On 01/24/2011 10:57 AM, Geoffrey Myers wrote:
> Adrian Klaver wrote:
>> On 01/24/2011 09:16 AM, Geoffrey Myers wrote:
>>
>>>
>>> We hope to identify the characters and fix them in the existing
>>> database, then convert. It appears to be very limited, but it would help
>>> if there was some way to identify these characters outside of simply
>>> doing the reload of the data and finding the errors.
>>>
>>> Hence the reason I asked about a resource that might identify the
>>> characters.
>>
>> The problem is that from the standpoint of the SQL_ASCII database
>> there is nothing wrong with the characters per se. AFAIK there is no
>> built in function to validate characters. The reason is that valid is
>> determined by the encoding and if you know the encoding then you
>> really don't need to determine validity. If you want to see one way
>> others have tackled this, search on iconv in the mailing list archive.
>> This requires working on an external copy of the data and knowing
>> something about the encodings involved. The nearest I could ever find
>> to an encoding detector is:
>>
>> http://chardet.feedparser.org/
>>
>> It is a Python program and the encodings it detects are limited but it
>> might work for you.
>>
>> Given all the above, when I was faced with the problem you are facing
>> I found it easiest to make an educated guess as to the original
>> encoding and then do test restores with client_encoding set to my guess.
>
> Understood. We had figured the problem to be small, and it appears it is
> and thus felt we could address it a character at a time. Then get this
> error:
>
> pg_restore: [archiver (db)] Error from TOC entry 5258; 0 17549 TABLE
> DATA fax postgres
> pg_restore: [archiver (db)] COPY failed: ERROR: invalid byte sequence
> for encoding "UTF8": 0xe28053
>
> That hex value doesn't translate to a single character. I've dumped the
> data to a file as you suggested, but reviewing the identified line
> brings no joy.
>
The only thing I can think of is to use iconv like:
iconv -c -t utf8 -f utf8 -o converted_txt.txt 'original.txt'
where original.txt is your plain text data dump. The -c switch causes
iconv not to convert any illegal characters.
You could then run a diff against converted_txt.txt and 'original.txt'
to see what characters in the original text are causing the problem.
--
Adrian Klaver
adrian.klaver@gmail.com