box is completely outside, skip */
    if (word->x0 >= x1_wrect || (word->x0+word->w0) <= x0_wrect 
                             || word->y0 >= y1_wrect || (word->y0+word->h0) <= y0_wrect)
        return(0);
    if (((cx_word >= x0_wrect && word->x0 <= x1_wrect) || (cx_wrect >= word->x0 && cx_wrect <= word->x0+word->w0))
            && ((cy_word >= y0_wrect && cy_word <= y1_wrect) || (cy_wrect >= word->y0 && cy_wrect <= word->y0+word->h0)))
        {
        int i,i1,i2;

        /* Find out which letters are in/out */
        /* From left... */
#if (WILLUSDEBUGX & 0x10000)
printf("Word:  '%s'\n",word->text);
#endif
        for (i=0;i<word->n;i++)
            {
            double x0,x1,x2,dx;

            x0=(i==0)?0.:word->cpos[i-1];
            dx=word->cpos[i]-x0;
            x1=word->x0+x0+dx*0.2;
            x2=word->x0+x0+dx*0.8;
#if (WILLUSDEBUGX & 0x10000)
printf("    i1=%d, x1=%g, x2=%g, x0_rect=%g, x1_rect=%g\n",i,x1,x2,x0_wrect,x1_wrect);
#endif
            if ((x1>=x0_wrect && x1<=x1_wrect) || (x2>=x0_wrect && x2<=x1_wrect))
                break;
            }
        i1=i;
        if (i1>=word->n)
            return(0);
        /* From right... */
        for (i=word->n-1;i>i1;i--)
            {
            double x0,x1,x2,dx;

            x0=(i==0)?0.:word->cpos[i-1];
            dx=word->cpos[i]-x0;
            x1=word->x0+x0+dx*0.2;
            x2=word->x0+x0+dx*0.8;
#if (WILLUSDEBUGX & 0x10000)
printf("    i2=%d, x1=%g, x2=%g, x0_rect=%g, x1_rect=%g\n",i,x1,x2,x0_wrect,x1_wrect);
#endif
            if ((x1>=x0_wrect && x1<=x1_wrect) || (x2>=x0_wrect && x2<=x1_wrect))
                break;
            }
        i2=i;
        (*index2)=i2;
#if (WILLUSDEBUGX & 0x10000)
printf("    i1=%d, i2=%d, n=%d\n",i1,i2,word->n);
#endif
        if (i2<i1)
            return(0);
#if (WILLUSDEBUGX & 0x10000)
printf("    Chars %d to %d\n",i1+1,i2+1);
printf("    Region = (%g,%g) - (%g,%g)\n",x0_wrect,y0_wrect,x1_wrect,y1_wrect);
printf("    Word = (%g,%g) - (%g,%g)\n",word->x0,word->y0,word->x0+word->w0,word->y0+word->h0);
#endif
        word->c=word->x0*wrectmap->srcdpiw/72.-wrectmap->coords[0].x+wrectmap->coords[1].x+.5;
        word->r=((double)word->r/100.)*wrectmap->srcdpih/72.-wrectmap->coords[0].y+wrectmap->coords[1].y+.5;
        word->w=(int)(word->w0*wrectmap->srcdpiw/72.+.5);
        word->h=(int)(word->h0*wrectmap->srcdpih/72.+.5);
#if (WILLUSDEBUGX & 0x10000)
printf("    Word mapped to (%d,%d), %d x %d\n",word->c,word->r,word->w,word->h);
printf("    Word mxheight = %g\n",word->maxheight);
#endif
        word->maxheight=word->maxheight*wrectmap->srcdpih/72.;
        word->lcheight=word->maxheight/1.7;
        word->rot=0;
        if (i1>0 || i2<word->n-1)
            {
            ocrword_truncate(word,i1,i2);
#if (WILLUSDEBUGX & 0x10000)
aprintf(ANSI_YELLOW "    Truncated='%s' (len=%d)" ANSI_NORMAL "\n",word->text,word->n);
printf("    Word = (%g,%g) - (%g,%g)\n",word->x0,word->y0,word->x0+word->w0,word->y0+word->h0);
printf("    Word mapped to (%d,%d), %d x %d\n",word->c,word->r,word->w,word->h);
printf("    Word mxheight = %g\n",word->maxheight);
#endif
            }
        return(1);
        }
    return(0);
    }


static void wtextchars_group_by_words(WTEXTCHARS *wtcs,OCRWORDS *words,
                                      K2PDFOPT_SETTINGS *k2settings)

    {
    double vert_spacing_threshold;
    static char *funcname="wtextchars_group_by_words";
    int i,i0;

    if (wtcs->n<=0)
        return;    
    /* Manually sort characters in OCR layer */
    if (k2settings->dst_ocr_visibility_flags&32)
        {
        double *ch; /* population of character heights */

        willus_dmem_alloc_warn(39,(void **)&ch,sizeof(double)*wtcs->n,funcname,10);
        /* Sort by position */
        wtextchars_sort_vertically_by_position(wtcs,0);
        /* Create population character heights to determine row spacing threshold */
        for (i=0;i<wtcs->n-1;i++)
            ch[i]=wtcs->wtextchar[i].y2-wtcs->wtextchar[i].y1;
        sortd(ch,wtcs->n);
        vert_spacing_threshold=0.67*ch[wtcs->n/10];
        willus_dmem_free(39,&ch,funcname);
        }
    else
        vert_spacing_threshold=-1;
    /* Group characters row by row and add one row at a time */
    for (i0=0,i=1;i<wtcs->n;i++)
        {
        double dx,dy;
        int newrow;

