Object persistence#

You can load and save most Sage object to disk using the load and save member functions and commands.

Note

It is impossible to save certain Sage objects to disk. For example, if \(x\) is a MAGMA object, i.e., a wrapper around an object that is defined in MAGMA, there is no way to save \(x\) to disk, since MAGMA doesn’t support saving of individual objects to disk.

  • Versions: Loading and saving of objects is guaranteed to work even if the version of Python changes. Saved objects can be loaded in future versions of Python. However, if the data structure that defines the object, e.g., in Sage code, changes drastically (or changes name or disappears), then the object might not load correctly or work correctly.

  • Objects are zlib compressed for space efficiency.

class sage.misc.persist.SagePickler(file_obj, persistent_id=None, py2compat=True)#

Bases: _BasePickler

Subclass pickle.Pickler with Sage-specific default options, and built-in support for external object persistence.

INPUT:

  • file_obj – a readable file-like object returning bytes from which the pickle data will be loaded.

  • persistent_id – callable or None; if given this callable takes a single object to be pickled, and returns an “ID” (a key with which to restore the object upon unpickling, which may itself be any pickleable object). See the Python documentation on pickling and unpickling external objects for more details.

  • py2compat – on Python 3 only, this creates pickles that have a better chance of being read on Python 2, by using protocol version 2 (instead of 4) and fixing up imports of standard library modules and types whose names changed between Python 2 and 3. This is enabled by default for the best chances of cross-Python compatibility.

  • Further arguments are passed to pickle.load(), where in Python-3 Sage sets the default encoding='latin1'. This is essential to make pickles readable in Python-3 that were created in Python-2. See github issue #28444 for details.

EXAMPLES:

sage: from sage.misc.persist import (
....:     unpickle_override, register_unpickle_override, SageUnpickler)
sage: from sage.rings.integer import make_integer
sage: from io import BytesIO
sage: def fake_constructor(x):
....:     print("unpickling an Integer")
....:     return make_integer(x)
sage: register_unpickle_override('sage.rings.integer', 'make_integer',
....:                            fake_constructor)
sage: unp = SageUnpickler(BytesIO(dumps(1, compress=False)))
sage: unp.load()
unpickling an Integer
1
sage: del unpickle_override[('sage.rings.integer', 'make_integer')]

The SagePickler can also be passed a persistent_id function:

sage: table = {1: 'a', 2: 'b'}
sage: # in practice this might be a database or something...
sage: def load_object_from_table(obj_id):
....:     tag, obj_id
....:     return table[obj_id]
classmethod dumps(obj, **kwargs)#

Equivalent to pickle.dumps() but using the sage.misc.persist.SagePickler.

INPUT:

OUTPUT:

  • pickle - the pickled object as bytes.

EXAMPLES:

sage: import pickle
sage: from sage.misc.persist import SagePickler
sage: gherkin = SagePickler.dumps(1)
sage: pickle.loads(gherkin)
1
class sage.misc.persist.SageUnpickler(file_obj, persistent_load=None, **kwargs)#

Bases: _BaseUnpickler

Subclass pickle.Unpickler to control how certain objects get unpickled (registered overrides, specifically).

This is only needed in Python 3 and up. On Python 2 the behavior of the cPickle module is customized differently.

This class simply overrides Unpickler.find_class to wrap sage.misc.persist.unpickle_global.

INPUT:

  • file_obj – a readable file-like object returning bytes from which the pickle data will be loaded.

  • persistent_load – callable or None; if given this callable implements loading of persistent external objects. The function should take a single argument, the persistent object ID. See the Python documentation on pickling and unpickling external objects for more details.

  • kwargs – additional keyword arguments passed to the pickle.Unpickler constructor.

EXAMPLES:

sage: from sage.misc.persist import (
....:     unpickle_override, register_unpickle_override, SageUnpickler)
sage: from sage.rings.integer import make_integer
sage: from io import BytesIO
sage: def fake_constructor(x):
....:     print("unpickling an Integer")
....:     return make_integer(x)
sage: register_unpickle_override('sage.rings.integer', 'make_integer',
....:                            fake_constructor)
sage: unp = SageUnpickler(BytesIO(dumps(1, compress=False)))
sage: unp.load()
unpickling an Integer
1
sage: del unpickle_override[('sage.rings.integer', 'make_integer')]

The SageUnpickler can also be passed a persistent_load function:

sage: table = {1: 'a', 2: 'b'}
sage: # in practice this might be a database or something...
sage: def load_object_from_table(obj_id):
....:     tag, obj_id
....:     return table[obj_id]
classmethod loads(data, **kwargs)#

Equivalent to pickle.loads() but using the sage.misc.persist.SagePickler.

