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 \(x\) it 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:
sage.misc.persist._BasePickler
Subclass \(pickle.Pickler\) with Sagespecific default options, and builtin support for external object persistence.
INPUT:
file_obj
– a readable filelike object returningbytes
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 crossPython compatibility. Further arguments are passed to
pickle.load()
, where in Python3 Sage sets the defaultencoding='latin1'
. This is essential to make pickles readable in Python3 that were created in Python2. See trac ticket #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 thesage.misc.persist.SagePickler
.INPUT:
obj
 the object to pickle.kwargs
 keyword arguments passed to thesage.misc.persist.SagePickler
constructor.
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)¶ Bases:
sage.misc.persist._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 filelike object returningbytes
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.dumps()
but using thesage.misc.persist.SagePickler
.INPUT:
data
 the pickle data as \(bytes\).kwargs
 keyword arguments passed to thesage.misc.persist.SageUnpickler
constructor.
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
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’simport
.This function also loads a
.sobj
file over a network by specifying the full URL. (Settingverbose = 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 typestr
.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. Ifverbose
is true (the default), then some logging is printed when accessing remote files. Further keyword arguments are passed topickle.load()
.EXAMPLES:
sage: u = 'http://www.sagemath.org/files/test.sobj' sage: s = load(u) # optional  internet Attempting to load remote file: http://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) sage: hello <fortran object>

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
EXAMPLES:
sage: a = matrix(2, [1,2,3,4/3]) sage: s = dumps(a) sage: loads(s) [ 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 Python2 to Python3. The problem is that Python3 by default would not be able to unpickle a nonASCII Python2 string appearing in a pickle. See trac ticket #28444 for details.
sage: class Foo(object): ....: 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 bydumps(Foo('\x80\x07')
in Python2.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 trac ticket #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') #py3 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')) None

sage.misc.persist.
picklejar
(obj, dir=None)¶ Create pickled sobj of
obj
indir
, with name the absolute value of the hash of the pickle of obj. This is used in conjunction withunpickle_all()
.To use this to test the whole Sage library right now, set the environment variable
SAGE_PICKLE_JAR
, which will make it sodumps()
will by default callpicklejar()
with the default dir. Once you do that and doctest Sage, you’ll find that theDOT_SAGE/pickle_jar
directory contains a bunch of pickled objects along with corresponding txt descriptions of them. Use theunpickle_all()
to see if they unpickle later.INPUT:
obj
– a pickleable objectdir
– a string or None; if None thendir
defaults toDOT_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 within_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 thesage
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 aCombinatorialObject
object using a class which also inherits fromElement
. 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 fromElement
which is not compatible with unpickling forCombinatorialObject
. 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) [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 asCombinatorialObject
it will be a tuple. In general, thestate
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(object): ....: def __init__(self,value): ....: self.original_attribute = value ....: def __repr__(self): ....: return 'A(%s)' % self.original_attribute sage: class B(object): ....: 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 thatf(*args)
returns (a copy of) the original object. The following code snippet is the__reduce__()
method fromsage.rings.integer.Integer
.def __reduce__(self): 'Including the documentation properly causes a doctest failure so we include it as a comment:' #* ''' #* This is used when pickling integers. #* #* EXAMPLES:: #* #* sage: n = 5 #* sage: t = n.__reduce__(); t #* (<builtin 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 namefilename
, which will have an.sobj
extension added if it doesn’t have one and ifobj
doesn’t have its ownsave()
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, iffilename
could be interpreted as already having some extension).Warning
This will replace the contents of the file if it already exists.
EXAMPLES:
sage: a = matrix(2, [1,2,3,5/2]) sage: objfile = os.path.join(SAGE_TMP, 'test.sobj') sage: objfile_short = os.path.join(SAGE_TMP, 'test') sage: save(a, objfile) sage: load(objfile_short) [ 1 2] [ 3 5/2] 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(SAGE_TMP, "sage.png"), xmin=2) sage: save(P, os.path.join(SAGE_TMP, "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(SAGE_TMP, 'test')) sage: load(objfile) 'A python string' sage: load(objfile_short) 'A python string'

sage.misc.persist.
unpickle_all
(dir, debug=False, run_test_suite=False)¶ Unpickle all sobj’s in the given directory, reporting failures as they occur. Also printed the number of successes and failure.
INPUT:
dir
– a string; the name of a directory (or of a .tar.bz2 file that decompresses to a directory) full of pickles.debug
– a boolean (default: False) whether to report a stacktrace in case of failurerun_test_suite
– a boolean (default: False) whether to runTestSuite(x).run()
on the unpickled objects
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') <type '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') <type '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') <type '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