catch UnicodeDecodeError

[jaroslav.dobrek]

Hello,

very often I have the following problem: I write a program that processes many files which it assumes to be encoded in utf-8. Then, some day, I there is a non-utf-8 character in one of several hundred or thousand (new) files.The program exits with an error message like this:

UnicodeDecodeError: 'utf8' codec can't decode byte 0xe4 in position 60: invalid continuation byte

I usually solve the problem by moving files around and by recoding them.

What I really want to do is use something like

try:
# open file, read line, or do something else, I don't care
except UnicodeDecodeError:
sys.exit("Found a bad char in file " + file + " line " + str(line_number)

Yet, no matter where I put this try-except, it doesn't work.

How should I use try-except with UnicodeDecodeError?

Jaroslav

 

[Andrew Berg]

What I really want to do is use something like

try:
# open file, read line, or do something else, I don't care
except UnicodeDecodeError:
sys.exit("Found a bad char in file " + file + " line " + str(line_number)

Yet, no matter where I put this try-except, it doesn't work.

How should I use try-except with UnicodeDecodeError?

The same way you handle any other exception. The traceback will tell you
the exact line that raised the exception. It helps us help you if you
include the full traceback and give more detail than "it doesn't work".

 

[Philipp Hagemeister]

Hi Jaroslav,

you can catch a UnicodeDecodeError just like any other exception. Can
you provide a full example program that shows your problem?

This works fine on my system:


import sys
open('tmp', 'wb').write(b'\xff\xff')
try:
buf = open('tmp', 'rb').read()
buf.decode('utf-8')
except UnicodeDecodeError as ude:
sys.exit("Found a bad char in file " + "tmp")


Note that you cannot possibly determine the line number if you don't
know what encoding the file is in (and what EOL it uses).

What you can do is count the number of bytes with the value 10 before
ude.start, like this:

lineGuess = buf[:ude.start].count(b'\n') + 1

- Philipp

it doesn't work

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)

iEYEAREKAAYFAlAP2e8ACgkQ9eq1gvr7CFxjIgCfZDryZu+HIQl4wSfH62sAEJl/
IlgAoJUqLDDWYZREqYe9O5PKYdlsMBki
=cGOq
-----END PGP SIGNATURE-----

 

[jaroslav.dobrek]

Hi Jaroslav,

you can catch a UnicodeDecodeError just like any other exception. Can
you provide a full example program that shows your problem?

This works fine on my system:


import sys
open('tmp', 'wb').write(b'\xff\xff')
try:
buf = open('tmp', 'rb').read()
buf.decode('utf-8')
except UnicodeDecodeError as ude:
sys.exit("Found a bad char in file " + "tmp&quot

Thank you. I got it. What I need to do is explicitly decode text.

But I think trial and error with moving files around will in most cases be faster. Usually, such a problem occurs with some (usually complex) program that I wrote quite a long time ago. I don't like editing old and complex programs that work under all normal circumstances.

What I am missing (especially for Python3) is something like:

try:
for line in sys.stdin:
except UnicodeDecodeError:
sys.exit("Encoding problem in line " + str(line_number))

I got the point that there is no such thing as encoding-independent lines. But if no line ending can be found, then the file simply has one single line.

 

[jaroslav.dobrek]

Hi Jaroslav,

you can catch a UnicodeDecodeError just like any other exception. Can
you provide a full example program that shows your problem?

This works fine on my system:


import sys
open('tmp', 'wb').write(b'\xff\xff')
try:
buf = open('tmp', 'rb').read()
buf.decode('utf-8')
except UnicodeDecodeError as ude:
sys.exit("Found a bad char in file " + "tmp&quot

Thank you. I got it. What I need to do is explicitly decode text.

But I think trial and error with moving files around will in most cases be faster. Usually, such a problem occurs with some (usually complex) program that I wrote quite a long time ago. I don't like editing old and complex programs that work under all normal circumstances.

What I am missing (especially for Python3) is something like:

try:
for line in sys.stdin:
except UnicodeDecodeError:
sys.exit("Encoding problem in line " + str(line_number))

I got the point that there is no such thing as encoding-independent lines. But if no line ending can be found, then the file simply has one single line.

 

[Dave Angel]

Thank you. I got it. What I need to do is explicitly decode text.

But I think trial and error with moving files around will in most cases be faster. Usually, such a problem occurs with some (usually complex) program that I wrote quite a long time ago. I don't like editing old and complex programs that work under all normal circumstances.

