#!/bin/bash
## 2009-07-15:  to allow my manual fixups (during tesseract-training) to be done on the SOURCEFILE form;
## this is the inverse of cvt-tesseract-box-to-source;
##
## boxfile has one char per line, with coord info, eg:
##	a 830 3513 849 3536 0	<--char "a" in a box with coords x:830..849 y:3513..3536 on page:0  <-- from multi-page tiff
##	a 830 3513 849 3536	<--char "a" in a box with coords x:830..849 y:3513..3536            <-- from simple tiff
##
## USAGE EXAMPLE:  cvt-tesseract-source-to-box  pde.img0005.box.latin1   --uses pde.img0005.box-SOURCEFILE.latin1 to revise pde.img0005.box.latin1
##
## Copyright © 2009 Eugene Reimer;  can be used, modified, copied, and distributed or sold under the terms of either the LGPL or the GPL (your choice);
## see http://www.gnu.org/licenses for the details of these terms.

SSTR=;  while read -r;do SSTR="$SSTR $REPLY"; done  <${1/box/box-SOURCEFILE}	##inhale the SOURCEFILE
SSTR="${SSTR}()()()()"								##add some trailing junk, just in case...
getSCHR() {								##function to get SSTR[SPOS] into SCHR, skipping spaces, handling parens (for pairs etc)
  SCHR=${SSTR:$SPOS:1}							##get char
  while [ "$SCHR" == " " ];do ((++SPOS)); SCHR=${SSTR:$SPOS:1}; done	##skip space(s)
##if [ "$SCHR" == "(" ];then SCHR=${SSTR:1+$SPOS:2}; ((SPOS+=3)); fi									##handle pair in parens
  if [ "$SCHR" == "(" ];then ((B=SPOS+1,E=B)); while [[ ${SSTR:$E:1} != ")" ]];do ((++E));done; SCHR=${SSTR:$B:$E-$B}; ((SPOS=E)); fi	##handle string in parens
##if [ "$SCHR" == "~" ];then SCHR=""; fi				##==YANK this line if charset includes tilde
}
SPOS=0; getSCHR								##init SPOS & SCHR to first char of SSTR

((XHIprv=999999,Y=999999))
while read -r BCHR XLO YLO XHI YHI PAGENBR;do				##read line of boxfile;  vars get empty-string if omitted (all if empty-line; PAGENBR if simple tiff)
  if [ "$BCHR" != "$SCHR" ];then echo "==revising $BCHR-->$SCHR" >&2;fi	##msg about fixup
  if [ "$SCHR" == ""      ];then echo "==empty-line at $Y"       >&2;fi	##msg about empty-line, may scrap??
  if [ "$SCHR" == "" ];then echo ""					##==write empty-line for a deletion
  else                      echo "$SCHR $XLO $YLO $XHI $YHI $PAGENBR"	##write revised boxfile-line
  fi
  ((++SPOS)); getSCHR							##advance to next char of SSTR
  if [[ $SCHR ]];then ((XHIprv=XHI,Y=YLO)); fi
done  <$1  >tmp$$-$1							##loop reads $1, writes tmp$$-$1
touch -r$1  tmp$$-$1							##preserve timestamp
mv -bfv tmp$$-$1 $1			  				##rename;  the tmp$$-$1 output replaces $1  (mv -b, rather than 2 mv-cmds, for one-line msg)


exit
CHANGE-LOG:
===========
TESTING:  converted pde.img0005.box-EDITED.latin1 to pde.img0005.box-SOURCEFILE-EDITED.latin1;  after correcting it, will use this script on that pair...
	have corrected it, using the new "(AB)" notation for pairs;
	note: cvt-tesseract-box-to-source lost some space chars at CONS:23;  but works properly at CONS:20;
not as easy as I'd hoped;  on page 0005, I'm getting box-errors, two chars in one box, eg: ft-->ü  wo-->m
Consider using SPACE to "resync" at end of each word??
Support "(AB)" character-pairs -- just pairs was sufficient for my pde cases;
Added discarding of tilde-lines, after further testing on all 10 test-images;
--NOTE: for general use, have switched to using () rather than tilde--!!--

2009-07-22: support arbitrary-length parenthecized-strings in SOURCEFILE;
	support "()" in SOURCEFILE for a deletion;  
	support & produce empty-line in boxfile for a deletion (so reapplying this source-to-box will work);
	briefly kept tilde as synonym for "()" deletion;  have since scrapped;
	==note:  may get sick of tesseract-warnings due to empty-lines??

2009-08-02: preseve timestamps, using touch -r;