How do I get info on an exception ?

[Erik Max Francis]

Using Python 2.2.2,
I want to catch all exceptions from "socket.gethostbyaddr(ip)"

From IDLE, I can generate:
Traceback (most recent call last):
File "<pyshell#28>", line 1, in ?
socket.gethostbyaddr('1.2')
herror: (11004, 'host not found') <=== what I want when I catch

When I run this code:
try:
hostname, aliases, hostip = socket.gethostbyaddr(ip)
return (hostip, hostname)
except:
print sys.exc_info()
print sys.exc_type
return

Use the format:

try:
...
except ErrorType, e:
... do something with exception object e ...
.... socket.gethostbyaddr('1.2')
.... except socket.error, e:
.... print e, dir(e), e.args
....
(1, 'Unknown host') ['__doc__', '__getitem__', '__init__', '__module__',
'__str__', 'args'] (1, 'Unknown host')

How do I get the "(11004, 'host not found')" part?
More importantly, where is the answer documented that I should
have looked?

Check the part on exception handling.

 

[Frank]

Using Python 2.2.2,
I want to catch all exceptions from "socket.gethostbyaddr(ip)"

From IDLE, I can generate:Traceback (most recent call last):
File "<pyshell#28>", line 1, in ?
socket.gethostbyaddr('1.2')
herror: (11004, 'host not found') <=== what I want when I catch

When I run this code:
try:
hostname, aliases, hostip = socket.gethostbyaddr(ip)
return (hostip, hostname)
except:
print sys.exc_info()
print sys.exc_type
return

The "print sys.exc_info()" gives-
(<class socket.herror at 0x00AE7AC0>,
<socket.herror instance at 0x00C725C0>,
<traceback object at 0x00C73920>)

And the "print sys.exc_type" gives-
socket.herror

How do I get the "(11004, 'host not found')" part?
More importantly, where is the answer documented that I should
have looked?

Frank

 

[Ian Bicking]

Using Python 2.2.2,
I want to catch all exceptions from "socket.gethostbyaddr(ip)"

Traceback (most recent call last):
File "<pyshell#28>", line 1, in ?
socket.gethostbyaddr('1.2')
herror: (11004, 'host not found') <=== what I want when I catch

When I run this code:
try:
hostname, aliases, hostip = socket.gethostbyaddr(ip)
return (hostip, hostname)
except:
print sys.exc_info()
print sys.exc_type
return

try:
do exceptional stuff...
except Exception, e:
print e.args, dir(e)

Probably that information is just in e.args, but you can add other
attributes to exceptions, any of which should be revealed by dir(e).

Exceptions are poorly documented, so I don't know where that might be
noted.

Ian

 

[Heiko Wundram]

How do I get the "(11004, 'host not found')" part?

More importantly, where is the answer documented that I should
have looked?

Look at the traceback module from the Python standard library. If you
just wish to print the exception, do the following:

import traceback

try:
nosuchobject
except:
traceback.print_exc()

HTH!

Heiko.

 

[Bengt Richter]

Use the format:

try:
...
except ErrorType, e:
... do something with exception object e ...

Indeed. Or if you want to catch all standard exceptions until you know what is
happening, you could pick from the following. I guess you could also catch all
exceptions defined in a module without knowing what they're named, and re-raise
if other, e.g., with (untested!)

if not repr(e.__class__).startswith('socket'): raise

is there another way to do that?
... socket.gethostbyaddr('1.2')
... except Exception, e:
.... print 'plain e:', e
.... print 'e class:', e.__class__
.... print 'e class name:', e.__class__.__name__
.... print 'vars keys:', vars(e).keys()
.... print 'vars dict:', vars(e)
.... print 'additional inherited clutter seen by dir:\n',dir(e)
....
plain e: (11004, 'host not found')
e class: socket.herror
e class name: herror
vars keys: ['args']
vars dict: {'args': (11004, 'host not found')}
additional inherited clutter seen by dir:
['__doc__', '__getitem__', '__init__', '__module__', '__str__', 'args']

You could also print a formatted vars(e) to see unanticipated attributes more nicely
with e.g., (untested!) print ''.join(['%16s = %r\n' %(k,v) for k,v in vars(e).items()])

... socket.gethostbyaddr('1.2')
... except socket.error, e:
... print e, dir(e), e.args
...
(1, 'Unknown host') ['__doc__', '__getitem__', '__init__', '__module__',
'__str__', 'args'] (1, 'Unknown host')

How do I get the "(11004, 'host not found')" part?
More importantly, where is the answer documented that I should
have looked?

