I'm writing a program that uses gsed to extract multibyte charactors from csv file.
It works well with csv file encoded UTF-8, but it doesn't work with csv file encoded SHIFT_JIS.
test % cat sjis_sample.csv | iconv -f shift_jis -t utf-8
"こんにちは","hello"%
test % cat sjis_sample.csv | iconv -f shift_jis -t utf-8 | gsed -r 's/"(.*)","(.*)"/\1 \2/'
こんにちは hello%
test % cat sjis_sample.csv | gsed -r 's/"(.*)","(.*)"/\1 \2/' | iconv -f shift_jis -t utf-8
"こんにちは","hello"%
LINE 1：
Read file with UTF-8
LINE 2：
Extracted text contents from csv file after converting encoding from SHIFT_JIS to UTF-8
-> Works well
LINE 3：
Extracted text contents from csv file without converting encoding
-> It seems that `gsed` failed to capture text contents with match pattern.
Does anybody know how to use gsed for SHIFT_JIS encoded file?
Thank you.
% gsed --version
gsed (GNU sed) 4.8
Copyright (C) 2020 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <https://gnu.org/licenses/gpl.html>.
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.
Written by Jay Fenlason, Tom Lord, Ken Pizzini,
Paolo Bonzini, Jim Meyering, and Assaf Gordon.
This sed program was built without SELinux support.
GNU sed home page: <https://www.gnu.org/software/sed/>.
General help using GNU software: <https://www.gnu.org/gethelp/>.
E-mail bug reports to: <bug-sed#gnu.org>.
test % locale
LANG="ja_JP.UTF-8"
LC_COLLATE="ja_JP.UTF-8"
LC_CTYPE="ja_JP.UTF-8"
LC_MESSAGES="ja_JP.UTF-8"
LC_MONETARY="ja_JP.UTF-8"
LC_NUMERIC="ja_JP.UTF-8"
LC_TIME="ja_JP.UTF-8"
LC_ALL=
Solved
Thanks to #KamilCuk
GNU sed is locale aware. If you want to work with raw bytes (ie. you can check what bytes represent " in Shift_JIS and feed that to sed) use:
LC_ALL=C sed ....
I set LANG instead of LC_ALL as C because I could not set LC_ALL as C.
test % cat sjis_convert.sh
#!/bin/bash
LANG=C
cat sjis_sample.csv |\
gsed -r 's/"(.*)","(.*)"/\1 \2/' |\
iconv -f shift_jis -t utf-8
test % ./sjis_convert.sh
こんにちは hello%
Appendix
I could not set C to LC_ALL.
test % cat sjis_convert.sh
#!/bin/bash
LC_ALL=C
locale
echo ''
cat sjis_sample.csv |\
gsed -r 's/"(.*)","(.*)"/\1 \2/' |\
iconv -f shift_jis -t utf-8
echo ''
locale
test % ./sjis_convert.sh
LANG="ja_JP.UTF-8"
LC_COLLATE="ja_JP.UTF-8"
LC_CTYPE="ja_JP.UTF-8"
LC_MESSAGES="ja_JP.UTF-8"
LC_MONETARY="ja_JP.UTF-8"
LC_NUMERIC="ja_JP.UTF-8"
LC_TIME="ja_JP.UTF-8"
LC_ALL=
"こんにちは","hello"
LANG="ja_JP.UTF-8"
LC_COLLATE="ja_JP.UTF-8"
LC_CTYPE="ja_JP.UTF-8"
LC_MESSAGES="ja_JP.UTF-8"
LC_MONETARY="ja_JP.UTF-8"
LC_NUMERIC="ja_JP.UTF-8"
LC_TIME="ja_JP.UTF-8"
LC_ALL=
Instead, I set C to LANG and it worked.
test % cat ./sjis_convert.sh
#!/bin/bash
LANG=C
locale
echo ''
cat sjis_sample.csv |\
gsed -r 's/"(.*)","(.*)"/\1 \2/' |\
iconv -f shift_jis -t utf-8
echo ''
locale
test % ./sjis_convert.sh
LANG="C"
LC_COLLATE="C"
LC_CTYPE="C"
LC_MESSAGES="C"
LC_MONETARY="C"
LC_NUMERIC="C"
LC_TIME="C"
LC_ALL=
こんにちは hello
LANG="C"
LC_COLLATE="C"
LC_CTYPE="C"
LC_MESSAGES="C"
LC_MONETARY="C"
LC_NUMERIC="C"
LC_TIME="C"
LC_ALL=
GNU sed is locale aware. If you want to work with raw bytes (ie. you can check what bytes represent " in Shift_JIS and feed that to sed) use:
LC_ALL=C sed ....