        newrow=0;
        dy=wtcs->wtextchar[i].yp - wtcs->wtextchar[i-1].yp;
        dx=wtcs->wtextchar[i].xp - wtcs->wtextchar[i-1].xp;
        if (vert_spacing_threshold>0)
            {
            double h;

            h=wtcs->wtextchar[i].y2 - wtcs->wtextchar[i].y1;
            if (dy >= h*.75 || dy >= vert_spacing_threshold)
                {
                if (i-1>i0)
                    wtextchar_array_sort_horizontally_by_position(&wtcs->wtextchar[i0],i-i0+1);
                newrow=1;
                }
            }
        else
            {
            if (dx<0 || dy<0)
                newrow=1;
            }
        if (newrow)
            {
            wtextchars_add_one_row(wtcs,i0,i-1,words);
            i0=i;
            }
        }
    wtextchars_add_one_row(wtcs,i0,wtcs->n-1,words);
    }


static void wtextchars_add_one_row(WTEXTCHARS *wtcs,int i0,int i1,OCRWORDS *words)

    {
    int i,j0;
    OCRWORD word;
    int *u16str;
    static char *funcname="wtextchars_add_one_row";

    ocrword_init(&word);
    willus_dmem_alloc_warn(40,(void **)&u16str,(i1-i0+1)*sizeof(int),funcname,10);
    willus_dmem_alloc_warn(41,(void **)&word.text,(i1-i0+2)*sizeof(int),funcname,10);
    willus_dmem_alloc_warn(42,(void **)&word.cpos,(i1-i0+2)*sizeof(double),funcname,10);
    for (j0=i=i0;i<=i1;i++)
        {
        int c,space;
        double h,dx;

        c=wtcs->wtextchar[i].ucs;
        /* If letters are physically separated, assume a space */
        if (i<i1)
            {
            h = fabs((wtcs->wtextchar[i+1].y2 - wtcs->wtextchar[i+1].y1)
                      +(wtcs->wtextchar[i].y2 - wtcs->wtextchar[i].y1)) / 2.;
            dx = wtcs->wtextchar[i+1].x1 - wtcs->wtextchar[i].x2;
            space = (dx > h*.05);
            }
        else
            {
            dx = 0.;
            space = 0;
            }
        /*
        ** Look for ligatured pattern--v2.33 with MuPDF v1.7
        ** Ligature pattern, e.g. "fi"
        ** As of MuPDF 1.7, a ligured "fi" is like this:
        ** f has zero width, followed by space at same position w/zero width,
        ** followed by i at same position, followed by space of zero width.
        */
        if (wtextchars_ligature_pattern(wtcs,i))
            {
            i+=3;
            continue;
            }
        if (space || c==' ' || i==i1)
            {
            if (c==' ' && i==j0)
                j0++;
            else
                {
                int j1,j,ligature,nc;
                double dx;

                j1=(c==' ')?i-1:i;
                word.r=(int)(wtcs->wtextchar[j0].yp*100.+.5);
                word.x0=wtcs->wtextchar[j0].x1;
                word.y0=wtcs->wtextchar[j0].y1;
                word.w0=wtcs->wtextchar[j1].x2-word.x0;
                word.h0=wtcs->wtextchar[j0].y2-word.y0;
                /*
                ** Narrow up the words slightly so that it is physically separated
                ** from other words--this makes selection work better in Sumatra.
                ** Note that only w0 is narrowed--the cpos[] values are not. (v2.20)
                */
                dx = word.h0*.05;
                if (word.w0<dx*4.)
                    dx = word.w0/10.;
                word.x0 += dx;
                word.w0 -= 2*dx;

                word.maxheight=0.;
                for (ligature=0,nc=0,j=j0;j<=j1;j++)
                    {
                    word.cpos[nc]=wtcs->wtextchar[j].x2-word.x0;
                    u16str[nc]=wtcs->wtextchar[j].ucs;
                    nc++;
                    if (wtcs->wtextchar[j].yp-wtcs->wtextchar[j].y1 > word.maxheight)
                        word.maxheight=(wtcs->wtextchar[j].yp-wtcs->wtextchar[j].y1);
                    /* Ligature check--new with MuPDF v1.7 */
                    if (ligature)
                        {
                        word.cpos[nc-1] += (wtcs->wtextchar[j+1].x2-wtcs->wtextchar[j].x2)/2.;
                        ligature=0;
                        j++;
                        }
                    else if (j<j1-2 && wtextchars_ligature_pattern(wtcs,j))
                        {
                        ligature=1;
                        j++;
                        }
                    }
                unicode_to_utf8(word.text,u16str,nc);
                word.n=nc;
                ocrwords_add_word(words,&word);
                j0= i+1;
                }
            }
        }
    willus_dmem_free(42,(double **)&word.cpos,funcname);
    willus_dmem_free(41,(double **)&word.text,funcname);
    willus_dmem_free(40,(double **)&u16str,funcname);
    }

/*
** For MuPDF v1.7 and up
*/
static int wtextchars_ligature_pattern(WTEXTCHARS *wtcs,int index)

    {
    return(index<wtcs->n-3
             && wtcs->wtextchar[index].ucs!=' ' 
             && wtcs->wtextchar[index+1].ucs==' '
             && wtcs->wtextchar[index+2].ucs!=' '
             && wtcs->wtextchar[index+3].ucs==' '
             && fabs(wtcs->wtextchar[index+1].x1-wtcs->wtextchar[index].x1)<.01
             && fabs(wtcs->wtextchar[index+2].x1 - wtcs->wtextchar[index].x1)<.01);
    }
#endif /* HAVE_MUPDF_LIB */