3
0
mirror of https://github.com/triqs/dft_tools synced 2024-12-25 13:53:40 +01:00
dft_tools/pytriqs/archive/hdf_archive.py
Michel Ferrero 15f0899016 Fix missing cached key in archives
When an object with __hdf_write__ was written in an archive, the
corresponding key was not cached. Therefore, one could not access
the object which had just been written.
2013-12-23 13:15:27 +01:00

412 lines
17 KiB
Python

################################################################################
#
# TRIQS: a Toolbox for Research in Interacting Quantum Systems
#
# Copyright (C) 2011 by M. Ferrero, O. Parcollet
#
# TRIQS is free software: you can redistribute it and/or modify it under the
# terms of the GNU General Public License as published by the Free Software
# Foundation, either version 3 of the License, or (at your option) any later
# version.
#
# TRIQS is distributed in the hope that it will be useful, but WITHOUT ANY
# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more
# details.
#
# You should have received a copy of the GNU General Public License along with
# TRIQS. If not, see <http://www.gnu.org/licenses/>.
#
################################################################################
import sys,numpy,string
from hdf_archive_basic_layer_h5py import HDFArchiveGroupBasicLayer
from pytriqs.archive.hdf_archive_schemes import hdf_scheme_access, register_class
# -------------------------------------------
#
# Various wrappers for basic python types.
#
# --------------------------------------------
def _my_str(ll, digs = 10) :
ns = str(ll)
for ii in xrange(digs-len(ns)): ns = '0'+ns
return ns
class PythonListWrap:
def __init__(self,ob) :
self.ob = ob
def __reduce_to_dict__(self) :
return dict( [ (_my_str(n),v) for (n,v) in enumerate (self.ob)])
@classmethod
def __factory_from_dict__(cls,D) :
return [x for (n,x) in sorted(D.items())]
class PythonTupleWrap:
def __init__(self,ob) :
self.ob = ob
def __reduce_to_dict__(self) :
return dict( [ (_my_str(n),v) for (n,v) in enumerate (self.ob)])
@classmethod
def __factory_from_dict__(cls,D) :
return tuple([x for (n,x) in sorted(D.items())])
class PythonDictWrap:
def __init__(self,ob) :
self.ob = ob
def __reduce_to_dict__(self) :
return dict( [ (str(n),v) for (n,v) in self.ob.items()])
@classmethod
def __factory_from_dict__(cls,D) :
return dict([(n,x) for (n,x) in D.items()])
register_class (PythonListWrap)
register_class (PythonTupleWrap)
register_class (PythonDictWrap)
# -------------------------------------------
#
# A view of a subgroup of the archive
#
# --------------------------------------------
class HDFArchiveGroup (HDFArchiveGroupBasicLayer) :
"""
"""
_wrappedType = { type([]) : PythonListWrap, type(()) : PythonTupleWrap, type({}) : PythonDictWrap}
_MaxLengthKey = 500
def __init__(self, parent, subpath) :
self.options = parent.options
HDFArchiveGroupBasicLayer.__init__(self, parent, subpath)
self.options = parent.options
self.key_as_string_only = self.options['key_as_string_only']
self._reconstruct_python_objects = self.options['reconstruct_python_object']
self.is_top_level = False
#-------------------------------------------------------------------------
def _key_cipher(self,key) :
if key in self.ignored_keys :
raise KeyError, "key %s is reserved"%key
if self.key_as_string_only : # for bacward compatibility
if type(key) not in [ type('') , type(u'a')] :
raise KeyError, "Key must be string only !"
return key
r = repr(key)
if len (r)> self._MaxLengthKey :
raise KeyError, "The Key is too large !"
# check that the key is ok (it can be reconstructed)
try :
ok = eval(r) == key
except :
ok =False
if not ok : raise KeyError, "The Key *%s*can not be serialized properly by repr !"%key
return r
#-------------------------------------------------------------------------
def _key_decipher(self,key) :
return key if self.key_as_string_only else eval(key)
#-------------------------------------------------------------------------
def __contains__(self,key) :
key= self._key_cipher(key)
return key in self.keys()
#-------------------------------------------------------------------------
def values(self) :
"""
Generator returning the values in the group
"""
def res() :
for name in self.keys() :
yield self[name]
return res()
#-------------------------------------------------------------------------
def items(self) :
"""
Generator returning couples (key, values) in the group.