INPUT:

OUTPUT:

  • obj - the object that was serialized to the given pickle data.

EXAMPLES:

sage: import pickle
sage: from sage.misc.persist import SageUnpickler
sage: gherkin = pickle.dumps(1)
sage: SageUnpickler.loads(gherkin)
1
sage.misc.persist.db(name)#

Load object with given name from the Sage database. Use x.db(name) or db_save(x, name) to save objects to the database.

The database directory is $HOME/.sage/db.

sage.misc.persist.db_save(x, name=None)#

Save x to the Sage database.

The database directory is $HOME/.sage/db.

sage.misc.persist.dumps(obj, compress=True)#

Dump obj to a string s. To recover obj, use loads(s).

See also

loads()

EXAMPLES:

sage: a = 2/3
sage: s = dumps(a)
sage: a2 = loads(s)
sage: type(a) is type(a2)
True
sage: a2
2/3
sage.misc.persist.load(compress=True, verbose=True, *filename, **kwargs)#

Load Sage object from the file with name filename, which will have an .sobj extension added if it doesn’t have one. Or, if the input is a filename ending in .py, .pyx, .sage, .spyx, .f, .f90 or .m, load that file into the current running session.

Loaded files are not loaded into their own namespace, i.e., this is much more like Python’s execfile than Python’s import.

This function also loads a .sobj file over a network by specifying the full URL. (Setting verbose = False suppresses the loading progress indicator.)

When a pickle created with Python 2 is unpickled in Python 3, Sage uses the default encoding latin1 to unpickle data of type str.

Finally, if you give multiple positional input arguments, then all of those files are loaded, or all of the objects are loaded and a list of the corresponding loaded objects is returned.

If compress is true (the default), then the data stored in the file are supposed to be compressed. If verbose is true (the default), then some logging is printed when accessing remote files. Further keyword arguments are passed to pickle.load().

EXAMPLES:

sage: u = 'https://www.sagemath.org/files/test.sobj'
sage: s = load(u)                                                  # optional - internet
Attempting to load remote file: https://www.sagemath.org/files/test.sobj
Loading started
Loading ended
sage: s                                                            # optional - internet
'hello SageMath'

We test loading a file or multiple files or even mixing loading files and objects:

sage: t = tmp_filename(ext='.py')
sage: with open(t, 'w') as f:
....:     _ = f.write("print('hello world')")
sage: load(t)
hello world
sage: load(t,t)
hello world
hello world
sage: t2 = tmp_filename(); save(2/3,t2)
sage: load(t,t,t2)
hello world
hello world
[None, None, 2/3]

Files with a .sage extension are preparsed. Also note that we can access global variables:

sage: t = tmp_filename(ext=".sage")
sage: with open(t, 'w') as f:
....:     _ = f.write("a += Mod(2/3, 11)")  # This evaluates to Mod(8, 11)
sage: a = -1
sage: load(t)
sage: a
7

We can load Fortran files:

sage: code = '      subroutine hello\n         print *, "Hello World!"\n      end subroutine hello\n'
sage: t = tmp_filename(ext=".F")
sage: with open(t, 'w') as f:
....:     _ = f.write(code)
sage: load(t)                                                                   # needs numpy
sage: hello                                                                     # needs numpy
<fortran ...>
sage.misc.persist.load_sage_element(cls, parent, dic_pic)#
sage.misc.persist.load_sage_object(cls, dic)#
sage.misc.persist.loads(s, compress=True, **kwargs)#

Recover an object x that has been dumped to a string s using s = dumps(x).

See also

dumps()

EXAMPLES:

sage: a = matrix(2, [1,2, 3,-4/3])                                              # needs sage.modules
sage: s = dumps(a)                                                              # needs sage.modules
sage: loads(s)                                                                  # needs sage.modules
[   1    2]
[   3 -4/3]

If compress is True (the default), it will try to decompress the data with zlib and with bz2 (in turn); if neither succeeds, it will assume the data is actually uncompressed. If compress=False is explicitly specified, then no decompression is attempted. Further arguments are passed to python’s pickle.load().

sage: v = [1..10]
sage: loads(dumps(v, compress=False)) == v
True
sage: loads(dumps(v, compress=False), compress=True) == v
True
sage: loads(dumps(v, compress=True), compress=False)
Traceback (most recent call last):
...
UnpicklingError: invalid load key, 'x'.