Check the part on exception handling.

Interactively, help('exceptions') is also helpful (note quotes)

Regards,
Bengt Richter

 

[Cameron Laird]

.
.
.

You could catch it with:

except socket.herror, inst:
print inst.args

or more broadly with:

except socket.error, (errno, string_message):
print code, message



The list of possible socket exceptions is in the docs for sockets.
It also describes the (errno, string) return tuple value of inst.args.

.
.
.
True.

But unsatisfying--at least to me. I'm curious whether anyone
else sees it the same way.

We're talking here about <URL:
http://www.python.org/doc/lib/module-socket.html >, I assume,
which tells us the socket module has three exception constants
(and what they are), and also such details as that it might be
except socket.error, (errno, string_message)
but that it also might be simply
except socket.error, string_message

It's a temptation to go rhetorical at this point, and protest
that we would not believe any other API documented if so little
were written about it, but that we expect this poverty for
exceptions. Rather than promote a general manifesto, though,
I'll just summarize that neither the manual entry, nor the
references it provides, give definite knowledge about even the
signature of socket exceptions. A programmer who wants preci-
sion in regard to exceptions must either experiment, or read
implementation sources--and sometimes both.

So, the answer to the question is the subject line is this:
read the manual, then play with a live system to see what
really happens.

 

[Thomas Heller]

The list of possible socket exceptions is in the docs for sockets.

.
.
I deserved that.

There was more to my post, of course. Part of what I was trying to
express is that exception interfaces are almost always relegated to
a footnote, with only a generalized description, even in the frequent
case that a method's exceptions are more complicated than its invoca-
tions.

Rather than tilt at the collection of all such windmills, I want to
first understand better why this is, and what response is appropriate.
To summarize: I don't know what patch is the right one. I also
thought it only fair to warn Mr. July that things are indeed more dif-
ficult than we were explaining, even though I didn't feel up to
detailing the difficulties.

So, Raymond, do you have general guidelines for how you think excep-
tions should be documented?

I don't know what Raymond will suggest, but for myself I always (well,
nearly always) when I find the docs insufficient, I'll submit a patch
which contains what I would like to read there.
Works pretty well so far...

Thomas

 

[Raymond Hettinger]

[Cameron Laird]

.
.
.
I deserved that.

It wasn't a jab.
Truly, if you see a way to clarify the docs, submit a patch.
If everyone finds it helpful, it will be accepted. That's how
the docs improve over time. I've personally written or
accepted over a hundred of these in the past year.

There was more to my post, of course. Part of what I was trying to
express is that exception interfaces are almost always relegated to
a footnote, with only a generalized description, even in the frequent
case that a method's exceptions are more complicated than its invoca-
tions.

No doubt, exceptions often get less explanation than everything else.
However, in this case, the docs seemed reasonably clear to me.

The part that could be improved is where it talks about returning a
message or a tuple, as if either one but not both could occur at any
time. It would be more accurate to say, that in all cases an exception
instance is returned; the instance has a __str__ method so it can be
printed as if it were a string; the instance has an attribute "args" which
is the tuple (errno, errmsg); and the instance has a __getitem__ method
so the args tuple can be directly unpacked as if the instance had been
the tuple itself.

OTOH, that is long-winded and distracts from the flow of knowledge
about sockets. It is also how all exception instances work.

So, Raymond, do you have general guidelines for how you think excep-
tions should be documented?

Unless there are interesting values being returned, it is probably
sufficient to name the exception, state what it represents, and
describe where it fits in the class structure (i.e. socket.herror is a
subclass of socket.error). When you think about it, why have
three paragraphs on an exception whose code is as simple as:

class xyzException(Exception): pass # raised when xyz occurs

When interesting values are returned, describe the meaning of
each element in the instance's arg tuple (i.e. (errno, errmsg)).
Afterall, the code behind it may be as simple as:

raise socket.herror(errno, errmsg) # reveal the packet's error information

Elsewhere in the docs, functions descriptions should list the names
of the exceptions they can raise unless it is obvious or pertains
to a whole collection of functions (i.e. all high-volume socket
operations can potentially raise a timeout).