"""
def res() :
for name in self.keys():
yield name, self[name]
return res()
#-------------------------------------------------------------------------
def __iter__(self) :
"""Returns the keys, like a dictionary"""
def res() :
for name in self.keys() :
yield name
return res()
#-------------------------------------------------------------------------
def __len__(self) :
"""Returns the length of the keys list """
return len(self.keys())
#-------------------------------------------------------------------------
def update(self,object_with_dict_protocol):
for k,v in object_with_dict_protocol.items() : self[k] = v
#-------------------------------------------------------------------------
def __delitem__(self,key) :
key= self._key_cipher(key)
self._clean_key(key,True)
#-------------------------------------------------------------------------
def __setitem__(self,key,val) :
key= self._key_cipher(key)# first look if key is a string or key
if key in self.keys() :
if self.options['do_not_overwrite_entries'] : raise KeyError, "key %s already exists"%key
self._clean_key(key) # clean things
# Transform list, dict, etc... into a wrapped type that will allow HDF reduction
if type(val) in self._wrappedType: val = self._wrappedType[type(val)](val)
# write the attributes
def write_attributes(g) :
"""Use the _hdf5_data_scheme_ if it exists otherwise the class name"""
ds = val._hdf5_data_scheme_ if hasattr(val,"_hdf5_data_scheme_") else val.__class__.__name__
try :
sch = hdf_scheme_access(ds)
except :
err = """
You are trying to store an object of type "%s", with the TRIQS_HDF5_data_scheme "%s".
But that data_scheme is not registered, so you will not be able to reread the class.
Didn't you forget to register your class in pytriqs.archive.hdf_archive_schemes?
""" %(val.__class__.__name__,ds)
raise IOError,err
g.write_attr("TRIQS_HDF5_data_scheme", ds)
if '__write_hdf5__' in dir(val) : # simplest protocol
val.__write_hdf5__(self._group,key)
self.cached_keys.append(key) # I need to do this here
SUB = HDFArchiveGroup(self,key)
write_attributes(SUB)
elif '__reduce_to_dict__' in dir(val) : # Is it a HDF_compliant object
self.create_group(key) # create a new group
d = val.__reduce_to_dict__() if '__reduce_to_dict__' in dir(val) else dict( [(x,getattr(val,x)) for x in val.__HDF_reduction__])
if type(d) != type({}) : raise ValueError, " __reduce_to_dict__ method does not return a dict. See the doc !"
if (d=={}) : raise ValueError, "__reduce_to_dict__ returns an empty dict"
SUB = HDFArchiveGroup(self,key)
for n,v in d.items() : SUB[n] = v
write_attributes(SUB)
elif type(val)== numpy.ndarray : # it is a numpy
try :
self._write_array( key, numpy.array(val,copy=1,order='C') )
except RuntimeError:
print "HDFArchive is in trouble with the array %s"%val
raise
elif isinstance(val, HDFArchiveGroup) : # will copy the group recursively
# we could add this for any object that has .items() in fact...
SUB = HDFArchiveGroup(self, key)
for k,v in val.items() : SUB[k]=v
else : # anything else... expected to be a scalar
try :
self._write_scalar( key, val)
except:
raise #ValueError, "Value %s\n is not of a type suitable to storage in HDF file"%val
self._flush()
#-------------------------------------------------------------------------
def get_raw (self,key):
"""Similar to __getitem__ but it does NOT reconstruct the python object,
it presents it as a subgroup"""
return self.__getitem1__(key,False)
#-------------------------------------------------------------------------
def __getitem__(self,key) :
"""Return the object key, possibly reconstructed as a python object if
it has been properly set up"""
return self.__getitem1__(key,self._reconstruct_python_objects)
#-------------------------------------------------------------------------
def __getitem1__(self,key,reconstruct_python_object) :
if key not in self :
key = self._key_cipher(key)
if key not in self : raise KeyError, "Key %s does not exists"%key
if self.is_group(key) :
SUB = HDFArchiveGroup(self,key) # View of the subgroup
if not reconstruct_python_object : return SUB
try :
hdf_data_scheme = SUB.read_attr("TRIQS_HDF5_data_scheme")
except:
return SUB
if hdf_data_scheme :
try :
sch = hdf_scheme_access(hdf_data_scheme)
except :
print "Warning : The TRIQS_HDF5_data_scheme %s is not recognized. Returning as a group. Hint : did you forgot to import this python class ?"%hdf_data_scheme
return SUB
r_class_name = sch.classname
r_module_name = sch.modulename
r_readfun = sch.read_fun
if not (r_class_name and r_module_name) : return SUB
try :
exec("from %s import %s as r_class" %(r_module_name,r_class_name)) in globals(), locals()
except KeyError :
raise RuntimeError, "I can not find the class %s to reconstruct the object !"%r_class_name
if r_readfun :
res = r_readfun(self._group,key)
elif "__factory_from_dict__" in dir(r_class) :
f = lambda K : SUB.__getitem1__(K,reconstruct_python_object) if SUB.is_group(K) else SUB._read(K)
values = dict( (self._key_decipher(K),f(K)) for K in SUB )