What I am missing (especially for Python3) is something like:

try:
for line in sys.stdin:
except UnicodeDecodeError:
sys.exit("Encoding problem in line " + str(line_number))

I got the point that there is no such thing as encoding-independent lines. But if no line ending can be found, then the file simply has one single line.

i can't understand your question. if the problem is that the system
doesn't magically produce a variable called line_number, then generate
it yourself, by counting
in the loop.

Don't forget that you can tell the unicode decoder to ignore bad
characters, or to convert them to a specified placeholder.

 

[Jaroslav Dobrek]

i can't understand your question.  if the problem is that the system
doesn't magically produce a variable called line_number, then generate
it yourself, by counting
in the loop.

That was just a very incomplete and general example.

My problem is solved. What I need to do is explicitly decode text when
reading it. Then I can catch exceptions. I might do this in future
programs.

I dislike about this solution that it complicates most programs
unnecessarily. In programs that open, read and process many files I
don't want to explicitly decode and encode characters all the time. I
just want to write:

for line in f:

or something like that. Yet, writing this means to *implicitly* decode
text. And, because the decoding is implicit, you cannot say

try:
for line in f: # here text is decoded implicitly
do_something()
except UnicodeDecodeError():
do_something_different()

This isn't possible for syntactic reasons.

The problem is that vast majority of the thousands of files that I
process are correctly encoded. But then, suddenly, there is a bad
character in a new file. (This is so because most files today are
generated by people who don't know that there is such a thing as
encodings.) And then I need to rewrite my very complex program just
because of one single character in one single file.

 

[Stefan Behnel]

Jaroslav Dobrek, 26.07.2012 09:46:

My problem is solved. What I need to do is explicitly decode text when
reading it. Then I can catch exceptions. I might do this in future
programs.

Yes, that's the standard procedure. Decode on the way in, encode on the way
out, use Unicode everywhere in between.

I dislike about this solution that it complicates most programs
unnecessarily. In programs that open, read and process many files I
don't want to explicitly decode and encode characters all the time. I
just want to write:

for line in f:

And the cool thing is: you can!

In Python 2.6 and later, the new Py3 open() function is a bit more hidden,
but it's still available:

from io import open

filename = "somefile.txt"
try:
with open(filename, encoding="utf-8") as f:
for line in f:
process_line(line) # actually, I'd use "process_file(f)"
except IOError, e:
print("Reading file %s failed: %s" % (filename, e))
except UnicodeDecodeError, e:
print("Some error occurred decoding file %s: %s" % (filename, e))


Ok, maybe with a better way to handle the errors than "print" ...

For older Python versions, you'd use "codecs.open()" instead. That's a bit
messy, but only because it was finally cleaned up for Python 3.

or something like that. Yet, writing this means to *implicitly* decode
text. And, because the decoding is implicit, you cannot say

try:
for line in f: # here text is decoded implicitly
do_something()
except UnicodeDecodeError():
do_something_different()

This isn't possible for syntactic reasons.

Well, you'd normally want to leave out the parentheses after the exception
type, but otherwise, that's perfectly valid Python code. That's how these
things work.

The problem is that vast majority of the thousands of files that I
process are correctly encoded. But then, suddenly, there is a bad
character in a new file. (This is so because most files today are
generated by people who don't know that there is such a thing as
encodings.) And then I need to rewrite my very complex program just
because of one single character in one single file.

Why would that be the case? The places to change should be very local in
your code.

Stefan

 

[Chris Angelico]

My problem is solved. What I need to do is explicitly decode text when
reading it. Then I can catch exceptions. I might do this in future
programs.

Apologies if it's already been said (I'm only skimming this thread),
but ISTM that you want to open the file in binary mode. You'll then
get back a bytes() instead of a str(), and you can attempt to decode
it separately. You may then need to do your own division into lines
that way, though.

