marshal bug?

[Anurag]

I have been chasing a problem in my code since hours and it bolis down
to this
import marshal
marshal.dumps(s tr(123)) != marshal.dumps(s tr("123"))

Can someone please tell me why?
when
str(123) == str("123")

or are they different?

it also means that
if s = str(123)
marshal.dumps(s ) != marshal.dumps(m arshal.loads(ma rshal.dumps(s)) )

rgds
Anurag

[Anurag]

Thanks for the reply.

It seems problem is due to
"""
Any string in Python can be "interned" or not, the difference being
how/where the value is stored internally. The marshal module includes
such information in its output. What you are doing is probably
considered a misuse of the marshal module. I'd suggest using the
pickle
(or cPickle) modules instead.
"""
as mentioned by Gary Herron

Now is there a easy way to by pass it (hack around it)
I tried various options but all fail e.g.
i= 123; marshal.dumps(" %d"%123) != marshal.dumps(" %d"%i)

So maybe I have to change all my code to use pickle, which also
consumes for memory per string.

[Gary Herron]

Anurag wrote:

I have been chasing a problem in my code since hours and it bolis down
to this
import marshal
marshal.dumps(s tr(123)) != marshal.dumps(s tr("123"))

Can someone please tell me why?
when
str(123) == str("123")

or are they different?

it also means that
if s = str(123)
marshal.dumps(s ) != marshal.dumps(m arshal.loads(ma rshal.dumps(s)) )

rgds
Anurag

Any string in Python can be "interned" or not, the difference being
how/where the value is stored internally. The marshal module includes
such information in its output. What you are doing is probably
considered a misuse of the marshal module. I'd suggest using the pickle
(or cPickle) modules instead.

Here's the relevant part of the manual for marshal:

version:
Indicates the format that the module uses. Version 0 is the
historical format, version 1 (added in Python 2.4) shares interned
strings and version 2 (added in Python 2.5) uses a binary format for
floating point numbers. The current version is 2



Gary Herron.

[David]

I'm not sure why, but marshal does dump the 2 differently. ie:

>

>marshal.dumps( str(123))

's\x03\x00\x00\ x00123'

>marshal.dumps( str("123"))

't\x03\x00\x00\ x00123'

I've just checked the source [1].

's' refers to a regular string, 't' refers to an interned[2] string.

In other words the difference between the 2 marshal dumps is an
internal python implementation detail. You probably shouldn't be
comparing marshal dumps in your app in the first place.

[1] http://coverage.livinglogic.de/Python/marshal.c.html. (Actually, I
checked the downloaded bz2, but this is the only URL for marshal.c I
could find)
[2] http://mindprod.com/jgloss/interned.html

[Gabriel Genellina]

En Fri, 28 Sep 2007 04:26:39 -0300, Anurag <an**********@g mail.com>
escribi�:

Now is there a easy way to by pass it (hack around it)
I tried various options but all fail e.g.
i= 123; marshal.dumps(" %d"%123) != marshal.dumps(" %d"%i)

You can't. Don't use marshal to compare objects. You appear to assume that
two objects that compare equal, should have the same marshalled
representation, and that is simply not true.

py[1, 2.0, 3+0j, "4"] == [1.0, 2+0j, 3, u"4"]
True

So maybe I have to change all my code to use pickle, which also
consumes for memory per string.

Neither marshal nor pickle guarantee that equal objects have equal
representations , so in this regard pickle won't help either.
Maybe if you explain what you really want to do someone could suggest a
solution.

--
Gabriel Genellina

[=?ISO-8859-1?Q?=22Martin_v=2E_L=F6wis=22?=]

[1] http://coverage.livinglogic.de/Python/marshal.c.html. (Actually, I

checked the downloaded bz2, but this is the only URL for marshal.c I
could find)

A better URL is

http://svn.python.org/projects/pytho...thon/marshal.c

or

http://svn.python.org/view/python/tr...thon/marshal.c

Regards,
Martin