APEX

I have to admit that I do not know how to determine if some bytes constitute a legitimate UTF-8 value, perhaps there is something in Character that might help.

However this if statement can't be what you want since as far as I can tell, it can never be true.

if(fileContent<0 && fileContent[i]!=10 && fileContent[i]!=13 && fileContent[i]>300)


What single value will satisfy the first and third conditions?

Edited by: johndjr on Oct 23, 2009 8:26 AM                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                

Sanath,

It is my considered opinion that you're in way over your head. That code has some really basic noob mistakes.

Pop quiz:

1. For what value(s) of x is this statement true (x<0 && x>300)? Putz!

2. How many bytes (a signed 8-bit integer value) exceed 300?

3. If you read the API doco for File.length() you'll find that it specifically warns against using it to size a byte-array to hold the whole files contents. Why do you imagine that you are special (and therefore this technique will work reliably for you anyways?

4. How where you planning to determine the size of each character by reading bytes? Whitefella mojo maybe? You might also want to try a nice Séance, but I don't much like your chances there either.

5. Have you ever considered a career in the armed services?

Cheers. Keith.

301 bytes exceed 300 bytes!

My challenge is to spot the multi-byte character hidden in this big dat file.
This is because the data entry team asked me to spot out the record and column that has issue out of
lakhs of records they sent inside this file.

Lets have the validation code like this...

   if( (fileContent<0 && fileContent[i]!=10 && fileContent[i]!=13) || fileContent[i]>300) // if invalid

{System.out.println("failure at position: "+i);break;}
< 0 - As I tested, some chars generated -ve values for some codes.
300 - was a try to find out if any characters exceeds actual chars. range.
10 and 13 are for line-feed.

any alternative (better code ofcourse) way to catch this black sheep ?                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            

My challenge is to spot the multi-byte character hidden in this big dat file.
This is because the data entry team asked me to spot out the record and column that has issue out of
lakhs of records they sent inside this file.

Lets have the validation code like this...

   if( (fileContent<0 && fileContent[i]!=10 && fileContent[i]!=13) || fileContent[i]>300) // if invalid

{System.out.println("failure at position: "+i);break;}
lessthan 0 - I saw some -ve values when I was testing with other files.
greaterthan 300 - was a try to find out if any characters exceeds actual chars. range.
if 10 and 13 are for line-feed.

with this, I randomly placed chinese, korean characters and program found them.
any alternative (better code ofcourse) way to catch this black sheep ?

Edited by: Sanath_K on Oct 23, 2009 8:06 PM                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            

6. How much data do you think you've read when you do this:

fin.read(fileContent);

You might have read as little as one byte, meaning you're skimming over all but one byte of the file.

from right-click, file, properties , I found size : 12512196 bytes
same is the response from file.length, byte array size before for loop and finally the value I am printing to verify, i.e. totalcharacters.
from this, I felt it is ok to goahead for checking each byte value as aim is to locate the first speacial character.

If this is a disposable program, fine. But you're doing it wrong. And you still have serious issues with how you actually read in data. Namely, you look through every byte of fileContent but it's more than likely that only the first few bytes actually contain data from your file.

If you want the entire contents of the file in a byte array, here's how you can do it:

FileInputStream fin;
ByteArrayOutputStream baos = new ByteArrayOutputStream();
int len;
byte[] buf = new byte[1024];
while ( (len = fin.read(buf)) != -1 ) {
   baos.write(buf, 0, len);
}

byte[] fileContents = baos.toByteArray();

But you're probably fine looking at it chunk-by-chunk.

In fact, if you're only interested in doing it byte-by-byte, just do this:

BufferedInputStream bin = new BufferedInputStream(fin);
for (int b = -1; (b = bin.read()) != -1; ) {
  //deal with this byte
}

Edited by: endasil on 23-Oct-2009 11:36 AM

lot of helpful comments on the logic...thanks.
question still hunts...
is there any way we can seek help from java code to locate multi-byte characters in a dat file ?
I hope to help the data-entry team to rectify the error and re-process the file.

This is UTF-8 encoded text? Look at each byte. If the high bit is set, it's a participant in a multi-byte character. [See here|http://en.wikipedia.org/wiki/UTF-8#Description]. tschodt tells you how to check for this in a previous reply.

If you want to scan a sequence of bytes from a file and identify locations where a subsequence of two or more bytes would represent a UTF-8 character which isn't in the ASCII character set, then the simplest way is to ask Java to do the UTF-8 part:

Reader r = new InputStreamReader(new FileReader(file), "UTF-8");
int character = 0;
while ((character = r.read()) >= 0) {
  // here we have a stream of characters decoded using UTF-8
  if (character > 127) {
    // this one isn't ASCII
  }
}

Assuming the file contains valid UTF-8.
If the file contains [invalid byte sequences|http://en.wikipedia.org/wiki/Utf-8#Invalid_byte_sequences] you get UTFDataFormatException.

That's true, and certainly a possibility. But we don't know whether that's the OP's problem. All we have is some guff about "multi-byte" characters. If I were doing this -- well I wouldn't be doing this because I would get around to asking the right questions -- I would start with that, then if it threw an exception I would change it to count the number of characters read before the exception was thrown.