ChrisA

 

[wxjmfauth]

On Jul 25, 8:50 pm, Dave Angel <[email protected]> wrote:
> On 07/25/2012 08:09 AM, (e-mail address removed) wrote:
>
>
>
>
>
>
>
>
>
> > On Wednesday, July 25, 2012 1:35:09 PM UTC+2, Philipp Hagemeister wrote:
> >> Hi Jaroslav,
>
> >> you can catch a UnicodeDecodeError just like any other exception. Can
> >> you provide a full example program that shows your problem?
>
> >> This works fine on my system:
>
> >> import sys
> >> open('tmp', 'wb&#39.write(b'\xff\xff&#39
> >> try:
> >>     buf = open('tmp', 'rb&#39.read()
> >>     buf.decode('utf-8&#39
> >> except UnicodeDecodeError as ude:
> >>     sys.exit("Found a bad char in file " + "tmp&quot
>
> > Thank you. I got it. What I need to do is explicitly decode text.
>
> > But I think trial and error with moving files around will in most cases be faster. Usually, such a problem occurs with some (usually complex) program that I wrote quite a long time ago. I don't like editing old and complex programs that work under all normal circumstances.
>
> > What I am missing (especially for Python3) is something like:
>
> > try:
> >     for line in sys.stdin:
> > except UnicodeDecodeError:
> >     sys.exit("Encoding problem in line " + str(line_number))
>
> > I got the point that there is no such thing as encoding-independent lines. But if no line ending can be found, then the file simply has one single line.
>
> i can't understand your question.  if the problem is that the system
> doesn't magically produce a variable called line_number, then generate
> it yourself, by counting
> in the loop.


That was just a very incomplete and general example.

My problem is solved. What I need to do is explicitly decode text when
reading it. Then I can catch exceptions. I might do this in future
programs.

I dislike about this solution that it complicates most programs
unnecessarily. In programs that open, read and process many files I
don't want to explicitly decode and encode characters all the time. I
just want to write:

for line in f:

or something like that. Yet, writing this means to *implicitly* decode
text. And, because the decoding is implicit, you cannot say

try:
for line in f: # here text is decoded implicitly
do_something()
except UnicodeDecodeError():
do_something_different()

This isn't possible for syntactic reasons.

The problem is that vast majority of the thousands of files that I
process are correctly encoded. But then, suddenly, there is a bad
character in a new file. (This is so because most files today are
generated by people who don't know that there is such a thing as
encodings.) And then I need to rewrite my very complex program just
because of one single character in one single file.

In my mind you are taking the problem the wrong way.

Basically there is no "real UnicodeDecodeError", you are
just wrongly attempting to read a file with the wrong
codec. Catching a UnicodeDecodeError will not correct
the basic problem, it will "only" show, you are using
a wrong codec.
There is still the possibility, you have to deal with an
ill-formed utf-8 codding, but I doubt it is the case.

Do not forget, a "bit of text" has only a meaning if you
know its coding.

In short, all your files are most probably ok, you do not read
them correctly.
Traceback (most recent call last):
File "<eta last command>", line 1, in <module>
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xea in
position 3: invalid continuation byte'abcêdef'

jmf

 

[Jaroslav Dobrek]

And the cool thing is: you can!

In Python 2.6 and later, the new Py3 open() function is a bit more hidden,
but it's still available:

    from io import open

    filename = "somefile.txt"
    try:
        with open(filename, encoding="utf-8") as f:
            for line in f:
                process_line(line)  # actually, I'd use"process_file(f)"
    except IOError, e:
        print("Reading file %s failed: %s" % (filename, e))
    except UnicodeDecodeError, e:
        print("Some error occurred decoding file %s: %s" % (filename, e))

Thanks. I might use this in the future.

Well, you'd normally want to leave out the parentheses after the exception
type, but otherwise, that's perfectly valid Python code. That's how these
things work.

You are right. Of course this is syntactically possible. I was too
rash, sorry. In confused
it with some other construction I once tried. I can't remember it
right now.

But the code above (without the brackets) is semantically bad: The
exception is not caught.

Why would that be the case? The places to change should be very local in
your code.

This is the case in a program that has many different functions which
open and parse different
types of files. When I read and parse a directory with such different
types of files, a program that
uses

for line in f:

will not exit with any hint as to where the error occurred. I just
exits with a UnicodeDecodeError. That
means I have to look at all functions that have some variant of

for line in f:

in them. And it is not sufficient to replace the "for line in f" part.
I would have to transform many functions that
work in terms of lines into functions that work in terms of decoded
bytes.

That is why I usually solve the problem by moving fles around until I
find the bad file. Then I recode or repair
the bad file manually.

 

[Stefan Behnel]

Jaroslav Dobrek, 26.07.2012 12:51:

the code above (without the brackets) is semantically bad: The
exception is not caught.

Sure it is. Just to repeat myself: if the above doesn't catch the
exception, then the exception did not originate from the place where you
think it did. Again: look at the traceback.