The next example demonstrates that Sage strives to avoid data loss in the transition from Python-2 to Python-3. The problem is that Python-3 by default would not be able to unpickle a non-ASCII Python-2 string appearing in a pickle. See github issue #28444 for details.

sage: class Foo():
....:     def __init__(self, s):
....:         self.bar = s
....:     def __reduce__(self):
....:         return Foo, (self.bar,)
....:
sage: import __main__
sage: __main__.Foo = Foo

The data that is passed to loads in the following line was created by dumps(Foo('\x80\x07') in Python-2.

sage: g = loads(b'x\x9ck`J\x8e\x8f\xcfM\xcc\xcc\x8b\x8f\xe7r\xcb\xcf\xe7*d\x0cej`/dj\r*d\xd6\x03\x00\x89\xc5\x08{')
sage: type(g), g.bar
(<class '__main__.Foo'>, '\x80\x07')

The following line demonstrates what would happen without github issue #28444:

sage: loads(b'x\x9ck`J\x8e\x8f\xcfM\xcc\xcc\x8b\x8f\xe7r\xcb\xcf\xe7*d\x0cej`/dj\r*d\xd6\x03\x00\x89\xc5\x08{', encoding='ASCII')
Traceback (most recent call last):
...
UnicodeDecodeError: 'ascii' codec can...t decode byte 0x80 in position 0: ordinal not in range(128)
sage.misc.persist.make_None(*args, **kwds)#

Do nothing and return None. Used for overriding pickles when that pickle is no longer needed.

EXAMPLES:

sage: from sage.misc.persist import make_None
sage: print(make_None(42, pi, foo='bar'))                                       # needs sage.symbolic
None
sage.misc.persist.picklejar(obj, dir=None)#

Create pickled sobj of obj in dir, with name the absolute value of the hash of the pickle of obj. This is used in conjunction with unpickle_all().

To use this to test the whole Sage library right now, set the environment variable SAGE_PICKLE_JAR, which will make it so dumps() will by default call picklejar() with the default dir. Once you do that and doctest Sage, you’ll find that the DOT_SAGE/pickle_jar directory contains a bunch of pickled objects along with corresponding txt descriptions of them. Use the unpickle_all() to see if they unpickle later.

INPUT:

  • obj – a pickleable object

  • dir – a string or None; if None then dir defaults to DOT_SAGE/pickle_jar

EXAMPLES:

sage: dir = tmp_dir()
sage: sage.misc.persist.picklejar(1, dir)
sage: sage.misc.persist.picklejar('test', dir)
sage: len(os.listdir(dir))   # Two entries (sobj and txt) for each object
4
sage.misc.persist.register_unpickle_override(module, name, callable, call_name=None)#

Python pickles include the module and class name of classes. This means that rearranging the Sage source can invalidate old pickles. To keep the old pickles working, you can call register_unpickle_override with an old module name and class name, and the Python callable (function, class with __call__ method, etc.) to use for unpickling. (If this callable is a value in some module, you can specify the module name and class name, for the benefit of explain_pickle() when called with in_current_sage=True).)

EXAMPLES:

Imagine that there used to be an old_integer module and old pickles essentially trying to do the following:

sage: unpickle_global('sage.rings.old_integer', 'OldInteger')
Traceback (most recent call last):
...
ImportError: cannot import OldInteger from sage.rings.old_integer, call register_unpickle_override('sage.rings.old_integer', 'OldInteger', ...) to fix this

After following the advice from the error message, unpickling works:

sage: from sage.misc.persist import register_unpickle_override
sage: register_unpickle_override('sage.rings.old_integer', 'OldInteger', Integer)
sage: unpickle_global('sage.rings.old_integer', 'OldInteger')
<... 'sage.rings.integer.Integer'>

In many cases, unpickling problems for old pickles can be resolved with a simple call to register_unpickle_override, as in the example above and in many of the sage source files. However, if the underlying data structure has changed significantly then unpickling may fail and it will be necessary to explicitly implement unpickling methods for the associated objects. The python pickle protocol is described in detail on the web and, in particular, in the python pickling documentation. For example, the following excerpt from this documentation shows that the unpickling of classes is controlled by their __setstate__() method.

object.__setstate__(state)

    Upon unpickling, if the class also defines the method :meth:`__setstate__`, it is
    called with the unpickled state. If there is no :meth:`__setstate__` method,
    the pickled state must be a dictionary and its items are assigned to the new
    instance's dictionary. If a class defines both :meth:`__getstate__` and
    :meth:`__setstate__`, the state object needn't be a dictionary and these methods
    can do what they want.

