123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171 |
- #!/bin/sh
- #
- # makelang.sh - multi-language support script
- # for generating lang_xx.bin (secondary language binary file)
- #
- # Input files:
- # lang_en.txt
- # lang_en_xx.txt
- #
- # Output files:
- # lang_en.tmp (temporary, will be removed when finished)
- # lang_en_xx.tmp ==||==
- # lang_en_xx.dif ==||==
- # lang_xx.txt
- #
- #
- # Selected language:
- LNG=$1
- if [ -z "$LNG" ]; then LNG='cz'; fi
- #
- #
- finish()
- {
- if [ $1 -eq 0 ]; then
- if [ -e lang_en.tmp ]; then rm lang_en.tmp; fi
- if [ -e lang_en_$LNG.tmp ]; then rm lang_en_$LNG.tmp; fi
- if [ -e lang_en_$LNG.dif ]; then rm lang_en_$LNG.dif; fi
- fi
- # echo >&2
- if [ $1 -eq 0 ]; then
- echo "make_lang.sh finished with success" >&2
- else
- echo "make_lang.sh finished with errors!" >&2
- fi
- exit $1
- }
- make_lang()
- {
- LNG=$1
- echo "make_lang.sh started" >&2
- echo "selected language=$LNG" >&2
- #check if input files exists
- echo -n " checking input files..." >&2
- if [ ! -e lang_en.txt ]; then echo "NG! file lang_en.txt not found!" >&2; exit 1; fi
- if [ ! -e lang_en_$LNG.txt ]; then echo "NG! file lang_en_$LNG.txt not found!" >&2; exit 1; fi
- echo "OK" >&2
- #filter comment and empty lines from key and dictionary files, create temporary files
- echo -n " creating tmp files..." >&2
- cat lang_en.txt | sed "/^$/d;/^#/d" > lang_en.tmp
- cat lang_en_$LNG.txt | sed "/^$/d;/^#/d" > lang_en_$LNG.tmp
- echo "OK" >&2
- #cat lang_en_$LNG.tmp | sed 'n;d' >test1.txt
- #compare files using diff and check for differences
- echo -n " comparing tmp files..." >&2
- if ! cat lang_en_$LNG.tmp | sed 'n;d' | diff lang_en.tmp - > lang_en_$LNG.dif; then
- echo "NG!" >&2
- echo "Entries in lang_en_$LNG.txt are different from lang_en.txt!" >&2
- echo "please check lang_en_$LNG.dif" >&2
- finish 1
- fi
- echo "OK" >&2
- #generate lang_xx.txt (secondary language text data sorted by ids)
- echo -n " generating lang_$LNG.txt..." >&2
- cat lang_en_$LNG.tmp | sed '1~2d' | sed "s/^\"\\\\x00/\"/" > lang_$LNG.txt
- echo "OK" >&2
- #generate lang_xx.dat (secondary language text data in binary form)
- echo -n " generating lang_$LNG.dat..." >&2
- cat lang_$LNG.txt | sed "s/\\\\/\\\\\\\\/g" | while read s; do
- s=${s#\"}
- s=${s%\"}
- /bin/echo -e -n "$s\x00"
- done >lang_$LNG.dat
- echo "OK" >&2
- #calculate variables
- lt_magic='\xa5\x5a\xb4\x4b'
- lt_count=$(grep -c '^' lang_$LNG.txt)
- lt_data_size=$(wc -c lang_$LNG.dat | cut -f1 -d' ')
- lt_offs_size=$((2 * $lt_count))
- lt_size=$((16 + $lt_offs_size + $lt_data_size))
- lt_chsum=0
- lt_code='\xff\xff'
- lt_resv1='\xff\xff\xff\xff'
- case "$LNG" in
- *en*) lt_code='\x6e\x65' ;;
- *cz*) lt_code='\x73\x63' ;;
- *de*) lt_code='\x65\x64' ;;
- *es*) lt_code='\x73\x65' ;;