Erm?

Think he was thinking InputStream.available(). I wouldn't use File.length either way, because then you can't use it against pipes, etc.

I don't see one.

But it is nothing more than obvious that you shouldn't declare a bytearray whose length is exact the file's length. That's lot of fun if the file length exceeds the available JVM heap memory.

It's not the byte array that's the problem in that case, but rather the fact that you're gong to read the whole file. However, if you do decide to read the whole file, and if you know it's a regular file, not a pipe or something, and if you can assume that the size won't change while you're reading it, then declaring a byte[] of exactly the file's length would be a good way to do it.

It's not the way I'd normally read a file, but I wouldn't rule it out.

Even more fun if the file length exceeds the maximum value of an integer and it gets truncated to fit in an integer, which you then use as the size of your array.

Any such problem would of course not be caused by a character but rather by a character set.

Although it is possible that an Oracle driver has a bug in it, the Oracle drivers have been handling character set conversions for years.
Naturally if you do not set up the driver correctly then it will cause a problem.

That of course is ridiculous.

Bytes are encoded to represent a character set. If you have a data file with text in it then it must be because the data file has at least one and perhaps more character sets.

Attempting to detemine the character set of a file is generally a non-deterministic problem - a computer cannot determine the answer all the time.
It will definitely not be able to do it with a misplaced single character from another character set.

So your problem is actually one of the following
- Determine what the character set is rather than using the wrong one.
- Attempt to scrub the incoming data to remove garbage (because that is what wrong character set characters would be) from the data before attempting to insert it into the database. And provide enough error detection that problems can be dealt with manually. In this case you are NOT attempting to recognize characters from another character set but rather excluding anything that doesn't fit into the set that you are using.
- Make the source of the file start producing a file in a single/correct character set.

EPIC FAIL ;-)

I'm stealing that word. It's just so... ummm.... "pithy"!

Folks,

Combining the just-read-single-bytes technique suggested by endasil with the bit-twiddling-highorder-test suggested by tschodt I get:

(*NOTE:* The source code is it's own test-data. It contains two "extended characters" in the comment, and therefore must be saved in a UTF-8 (or equivalent) encoded file, and compiled using [javac's|http://www.manpagez.com/man/1/javac/] -encoding UTF8 argument.)

ExtendedCharacterDetector.java

package forums;

import java.io.FileInputStream;
import java.io.BufferedInputStream;

// ÊPIC FÃIL!

public class ExtendedCharacterDetector
{
  public static void main(String[] args) {
    String filename = args.length>0 ? args[0] : "ExtendedCharacterDetector.java";
    try {
      BufferedInputStream input = null;
      try {
        input = new BufferedInputStream(new FileInputStream(filename));
        int count = 0;
        for ( int i=0,b=-1; (b=input.read()) != -1; ++i ) {
          if ( (b&0x80) != 0 ) {
            System.out.println("byte "+i+" is "+b);
            count++;
          }
        }
        System.out.println("Number of bytes exceeding "+0x80+" = "+count);
      } finally {
        if(input!=null)input.close();
      }
    } catch (Exception e) {
      e.printStackTrace();
    }
  }
}

output

byte 94 is 195
byte 95 is 138
byte 101 is 195
byte 102 is 131
Number of bytes exceeding 128 = 4

This still isn't very useful though, is it? Bytes 94 and 95 are "busted"... Goodo! So WTF is byte 94? I suppose you could download a [free HEX editor|http://www.google.com/search?q=free+hex+editor] and use that to spy-out the offending characters... (I use NEO, it works).

So... presuming that you know what encoding the file is supposed to be in... I am still of the humble opinion that your users would be better-served if you read characters from the file, and reported character offsets and values to the user... the "extendedness test" remains logically the same... extended (non 7-bit-ascii) characters have a value exceeding 128 (2^7), and this (AFAIK) the same no matter which charset the character has been encoded in, because all modern charsets use the same "the basic ascii table" (i.e. code points <=128).

Cheers. Keith.

It occurs to me that the underlying codecs, defined in java.nio.charset or thereabouts, work between ByteBuffer and CharacterBuffer, so if you put your data file contents into a ByteBuffer (or map the file to a ByteBuffer) and run the decoder directly it should leave the buffer pointers at the point of failure.

Could save you a lot of bit twidling.

But I'm not sure how knowing the offset of the problem is is going to help you. The likely cause is that you've somehow told the middleware that this text is UTF-8 when it's in some other encoding, and byte offsets won't help you there.

In other words you probably have a configuration problem.

Laks is Dutch for "extremely negligent".

Just like the English word lax only more emphatic.

As DBA suggested, the file was extracted with the the select query, where CONVERT (column_name, 'UTF8', 'WE8ISO8859P1') was used and this resolved the issue in file processing. It was done for all the description columns since we couldn't spot the exact field.

I wonder why considering that you had so many bugs in your original code.

Concern from the quality standpoint is that using convert function for all columns instead of finding out just one record's field where data could possibly be corrected is a much time consuming process.

Thanks for all the valuable responses.