By implementing a __setstate__() method for a class it should be possible to fix any unpickling problems for the class. As an example of what needs to be done, we show how to unpickle a CombinatorialObject object using a class which also inherits from Element. This exact problem often arises when refactoring old code into the element framework. First we create a pickle to play with:

sage: from sage.structure.element import Element
sage: class SourPickle(CombinatorialObject): pass
sage: class SweetPickle(CombinatorialObject, Element): pass
sage: import __main__
sage: __main__.SourPickle = SourPickle
sage: __main__.SweetPickle = SweetPickle  # a hack to allow us to pickle command line classes
sage: gherkin = dumps(SourPickle([1, 2, 3]))

Using register_unpickle_override() we try to sweeten our pickle, but we are unable to eat it:

sage: from sage.misc.persist import register_unpickle_override
sage: register_unpickle_override('__main__', 'SourPickle', SweetPickle)
sage: loads(gherkin)
Traceback (most recent call last):
...
KeyError: 0

The problem is that the SweetPickle has inherited a __setstate__() method from Element which is not compatible with unpickling for CombinatorialObject. We can fix this by explicitly defining a new __setstate__() method:

sage: class SweeterPickle(CombinatorialObject, Element):
....:     def __setstate__(self, state):
....:         # a pickle from CombinatorialObject is just its instance
....:         # dictionary
....:         if isinstance(state, dict):
....:             # this is a fudge: we need an appropriate parent here
....:             self._set_parent(Tableaux())
....:             self.__dict__ = state
....:         else:
....:             P, D = state
....:             if P is not None:
....:                 self._set_parent(P)
....:             self.__dict__ = D
sage: __main__.SweeterPickle = SweeterPickle
sage: register_unpickle_override('__main__', 'SourPickle', SweeterPickle)
sage: loads(gherkin)                                                            # needs sage.combinat
[1, 2, 3]
sage: loads(dumps(SweeterPickle([1, 2, 3])))  # check that pickles work for SweeterPickle
[1, 2, 3]

The state passed to __setstate__() will usually be something like the instance dictionary of the pickled object, however, with some older classes such as CombinatorialObject it will be a tuple. In general, the state can be any python object. Sage provides a special tool, explain_pickle(), which can help in figuring out the contents of an old pickle. Here is a second example.

sage: class A():
....:    def __init__(self,value):
....:        self.original_attribute = value
....:    def __repr__(self):
....:        return 'A(%s)' % self.original_attribute
sage: class B():
....:    def __init__(self,value):
....:        self.new_attribute = value
....:    def __setstate__(self,state):
....:        try:
....:            self.new_attribute = state['new_attribute']
....:        except KeyError:      # an old pickle
....:            self.new_attribute = state['original_attribute']
....:    def __repr__(self):
....:        return 'B(%s)' % self.new_attribute
sage: import __main__
sage: # a hack to allow us to pickle command line classes
sage: __main__.A = A
sage: __main__.B = B
sage: A(10)
A(10)
sage: loads(dumps(A(10)))
A(10)
sage: sage.misc.explain_pickle.explain_pickle(dumps(A(10)))
pg_A = unpickle_global('__main__', 'A')
si = unpickle_newobj(pg_A, ())
pg_make_integer = unpickle_global('sage.rings.integer', 'make_integer')
unpickle_build(si, {'original_attribute':pg_make_integer('a')})
si
sage: from sage.misc.persist import register_unpickle_override
sage: register_unpickle_override('__main__', 'A', B)
sage: loads(dumps(A(10)))
B(10)
sage: loads(dumps(B(10)))
B(10)

Pickling for python classes and extension classes, such as cython, is different – again this is discussed in the python pickling documentation. For the unpickling of extension classes you need to write a __reduce__() method which typically returns a tuple (f, args,...) such that f(*args) returns (a copy of) the original object. The following code snippet is the __reduce__() method from sage.rings.integer.Integer.

def __reduce__(self):
    'Including the documentation properly causes a doc-test failure so we include it as a comment:'
    #* '''
    #* This is used when pickling integers.
    #*
    #* EXAMPLES::
    #*
    #*     sage: n = 5
    #*     sage: t = n.__reduce__(); t
    #*     (<built-in function make_integer>, ('5',))
    #*     sage: t[0](*t[1])
    #*     5
    #*     sage: loads(dumps(n)) == n
    #*     True
    #* '''
    # This single line below took me HOURS to figure out.
    # It is the *trick* needed to pickle Cython extension types.
    # The trick is that you must put a pure Python function
    # as the first argument, and that function must return
    # the result of unpickling with the argument in the second
    # tuple as input. All kinds of problems happen
    # if we don't do this.
    return sage.rings.integer.make_integer, (self.str(32),)
sage.misc.persist.save(obj, filename, compress=True, **kwargs)#

Save obj to the file with name filename, which will have an .sobj extension added if it doesn’t have one and if obj doesn’t have its own save() method, like e.g. Python tuples.