- *it*) lt_code='\x74\x69' ;;
- *pl*) lt_code='\x6c\x70' ;;
- esac
- #generate lang_xx.ofs (secondary language text data offset table)
- echo -n " generating lang_$LNG.ofs..." >&2
- cat lang_$LNG.txt | sed "s/\\\\x[0-9a-f][0-9a-f]/\./g;s/\\\\[0-7][0-7][0-7]/\./g" |\
- awk 'BEGIN { o='$((16 + $lt_offs_size))';} { printf("%d\n",o); o+=(length($0)-1); }' > lang_$LNG.ofs
- echo "OK" >&2
- #generate lang_xx.bin (secondary language result binary file)
- echo " generating lang_$LNG.bin:" >&2
- #create empty file
- dd if=/dev/zero of=lang_$LNG.bin bs=1 count=$lt_size 2>/dev/null
- #awk code to format ui16 variables for dd
- awk_ui16='{ h=int($1/256); printf("\\x%02x\\x%02x\n", int($1-256*h), h); }'
- #write data to binary file with dd
- echo -n " writing header (16 bytes)..." >&2
- /bin/echo -n -e "$lt_magic" |\
- dd of=lang_$LNG.bin bs=1 count=4 seek=0 conv=notrunc 2>/dev/null
- /bin/echo -n -e $(echo -n "$lt_size" | awk "$awk_ui16") |\
- dd of=lang_$LNG.bin bs=1 count=2 seek=4 conv=notrunc 2>/dev/null
- /bin/echo -n -e $(echo -n "$lt_count" | awk "$awk_ui16") |\
- dd of=lang_$LNG.bin bs=1 count=2 seek=6 conv=notrunc 2>/dev/null
- /bin/echo -n -e $(echo -n "$lt_chsum" | awk "$awk_ui16") |\
- dd of=lang_$LNG.bin bs=1 count=2 seek=8 conv=notrunc 2>/dev/null
- /bin/echo -n -e "$lt_code" |\
- dd of=lang_$LNG.bin bs=1 count=2 seek=10 conv=notrunc 2>/dev/null
- /bin/echo -n -e "$lt_resv1" |\
- dd of=lang_$LNG.bin bs=1 count=4 seek=12 conv=notrunc 2>/dev/null
- echo "OK" >&2
- echo -n " writing offset table ($lt_offs_size bytes)..." >&2
- /bin/echo -n -e $(cat lang_$LNG.ofs | awk "$awk_ui16" | tr -d '\n'; echo) |\
- dd of=./lang_$LNG.bin bs=1 count=$lt_offs_size seek=16 conv=notrunc 2>/dev/null
- echo "OK" >&2
- echo -n " writing text data ($lt_data_size bytes)..." >&2
- dd if=./lang_$LNG.dat of=./lang_$LNG.bin bs=1 count=$lt_data_size seek=$((16 + $lt_offs_size)) conv=notrunc 2>/dev/null
- echo "OK" >&2
- #update signature
- echo -n " updating signature..." >&2
- dd if=lang_en.bin of=lang_$LNG.bin bs=1 count=4 skip=6 seek=12 conv=notrunc 2>/dev/null
- echo "OK" >&2
- #calculate and update checksum
- lt_chsum=$(cat lang_$LNG.bin | xxd | cut -c11-49 | tr ' ' "\n" | sed '/^$/d' | awk 'BEGIN { sum = 0; } { sum += strtonum("0x"$1); if (sum > 0xffff) sum -= 0x10000; } END { printf("%x\n", sum); }')
- /bin/echo -n -e $(echo -n $((0x$lt_chsum)) | awk "$awk_ui16") |\
- dd of=lang_$LNG.bin bs=1 count=2 seek=8 conv=notrunc 2>/dev/null
- echo " lang_table details:" >&2
- echo " lt_count = $lt_count" >&2
- echo " lt_size = $lt_size" >&2
- echo " lt_chsum = $lt_chsum" >&2
- }
- echo $LNG
- if [ "$LNG" = "all" ]; then
- ./lang-build.sh en
- make_lang cz
- make_lang de
- make_lang es
- make_lang it
- make_lang pl
- exit 0
- else
- make_lang $LNG
- fi
- finish 0
|