res = r_class.__factory_from_dict__(values)
else :
raise ValueError, "Impossible to reread the class %s for group %s and key %s"%(r_class_name,self, key)
return res
elif self.is_data(key) :
return self._read(key)
else :
raise KeyError, "Key %s is of unknown type !!"%Key
#---------------------------------------------------------------------------
def __str__(self) :
def pr(name) :
if self.is_group(name) :
return "%s : subgroup"%name
elif self.is_data(name) : # can be an array of a number
return "%s : data "%(name)
else :
raise ValueError, "oopps %s"%(name)
s= "HDFArchive%s with the following content:\n"%(" (partial view)" if self.is_top_level else '')
s+=string.join([ ' '+ pr(n) for n in self.keys() ], '\n')
return s
#-------------------------------------------------------------------------
def __repr__(self) :
return self.__str__()
#-------------------------------------------------------------------------
def apply_on_leaves (self,f) :
"""
For each named leaf (name,value) of the tree, it calls f(name,value)
f should return :
- `None` : no action is taken
- an `empty tuple` () : the leaf is removed from the tree
- an hdf-compliant value : the leaf is replaced by the value
"""
def visit_tree(n,d):
for k in d:# Loop over the subgroups in d
if d.is_group(k) : visit_tree(k,d[k])
else :
r = f(k,d[k])
if r != None : d[k] = r
elif r == () : del d[k]
visit_tree('/',self['/'])
# -------------------------------------------
#
# The main class
#
# --------------------------------------------
class HDFArchive(HDFArchiveGroup):
"""
"""
_class_version = "HDFArchive | 1.0"
def __init__(self, url_name, open_flag = 'a', key_as_string_only = True,
reconstruct_python_object = True, init = {}):
r"""
Parameters
-----------
url_name : string
The url of the hdf5 file.
* If url is a simple string, it is interpreted as a local file name
* If url is a remote url (e.g. `http://ipht.cea.fr/triqs/data/single_site_bethe.output.h5` )
then the h5 file is downloaded in temporary file and opened.
In that case, ``open_flag`` must be 'r', read-only mode.
The temporary file is deleted at exit.
open_flag : Legal modes: r, w, a (default)
key_as_string_only : True (default)
init : any generator of tuple (key,val), e.g. a dict.items().
It will fill the archive with these values.
Attributes
----------
LocalFileName : string
the name of the file or of the local downloaded copy
url_name : string
the name of the Url
Examples
--------
>>> # retrieve a remove archive (in read-only mode) :
>>> h = HDFArchive( 'http://ipht.cea.fr/triqs/data/single_site_bethe.output.h5')
>>>
>>> # full copy of an archive
>>> HDFArchive( f, 'w', init = HDFArchive(fmp,'r').items()) # full
>>>
>>> # partial copy of file of name fmp, with only the key 'G'
>>> HDFArchive( f, 'w', init = [ (k,v) for (k,v) in HDFArchive(fmp,'r') if k in ['G'] )
>>>
>>> # faster version : the object are only retrieved when needed (list comprehension vs iterator comprehension)
>>> HDFArchive( f, 'w', init = ( (k,v) for (k,v) in HDFArchive(fmp,'r') if k in ['G'] ) )
>>>
>>> # partial copy with processing on the fly with the P function
>>> HDFArchive( f, 'w', init = ( (k,P(v)) for (k,v) in HDFArchive(fmp,'r') if k in ['G'] ) )
>>>
>>> # another variant with a filtered dict
>>> HDFArchive( f, 'w', init = HDFArchive(fmp,'r').items(lambda k : k in ['G'] ))
"""
import os,os.path
assert open_flag in ['r','w','a'], "Invalid mode"
assert type(url_name)==type(''), "url_name must be a string"
# If it is an url , retrieve if and check mode is read only
import urllib
LocalFileName, http_message = urllib.urlretrieve (url_name) if open_flag == 'r' else (url_name, None)
if LocalFileName != url_name : # this was not a local file, so it must be read only
assert open_flag == 'r', "You retrieve a distant Url %s which is not local, so it must be read-only. Use 'r' option"%url_name
if open_flag == 'w' :
os.system("rm -f %s"%(os.path.abspath(LocalFileName))) # destroys the file, ignoring errors
self._init_root( LocalFileName, open_flag)
self.options = {'key_as_string_only' : key_as_string_only,
'do_not_overwrite_entries' : False,
'reconstruct_python_object': reconstruct_python_object,
'UseAlpsNotationForComplex' : True
}
HDFArchiveGroup.__init__(self,self,"")
self.is_top_level = True
for k,v in init : self[k]=v
# These two methods are necessary for "with"
def __enter__(self): return self
def __exit__(self, type, value, traceback):
self._flush()
self._close()
#--------------------------------------------------------------------------------
class HDFArchiveInert:
"""
A fake class for the node in MPI. It does nothing, but
permits to write simply :
a= mpi.bcast(H['a']) # run on all nodes
-[] : __getitem__ returns self so that H['a']['b'] is ok...
- setitem : does nothing.
"""
def HDFArchive_Inert(self):
pass
def __getitem__(self,x) : return self
def __setitem__(self,k,v) : pass
#--------------------------------------------------------------------------------