For image objects and the like (which have their own save() method), you may have to specify a specific extension, e.g. .png, if you don’t want the object to be saved as a Sage object (or likewise, if filename could be interpreted as already having some extension).

Warning

This will replace the contents of the file if it already exists.

EXAMPLES:

sage: import tempfile
sage: d = tempfile.TemporaryDirectory()
sage: a = matrix(2, [1,2, 3,-5/2])                                              # needs sage.modules
sage: objfile = os.path.join(d.name, 'test.sobj')
sage: objfile_short = os.path.join(d.name, 'test')
sage: save(a, objfile)                                                          # needs sage.modules
sage: load(objfile_short)                                                       # needs sage.modules
[   1    2]
[   3 -5/2]

sage: # needs sage.plot sage.schemes
sage: E = EllipticCurve([-1,0])
sage: P = plot(E)
sage: save(P, objfile_short)   # saves the plot to "test.sobj"
sage: save(P, filename=os.path.join(d.name, "sage.png"), xmin=-2)
sage: save(P, os.path.join(d.name, "filename.with.some.wrong.ext"))
Traceback (most recent call last):
...
ValueError: allowed file extensions for images are
'.eps', '.pdf', '.pgf', '.png', '.ps', '.sobj', '.svg'!
sage: print(load(objfile))
Graphics object consisting of 2 graphics primitives

sage: save("A python string", os.path.join(d.name, 'test'))
sage: load(objfile)
'A python string'
sage: load(objfile_short)
'A python string'
sage: d.cleanup()
sage.misc.persist.unpickle_all(target, debug=False, run_test_suite=False)#

Unpickle all .sobj files in a directory or tar archive.

INPUT:

  • target – a string; the name of a directory or of a (possibly compressed) tar archive that contains a single directory of .sobj files. The tar archive can be in any format that python’s tarfile module understands; for example, .tar.gz or .tar.bz2.

  • debug – a boolean (default: False) whether to report a stacktrace in case of failure

  • run_test_suite – a boolean (default: False) whether to run TestSuite(x).run() on the unpickled objects

OUTPUT:

Typically, two lines are printed: the first reporting the number of successfully unpickled files, and the second reporting the number (zero) of failures. If there are failures, however, then a list of failed files will be printed before either of those lines, and the failure count will of course be non-zero.

Warning

You must only pass trusted data to this function, including tar archives. We use the “data” filter from PEP 706 if possible while extracting the archive, but even that is not a perfect solution, and it is only available since Python 3.11.4.

EXAMPLES:

sage: dir = tmp_dir()
sage: sage.misc.persist.picklejar('hello', dir)
sage: sage.misc.persist.unpickle_all(dir)
Successfully unpickled 1 objects.
Failed to unpickle 0 objects.
sage.misc.persist.unpickle_global(module, name)#

Given a module name and a name within that module (typically a class name), retrieve the corresponding object. This normally just looks up the name in the module, but it can be overridden by register_unpickle_override. This is used in the Sage unpickling mechanism, so if the Sage source code organization changes, register_unpickle_override can allow old pickles to continue to work.

EXAMPLES:

sage: from sage.misc.persist import unpickle_override, register_unpickle_override
sage: unpickle_global('sage.rings.integer', 'Integer')
<class 'sage.rings.integer.Integer'>

Now we horribly break the pickling system:

sage: register_unpickle_override('sage.rings.integer', 'Integer', Rational, call_name=('sage.rings.rational', 'Rational'))
sage: unpickle_global('sage.rings.integer', 'Integer')
<class 'sage.rings.rational.Rational'>

and we reach into the internals and put it back:

sage: del unpickle_override[('sage.rings.integer', 'Integer')]
sage: unpickle_global('sage.rings.integer', 'Integer')
<class 'sage.rings.integer.Integer'>

A meaningful error message with resolution instructions is displayed for old pickles that accidentally got broken because a class or entire module was moved or renamed:

sage: unpickle_global('sage.all', 'some_old_class')
Traceback (most recent call last):
...
ImportError: cannot import some_old_class from sage.all, call
register_unpickle_override('sage.all', 'some_old_class', ...)
to fix this

sage: unpickle_global('sage.some_old_module', 'some_old_class')
Traceback (most recent call last):
...
ImportError: cannot import some_old_class from sage.some_old_module, call
register_unpickle_override('sage.some_old_module', 'some_old_class', ...)
to fix this