If you want to work with UTF-8, set UTF-8 locale, which most probably is your default:
LC_ALL=en_US.UTF-8 sed ...
And if you want to work with any other locale, tell it to sed:
LC_ALL=ja_JP.Shift_JIS sed ...
The hexdump command converts any file to hex values.
But what if I have hex values and I want to reverse the process, is this possible?
There is a similar tool called xxd. If you run xxd with just a file name it dumps the data in a fairly standard hex dump format:
# xxd bdata
0000000: 0001 0203 0405
......
Now if you pipe the output back to xxd with the -r option and redirect that to a new file, you can convert the hex dump back to binary:
# xxd bdata | xxd -r >bdata2
# cmp bdata bdata2
# xxd bdata2
0000000: 0001 0203 0405
I've written a short AWK script which reverses hexdump -C output back to the
original data. Use like this:
reverse-hexdump.sh hex.txt > data
Handles '*' repeat markers and generating original data even if binary.
hexdump -C and reverse-hexdump.sh make a data round-trip pair. It is
available here:
GitHub reverse-hexdump repo
Direct to reverse-hexdump.sh
Restore file, given only the output of hexdump file
If you only have the output of hexdump file and want to restore the original file, first note that hexdump's default output depends on the endianness of the system you ran hexdump on!
If you have access to the system that created the dump, you can determinate its endianness using below command:
[[ "$(printf '\01\03' | hexdump)" == *0103* ]] && echo big || echo little
Reversing little-endian hexdump
This is the most common case. All x86/x64 systems are little-endian. If you don't know the endianness of the system that ran hexdump file, try this.
sed 's/ \(..\)\(..\)/ \2\1/g;$d' dump | xxd -r
The sed part converts hexdump's format into xxd's format, at least so far that xxd -r works.
Reversing big-endian hexdump
sed '$d' dump | xxd -r
Known Bugs (see comment section)
A trailing null byte is added if the original file was of odd length (e.g. 1, 3, 5, 7, ..., byte long).
Repeating sections of the original file are not restored correctly if they were hexdumped using a *.
You can check your dump for above problematic cases by running below command:
grep -qE '^\*|^[0-9a-f]*[13579bdf] *$' dump && echo bug || echo ok
Better alternative to create hexdumps in the first place
Besides the non-posix (and therefore not so portable) xxd there is od (octal dump) which should be available on all unix-like systems as it is specified by posix:
od -tx1 -An -v
Will print a hexadecimal dump, grouping digits as single bytes (-tx1), with no Address prefixes (-An, similar to xxd -p) and without abbreviating repeated sections as * (-v). You can reverse such a dump using xxd -r -p.
As someone who sucks at bash, I could not understand the examples already posted.
Here is what would have helped me when I was originally searching:
Take your text file "AYE.TXT" and convert it into a hex dump called "BEE.TXT"
xxd -p "AYE.TXT" > "BEE.TXT"
Take your hex dump file ("BEE.TXT") and covert it back to ascii file "CEE.TXT"
xxd -r -p "BEE.TXT" > "CEE.TXT"
Now that you have some simple working code, feel free to check out
"xxd -help" on the command line for an explanation of what all those flags do.
(That part is the easy part, the hard part is the specifics of the bash syntax)
There is a tonne of more elegant ways to get this done, but I've quickly hacked something together that Works for Me (tm) when regenerating a binary file from a hex dump generated by hexdump -C some_file.bin:
sed 's/\(.\{8\}\) \(..\) \(..\) \(..\) \(..\) \(..\) \(..\) \(..\) \(..\)/\1: \2\3 \4\5 \6\7 \8\9/g' some_file.hexdump | sed 's/\(.*\) \(..\) \(..\) \(..\) \(..\) \(..\) \(..\) \(..\) \(..\) |/\1 \2\3 \4\5 \6\7 \8\9 /g' | sed 's/.$//g' | xxd -r > some_file.restored
Basically, uses 2 sed processeses, each handling it's part of each line. Ugly, but someone might find it useful.
