B |ba4@sddlmZddlZddlZddlZddlZddlZddlZddl m Z ddl m Z ddl m Z mZmZddlmZmZdd gZejd krd nd Zd dZGddde ZGdddeZGdddeZGdddeZGdddeZGdddeZGdddeZddZdd ZdS))absolute_importN)compat) numeric_types) BaseHandlerregister unregister) b64decode b64encoderegister_handlersunregister_handlerslittle<>cCs|jj}|dkrtS|S)z%translate equals sign to native order=)dtype byteordernative_byteorder)arrrr3lib/python3.7/site-packages/jsonpickle/ext/numpy.py get_byteordersrc@seZdZddZddZdS)NumpyBaseHandlercCsLt|dr||d<n0t|}d}||r@|t|d}||d<dS)Ntostringrz(numpy.record, )hasattrrrZustr startswithlen)selfrdataprefixrrr flatten_dtypes   zNumpyBaseHandler.flatten_dtypecCs&|d}|drt|}t|S)Nr){[)rastZ literal_evalnpr)rrrrrr restore_dtype&s  zNumpyBaseHandler.restore_dtypeN)__name__ __module__ __qualname__r!r&rrrrrs rc@seZdZddZddZdS)NumpyDTypeHandlercCs||||S)N)r!)robjrrrrflatten.s zNumpyDTypeHandler.flattencCs ||S)N)r&)rrrrrrestore2szNumpyDTypeHandler.restoreN)r'r(r)r,r-rrrrr*-sr*c@seZdZddZddZdS)NumpyGenericHandlercCs0||jd||jj|dd|d<|S)NNF)resetvalue)r!r newbyteordercontextr,tolist)rr+rrrrr,7szNumpyGenericHandler.flattencCs$|jj|ddd}|||S)Nr1F)r0)r3r-r&type)rrr1rrrr-<szNumpyGenericHandler.restoreN)r'r(r)r,r-rrrrr.6sr.cs(eZdZdZfddZddZZS)UnpickleableNumpyGenericHandlerz From issue #381, this is used for simplifying the output of numpy arrays when unpicklable=False (the default is True). cs2|jjs|jj|ddStt|||SdS)NF)r0)r3Z unpicklabler,r4superr.)rr+r) __class__rrr,Gsz'UnpickleableNumpyGenericHandler.flattencCstdS)N)NotImplementedError)rrrrrr-Msz'UnpickleableNumpyGenericHandler.restore)r'r(r)__doc__r,r- __classcell__rr)r8rr6As r6c@s0eZdZdZddZddZddZdd Zd S) NumpyNDArrayHandlerz>Stores arrays as text representation, without regard for viewscCs|jjdkrd|d<dS)NF writeable)flagsr=)rr+rrrr flatten_flagsTs z!NumpyNDArrayHandler.flatten_flagscCs|ddsd|j_dS)Nr=TF)getr>r=)rrrrrr restore_flagsXs z!NumpyNDArrayHandler.restore_flagscCsP||jd|||||jj|dd|d<d|jkrL|j|d<|S)Nr/F)r0valuesrshape)r!rr2r?r3r,r4rC)rr+rrrrr,\s    zNumpyNDArrayHandler.flattencCs`|jj|ddd}tj||||ddd}|dd}|dk rP||}||||S)NrBF)r0orderC)rrDrC)r3r-r%arrayr&r@reshaperA)rrrBrrCrrrr-fs   zNumpyNDArrayHandler.restoreN)r'r(r)r:r?rAr,r-rrrrr<Qs  r<csJeZdZdZdefddZddZddZfd d Zfd d Z Z S) NumpyNDArrayHandlerBinaryzstores arrays with size greater than 'size_threshold' as (optionally) compressed base64 Notes ----- This would be easier to implement using np.save/np.load, but that would be less language-agnostic cCs||_||_dS)a :param size_threshold: nonnegative int or None valid values for 'size_threshold' are all nonnegative integers and None if size_threshold is None, values are always stored as nested lists :param compression: a compression module or None valid values for 'compression' are {zlib, bz2, None} if compression is None, no compression is applied N)size_threshold compression)rrJrKrrr__init__}s z"NumpyNDArrayHandlerBinary.