This is the case in a program that has many different functions which
open and parse different
types of files. When I read and parse a directory with such different
types of files, a program that
uses

for line in f:

will not exit with any hint as to where the error occurred. I just
exits with a UnicodeDecodeError.

.... that tells you the exact code line where the error occurred. No need to
look around.

Stefan

 

[jaroslav.dobrek]

that tells you the exact code line where the error occurred. No need to

look around.

You are right:

try:
for line in f:
do_something()
except UnicodeDecodeError:
do_something_different()

does exactly what one would expect it to do.

Thank you very much for pointing this out and sorry for all the posts. This is one of the days when nothing seems to work and when I don't seem to able to read the simplest error message.

 

[jaroslav.dobrek]

that tells you the exact code line where the error occurred. No need to

look around.

You are right:

try:
for line in f:
do_something()
except UnicodeDecodeError:
do_something_different()

does exactly what one would expect it to do.

Thank you very much for pointing this out and sorry for all the posts. This is one of the days when nothing seems to work and when I don't seem to able to read the simplest error message.

 

[Philipp Hagemeister]

... that tells you the exact code line where the error occurred.

Which property of a UnicodeDecodeError does include that information?

On cPython 2.7 and 3.2, I see only start and end, both of which refer to
the number of bytes read so far.

I used the followin test script:

e = None
try:
b'a\xc3\xa4\nb\xff0'.decode('utf-8')
except UnicodeDecodeError as ude:
e = ude
print(e.start) # 5 for this input, 3 for the input b'a\nb\xff0'
print(dir(e))

But even if you would somehow determine a line number, this would only
work if the actual encoding uses 0xa for newline. Most encodings (101
out of 108 applicable ones in cPython 3.2) do include 0x0a in their
representation of '\n', but multi-byte encodings routinely include 0x0a
bytes in their representation of non-newline characters. Therefore, the
most you can do is calculate an upper bound for the line number.

- Philipp






-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)

iEYEAREKAAYFAlARNWoACgkQ9eq1gvr7CFy8kACeMeAslB7dwOIlDSOlZd7fq0TO
0o0AnAz9yvd2pErICNfJTvh+ilrqsMhC
=DVt/
-----END PGP SIGNATURE-----

 

[Stefan Behnel]

Philipp Hagemeister, 26.07.2012 14:17:

Which property of a UnicodeDecodeError does include that information?

On cPython 2.7 and 3.2, I see only start and end, both of which refer to
the number of bytes read so far.

Read again: "*code* line". The OP was apparently failing to see that the
error did not originate in the source code lines that he had wrapped with a
try-except statement but somewhere else, thus leading to the misguided
impression that the exception was not properly caught by the except clause.

Stefan

 

[Philipp Hagemeister]

Read again: "*code* line". The OP was apparently failing to see that >

the error did not originate in the source code lines that he had

wrapped with a try-except statement but somewhere else, thus leading

to the misguided impression that the exception was not properly caught
by the except clause.

Oops, over a dozen posts and I still haven't grasped the OP's problem.
Sorry! and thanks for noting that.

- Philipp


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)

iEYEAREKAAYFAlARO3IACgkQ9eq1gvr7CFzB7wCfc2+iRzUQVmaysDBNPq8JmM86
/FIAoIDrIzKLc3CpdsONnOyg+3Xi6MRh
=Awd6
-----END PGP SIGNATURE-----

 

[Robert Miles]

Thanks. I might use this in the future.


You are right. Of course this is syntactically possible. I was too
rash, sorry. In confused
it with some other construction I once tried. I can't remember it
right now.

But the code above (without the brackets) is semantically bad: The
exception is not caught.



This is the case in a program that has many different functions which
open and parse different
types of files. When I read and parse a directory with such different
types of files, a program that
uses

for line in f:

will not exit with any hint as to where the error occurred. I just
exits with a UnicodeDecodeError. That
means I have to look at all functions that have some variant of

for line in f:

in them. And it is not sufficient to replace the "for line in f" part.
I would have to transform many functions that
work in terms of lines into functions that work in terms of decoded
bytes.

That is why I usually solve the problem by moving fles around until I
find the bad file. Then I recode or repair
the bad file manually.

Would it be reasonable to use pieces of the old program to write a
new program that prints the name for an input file, then searches
that input file for bad characters? If it doesn't find any, it can
then go on to the next input file, or show a message saying that no
bad characters were found.