In general, I like the documents to be as specific as possible without
replicating the code or locking in a particular implementation. However,
there is a competing force to keep module documents centered on how
to use the module. For example, I found the unittest documentation to
be highly specific and thorough, yet unsatisfactory because you practically
had to read a book on the subject to be able to make basic use of the
module. To fix it, I added a new introductory section that covers
(in one page) the 5% of the module that will meet 95% of your needs
(enough to get you up and running).


Raymond Hettinger

 

[Frank]

[Cameron Laird]

.
.
.
I deserved that.

It wasn't a jab.
Truly, if you see a way to clarify the docs, submit a patch.
If everyone finds it helpful, it will be accepted. That's how
the docs improve over time. I've personally written or
accepted over a hundred of these in the past year.

There was more to my post, of course. Part of what I was trying to
express is that exception interfaces are almost always relegated to
a footnote, with only a generalized description, even in the frequent
case that a method's exceptions are more complicated than its invoca-
tions.

No doubt, exceptions often get less explanation than everything else.
However, in this case, the docs seemed reasonably clear to me.

The part that could be improved is where it talks about returning a
message or a tuple, as if either one but not both could occur at any
time. It would be more accurate to say, that in all cases an exception
instance is returned; the instance has a __str__ method so it can be
printed as if it were a string; the instance has an attribute "args" which
is the tuple (errno, errmsg); and the instance has a __getitem__ method
so the args tuple can be directly unpacked as if the instance had been
the tuple itself.

OTOH, that is long-winded and distracts from the flow of knowledge
about sockets. It is also how all exception instances work.

So, Raymond, do you have general guidelines for how you think excep-
tions should be documented?

Unless there are interesting values being returned, it is probably
sufficient to name the exception, state what it represents, and
describe where it fits in the class structure (i.e. socket.herror is a
subclass of socket.error). When you think about it, why have
three paragraphs on an exception whose code is as simple as:

class xyzException(Exception): pass # raised when xyz occurs

When interesting values are returned, describe the meaning of
each element in the instance's arg tuple (i.e. (errno, errmsg)).
Afterall, the code behind it may be as simple as:

raise socket.herror(errno, errmsg) # reveal the packet's error information

Elsewhere in the docs, functions descriptions should list the names
of the exceptions they can raise unless it is obvious or pertains
to a whole collection of functions (i.e. all high-volume socket
operations can potentially raise a timeout).

In general, I like the documents to be as specific as possible without
replicating the code or locking in a particular implementation. However,
there is a competing force to keep module documents centered on how
to use the module. For example, I found the unittest documentation to
be highly specific and thorough, yet unsatisfactory because you practically
had to read a book on the subject to be able to make basic use of the
module. To fix it, I added a new introductory section that covers
(in one page) the 5% of the module that will meet 95% of your needs
(enough to get you up and running).


Raymond Hettinger

Thanks to all for your help.

After checking that the input is a string, I try calling
gethostbyname/addr depending on whether the first char is numeric and
catch:
"except socket.error, err:"
Will this catch all possible errors? I've tried feeding various
errors into my def and caught them all so far. Input might come from
users at the keyboard, and crashing the program is not an option, no
matter what is typed in.
------------------------------------
I did look for the solution before posting; I checked:
Python Lib Ref 2.2.2
Python Language Ref 2.2.2
Lutz - Learning Python
Lutz - Programming Python, 2nd
Holden - Python Web Programming
Brueck - Python 2.1 Bible
Martelli - Python in a Nutshell
and even Guido's Tutorial. If the complete answer is there, I could
not find it or assemble all the parts from the various references.
Perhaps with more experience in Python it would have been clearer, but
I have to agree with those that say exceptions are not well
documented. Complete and precise documentation is probably more
important for a language/compiler/interpreter system than any anything
else.

Frank

 

[Bengt Richter]

.
.
.
I deserved that.

There was more to my post, of course. Part of what I was trying to
express is that exception interfaces are almost always relegated to
a footnote, with only a generalized description, even in the frequent
case that a method's exceptions are more complicated than its invoca-
tions.

Rather than tilt at the collection of all such windmills, I want to
first understand better why this is, and what response is appropriate.
To summarize: I don't know what patch is the right one. I also
thought it only fair to warn Mr. July that things are indeed more dif-
ficult than we were explaining, even though I didn't feel up to
detailing the difficulties.

So, Raymond, do you have general guidelines for how you think excep-
tions should be documented?