__init__cCs |jj}|dkrt||d<dS)N|r)rrr)rr+rrrrrflatten_byteordersz+NumpyNDArrayHandlerBinary.flatten_byteordercCs"|dd}|r|j||_dS)Nr)r@rr2)rrrrrrrrestore_byteorders z+NumpyNDArrayHandlerBinary.restore_byteordercs|jdks|j|jkr*tt|||}n|jtkrHt|  }n$t |dr`|j dd}n |j dd}|jr~|j|}t||d<|j|d<||jd||||||||jjsd|d <|S) zencode numpy to jsonNtobytesa)rDrBrCr/FrD)rJsizer7rHr,robjectjsondumpsr4encoderrPrrKcompressr rCr!r2rNr?r> c_contiguous)rr+rbuf)r8rrr,s"        z!NumpyNDArrayHandlerBinary.flattencs|d}t|tr$tt||}nt|trFtj|g||d}n||}t |}|j rj|j |}|t krt |}tj|||ddd}|dd}|dk r||}n0tj|||d|ddd}|||||||S) zdecode numpy from jsonrB)rrDrE)rrDrCN)bufferrrCrD) isinstancelistr7rHr-rr%rFr&r rKZ decompressrTrUloadsdecoder@rGndarraycopyrOrA)rrrBrrrZrC)r8rrr-s.        z!NumpyNDArrayHandlerBinary.restore) r'r(r)r:zlibrLrNrOr,r-r;rr)r8rrHss   *rHcs@eZdZdZddeffdd ZfddZfdd ZZS) NumpyNDArrayHandlerViewaPickles references inside ndarrays, or array-views Notes ----- The current implementation has some restrictions. 'base' arrays, or arrays which are viewed by other arrays, must be f-or-c-contiguous. This is not such a large restriction in practice, because all numpy array creation is c-contiguous by default. Relaxing this restriction would be nice though; especially if it can be done without bloating the design too much. Furthermore, ndarrays which are views of array-like objects implementing __array_interface__, but which are not themselves nd-arrays, are deepcopied with a warning (by default), as we cannot guarantee whatever custom logic such classes implement is correctly reproduced. warnrIcstt|||||_dS)aO :param mode: {'warn', 'raise', 'ignore'} How to react when encountering array-like objects whose references we cannot safely serialize :param size_threshold: nonnegative int or None valid values for 'size_threshold' are all nonnegative integers and None if size_threshold is None, values are always stored as nested lists :param compression: a compression module or None valid values for 'compression' are {zlib, bz2, None} if compression is None, no compression is applied N)r7rcrLmode)rrerJrK)r8rrrLs z NumpyNDArrayHandlerView.__init__csl|j}|dkr<|jjrforcr7rcr,rYr\r%r`r3ZctypesrrhrCr!rr2r?rrJrSrHrewarningsrd ValueErrorra)rr+rrfrgrmsg)r8rrr,s:        zNumpyNDArrayHandlerView.flattenc s|dd}|dkr&tt||}nl|jj|dd}|jjsFtdtj |j | | |dd|d|d d |d dd }| |||S) zdecode numpy from jsonrfNF)r0z8Current implementation assumes base is C or F contiguousrrMrCrgrrh)r[rrCrgrh)r@r7rcr-r3r>rkAssertionErrorr%r`rr&r2rA)rrrfr)r8rrr->s    zNumpyNDArrayHandlerView.restore) r'r(r)r:rbrLr,r-r;rr)r8rrcs 6rccCsttjtddttjtddttjtddtttjj tddtttj j tddtttj j tddtttj j tdddS)NT)rf) rr%rr*genericr.r`rcvoidr8float32int32 datetime64rrrrr XscCsjttjttjttjtttjjtttjjtttjjtttj jdS)N) rr%rrpr`rqr8rrrsrtrrrrr cs   ) Z __future__rr$rUsysrlrbZnumpyr%rrZhandlersrrrutilr r __all__rrrrr*r.r6r<rHrcr r rrrrs,     "ov