"""
Introduction
============

Python module for reading and writing GRIB (editions 1 and 2) files.
GRIB is the World Meterological Organization
U{standard<http://www.wmo.ch/pages/prog/www/WMOCodes/GRIB.html>} 
for distributing gridded data. 
The module is a python interface to the
U{GRIB API<http://www.ecmwf.int/products/data/software/grib_api.html>} C library
from the European Centre for Medium-Range Weather Forecasts
(U{ECMWF<http://www.ecmwf.int>}).

Required
========

- U{Python<http://python.org>} 2.4 or higher.
- U{numpy<http://sourceforge.net/project/showfiles.php?group_id=1369>}
  N-dimensional array object for python. Version 1.2.1 or higher (version
  1.5.1 required for Python 3).
- U{pyproj<http://code.google.com/p/pyproj/>} Python interface to 
  U{PROJ.4<http://trac.osgeo.org/proj>} library for cartographic
  transformations B{or} U{matplotlib<http://matplotlib.sf.net>} and
  the U{basemap<http://matplotlib.sf.net/basemap/doc/html>} toolkit.
  Pyproj 1.8.9 is required for Python 3.
- U{GRIB API<http://www.ecmwf.int/products/data/software/grib_api.html>} C library
  for encoding and decoding GRIB messages (edition 1 and edition 2).
  Version 1.8.0 or higher required.
  To be fully functional, the GRIB API library requires
  U{Jasper<http://www.ece.uvic.ca/~mdadams/jasper>} or 
  U{OpenJPEG<http://www.openjpeg.org>} for JPEG200 encoding,
  and U{PNG<http://www.libpng.org/pub/png/libpng.html>} for PNG encoding.

These dependencies are available via the package management system of most
Linux distributions, and on MacOS X using U{macports<http://macports.org/>}. 
If you build grib_api yourself as a static library on a 64-bit system
you may need to set C{CFLAGS} to C{'-O2 -fPIC'} before running the C{configure}
script.  To use pygrib on Windows, you must use the
U{Cygwin<http://cygwin.com>} environment, since the grib_api library requires a 
posix environment.  Cygwin installation instructions are available
U{here<http://code.google.com/p/pygrib/wiki/CygwinWindowsInstall>}.

Installation
============

 - U{Download<http://code.google.com/p/pygrib/downloads/list>} the source code. 
 - pygrib installation options can either be set with environment variables,
 or specified in a text file (setup.cfg).  To use environment variables,
 set C{$GRIBAPI_DIR}, C{$JASPER_DIR}, C{$OPENJPEG_DIR},
 C{$PNG_DIR} and C{$ZLIB_DIR} so that the include files and libraries for
 GRIB API, Jasper, OpenJPEG, PNG and zlib will be found.  
 For example, the include files for 
 jasper should be found in C{$JASPER_DIR/include} or
 C{$JASPER_DIR/include/jasper}, and the jasper
 library should be found in C{$JASPER_DIR/lib} or C{$JASPER_DIR/lib64}. If any of
 those environment variables are not set, then it is assumed that GRIB API was
 not built with support for that library.
 If the libraries and
 include files are installed in separate locations, the environment variables
 C{$GRIBAPI_INCDIR} and C{$GRIBAPI_LIBDIR} can be used to define the locations
 separately (same goes for C{JASPER}, C{OPENJPEG}, C{PNG} and C{ZLIB}).
 To use setup.cfg, copy setup.cfg.template to setup.cfg, open setup.cfg in a 
 text editor and follow the instructions in the comments for editing.
 - Run 'python setup.py build' and then 'python setup.py install', as root if necessary.
 Note that if you are using environment variables to specify the build options,
 you cannot build and install in one step with 'sudo python setup.py install',
 since sudo does not pass environment variables.  Instead, run 'python setup.py
 build' first as a regular user, then run 'sudo python setup.py install' if the
 install directory requires admin or root privileges.
 - Run 'python test.py' to test your installation.
 - Look at examples in C{test} directory (most require 
 U{matplotlib<http://matplotlib.sf.net>} and
 U{basemap<http://matplotlib.sourceforge.net/basemap/doc/html/>}).
 - If you're on MacOS X, see
 U{README.macosx<http://pygrib.googlecode.com/svn/trunk/README.macosx>}
 for special instruction on how to install pygrib and all it's
 dependencies using U{macports<http://macports.org>}.