ISTM there are (at least) two important places for doc info: Code-associated and
language/tool-usage-associated. There are docs we approach as docs like books to read,
and there are embedded docs as in doc strings and comments. And pydoc sort of
bridges the gap and makes a book from code and whatall.

I wonder where the best place is to enhance the documentation accessible to users.
I suspect it would be in the area of pydoc and the structure of "code and whatall."
The most likely place to find up-to-date info is in the latest code. (See below
for an idea of how to enhance the latest code doc info without changing known-good
code source).

Exceptions (subject topic ;-) are a special case. In principle any code documentation
could be enhanced by adopting source conventions to help pydoc present better info.
The question then becomes how best to format the doc-enhancing info, where to put it,
and how best to support its presentation.

One thing that is a deterrent to improved documentation *within* code is that people
like to minimize the number of official code versions, whether by CVS or just noting
MD5 digests etc., and/or by burning an "official" CDROM. This makes me wonder if it would
make sense to have a way to store doc info separately, yet still integratable into a pydoc
presentation.

One way would be to have a file name convention, so that the optional doc-enhancement
info for e.g., xxx.py would be found in, e.g., xxx.dpy. This could make a clutter of small
files though, so maybe one or more .dpx files could contain the mapping from multiple
small related sources to a single .dpy. And by convention .dpy info for packaged files
would be looked for in a .dpy for the overall package if not found separately.

Ok, we could hone this search mechanism. What about the structure of .dpy files themselves,
which now potentially could have info enhancing docs for both xxx.py and yyy.py?

Some kind of easy internal labeled sections to separate xxx.py from yyy.py info
would seem obvious. Then, the real question: what goes in the .dpy section for xxx.py
so as to tie into certain parts of xxx.py code, without requiring changes to that code
(which, remember, may be on a CDROM etc)?

The context-finding mechanism of diff/patch would be one option, but to use that literally
would make the .dpy files into diff files, and probably not something easy to generate
manually and readably. And you don't want to carry a whole source fork just to maintain
doc enhancement .dpy files. So we need some conventions functionally to enable human-doable
doc diff patches. Maybe with special easy syntax to refer to exception raising loci
in the code vs catching, vs functions, classes, methods, existing doc strings, etc, etc.

An advantage of pure doc-enhancement info is that a mistake can't break the actual code,
so more people could be given commit privileges for that kind of files, if maintained
separately on CVS. Even wide-open wiki methods could be considered. You could even configure
pydoc to query python.org efficiently to check if a local cache is out of date, and prompt
for optional download.

But this is a matter of arriving at some PEP consensus after discussing 1) whether it's
worth discussing, and 2) what the usability goals are for both users and creators of
doc enhancement info, and 3) who has the time to follow up on stuff like this.
Just writing this post cost something, as will your replies (if any ;-)

Regards,
Bengt Richter

 

[Raymond Hettinger]

[Raymond]

The part that could be improved is where it talks about returning a
message or a tuple, as if either one but not both could occur at any
time. It would be more accurate to say, that in all cases an exception
instance is returned; the instance has a __str__ method so it can be
printed as if it were a string; the instance has an attribute "args" which
is the tuple (errno, errmsg); and the instance has a __getitem__ method
so the args tuple can be directly unpacked as if the instance had been
the tuple itself.

[Frank]
After checking that the input is a string, I try calling
gethostbyname/addr depending on whether the first char is numeric and
catch:
"except socket.error, err:"
Will this catch all possible errors? I've tried feeding various
errors into my def and caught them all so far. Input might come from
users at the keyboard, and crashing the program is not an option, no
matter what is typed in.

Yes, that will catch all errors. You have bound the instance to "err".

As I said earlier, the part of the docs that some folks may
find confusing is that part about returning a string or a tuple.
In fact, it always returns an instance that can be printed like
a string *and* unpacked like a tuple.

I'll modify the docs to make this more clear.

If you're in an experimental frame of mind, it is straight-forward to
learn exactly what is going on by adding a few print statements:

except socket.error, inst:
print 'Caught a', type(inst) # show that an exception instance was
recd
print 'The inst can be printed like a string:', inst
errno, errmsg = inst # show the exception can be unpacked
print 'It contains a tuple with', errno, 'and', errmsg
print 'Its methods are', dir(inst)
print 'The data is kept in inst.args', inst.args
print 'It prints because of the __str__ method:', inst.__str__()
print 'That is my story and I am sticking with it!'


Raymond Hettinger




Raymond Hettinger