If you don't have xxd, use hexdump, od, perl or python:
The following all give the same output:
# If you only have hexdump
hexdump -ve '1/1 "%.2x"' mybinaryfile > mydump
# This gives exactly the same output as:
xxd -p mybinaryfile > mydump
# Or, much slower:
od -v -t x1 -An < mybinaryfile | tr -d "\n " > mydump
# Or, the fastest:
perl -pe 'BEGIN{$/=\1e6} $_=unpack "H*"' < mybinaryfile > mydump
# Or, if you somehow have Python, and not Perl:
python -c "print(open('mybinaryfile','rb').read().hex())" > mydump
Then you can copy and paste, or pipe the output, and convert back with:
xxd -r -p mydump mybinaryfileagain
# Or
xxd -r -p < mydump > mybinaryfileagain
The hexdump command is available almost everywhere, and is usually part of the default busybox - if it's not linked, you can try running busybox hexdump or busybox xxd.
If xxd is not available to reverse the data, then you can try awk
The old days: Zmodem
In the old days we used to use X/Y/Zmodem which is available in the package lrzsz which can tolerate lossy comms - but it's a bidirectional protocol so the binaries need to be running at the same time and there needs to be bidirectional comms:
# Demo on local machine, using FIFOs
mkfifo /tmp/fifo-in
mkfifo /tmp/fifo-out
sz -b mybinaryfile > /tmp/fifo-out < /tmp/fifo-in
mkdir out; cd out
rz -b < /tmp/fifo-out > /tmp/fifo-in
Luckily, screen supports receiving Zmodem, so if you're in a screen session:
screen
telnet somehost
Then type Ctrl+A and : and then zmodem catch and Enter. Then inside the screen on the remote host, issue:
# sz -b mybinaryfile
Press Enter when you see the string starting with "!!!".
When you see "Transfer Complete", you may want to run reset if you want to continue the terminal session normally.
This program reverses hexdump -C output back to the original data.
Usage:
make
make test
./unhexdump -i inputfile -o outputfile
see https://github.com/zhouzq-thu/unhexdump!
i found more simple solution:
bin2hex
echo -n "abc" | hexdump -ve '1/1 "%02x"'
hex2bin
echo -n "616263" | grep -Eo ".{2}" | sed 's/\(.*\)/\\x\1/' | tr -d '\n' | xargs -0 echo -ne
I have done this command
iconv -f us-ascii -t SHIFT_JIS qpr_specify_item_def.csv > specify_item_def.csv
but is getting an exception
iconv: illegal input sequence at position 548
I have an UTF-16 encoded file and I want replace UNIX line endings with Windows line endings. I don't want to touch anything else.
Is there a linux command line tool that can search for two bytes "0A 00" and replace it with four bytes "0D 00 0A 00"?
Perl to the rescue:
perl -we 'binmode STDIN, ":encoding(UTF-16le)";
binmode STDOUT, ":encoding(UTF-16le):crlf";
print while <STDIN>;
' < input.txt > output.txt
You may use unix2dos, but you have to convert the file to a 8-bit encoding before, and back to UTF-16 after. The obvious intermediate candidate is UTF-8:
$ cat in.txt | iconv -f UTF-16 -t UTF-8 | unix2dos | iconv -f UTF-8 -t UTF-16 > out.txt
You can wrap these three piped commands in a handy script, if you wish.
#/bin/sh
iconv -f UTF-16 -t UTF-8 | unix2dos | iconv -f UTF-8 -t UTF-16
unix2dos is what you're looking for. See its different options to find the one that's right for your UTF-16 encoding.
Solution:
perl -pe "BEGIN { binmode $_, ':raw:encoding(UTF-16LE)' for *STDIN, *STDOUT }; s/\n\0/\r\0\n\0/g;" < input.file > output.file
Credit to my coworker Manu and Stream-process UTF-16 file with BOM and Unix line endings in Windows perl
Hi I am trying to convert some log files from a Microsoft SQL server, but the files are encoded using UTf-16LE and iconv does not seem to be able to convert them.
I am doing:
iconv -f UTF-16LE -t UTF-8 <filename>
I also tried to delete any carriage returns from the end of the line if there are any, but that did not fix it either. If I save it using gedit that works, but this is not a viable solution since I have hundreds of those files.
EDIT: Please see the new answer for the missing option
I forgot the -o switch!
The final command is :
iconv -f UTF-16LE -t UTF-8 <filename> -o <new-filename>
The command you specified will output to stdout. You can either use the -o parameter, or redirect your output:
with -o:
iconv -f UTF-16LE -t UTF-8 infile -o outfile
with piping:
iconv -f UTF-16LE -t UTF-8 infile > outfile
Both will yield the desired result.
However some versions of iconv (v1 on macOS for example) do not support the -o parameter and you will see that the converted text is echoed to stdout. In that case, use the piping option.