Example usage
=============

 - from the python interpreter prompt, import the package::
    >>> import pygrib
 - open a GRIB file, create a grib message iterator::
    >>> grbs = pygrib.open('sampledata/flux.grb')  
 - pygrib open instances behave like regular python file objects, with
 C{seek}, C{tell}, C{read}, C{readline} and C{close} methods, except that offsets
 are measured in grib messages instead of bytes::
    >>> grbs.seek(2)
    >>> grbs.tell()
    2
    >>> grb = grbs.read(1)[0] # read returns a list with the next N (N=1 in this case) messages.
    >>> grb # printing a grib message object displays summary info
    3:Maximum temperature:K (instant):regular_gg:heightAboveGround:level 2 m:fcst time 108-120 hrs:from 200402291200
    >>> grbs.tell()
    3
 - print an inventory of the file::
    >>> grbs.seek(0)
    >>> for grb in grbs:
    >>>     grb 
    1:Precipitation rate:kg m**-2 s**-1 (avg):regular_gg:surface:level 0:fcst time 108-120 hrs (avg):from 200402291200
    2:Surface pressure:Pa (instant):regular_gg:surface:level 0:fcst time 120 hrs:from 200402291200
    3:Maximum temperature:K (instant):regular_gg:heightAboveGround:level 2 m:fcst time 108-120 hrs:from 200402291200
    4:Minimum temperature:K (instant):regular_gg:heightAboveGround:level 2 m:fcst time 108-120 hrs:from 200402291200
 - find the first grib message with a matching name::
    >>> grb = grbs.select(name='Maximum temperature')[0]
 - extract the data values using the 'values' key
 (grb.keys() will return a list of the available keys)::
    # The data is returned as a numpy array, or if missing values or a bitmap
    # are present, a numpy masked array.  Reduced lat/lon or gaussian grid
    # data is automatically expanded to a regular grid. Details of the internal
    # representation of the grib data (such as the scanning mode) are handled
    # automatically.
    >>> maxt = grb.values # same as grb['values']
    >>> maxt.shape, maxt.min(), maxt.max()
    (94, 192) 223.7 319.9
 - get the latitudes and longitudes of the grid::
    >>> lats, lons = grb.latlons()
    >>> lats.shape, lats.min(), lats.max(), lons.shape, lons.min(), lons.max()
    (94, 192) -88.5419501373 88.5419501373  0.0 358.125
 - get the second grib message::
    >>> grb = grbs.message(2) # same as grbs.seek(1); grb=grbs.readline()
    >>> grb
    2:Surface pressure:Pa (instant):regular_gg:surface:level 0:fcst time 120 hrs:from 200402291200
 - modify the values associated with existing keys (either via attribute or
 dictionary access)::
    >>> grb['forecastTime'] = 240
    >>> grb.dataDate = 20100101
 - get the binary string associated with the coded message::
    >>> msg = grb.tostring()
    >>> grbs.close() # close the grib file.
 - write the modified message to a new GRIB file::
    >>> grbout = open('test.grb','wb')
    >>> grbout.write(msg)
    >>> grbout.close()
    >>> pygrib.open('test.grb').readline() 
    1:Surface pressure:Pa (instant):regular_gg:surface:level 0:fcst time 240 hrs:from 201001011200

Documentation
=============

 - see below for the full python API documentation.
  
Changelog
=========

 - see U{Changelog<http://pygrib.googlecode.com/svn/trunk/Changelog>} file.

@author: Jeffrey Whitaker.

@contact: U{Jeff Whitaker<mailto:jeffrey.s.whitaker@noaa.gov>}

@version: 1.9.4

@copyright: copyright 2010 by Jeffrey Whitaker.

@license: Permission to use, copy, modify, and distribute this software and its
documentation for any purpose and without fee is hereby granted,
provided that the above copyright notice appear in all copies and that
both that copyright notice and this permission notice appear in
supporting documentation.
THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE,
INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO
EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, INDIRECT OR
CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF
USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR
OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
PERFORMANCE OF THIS SOFTWARE."""
__test__ = None
del __test__ # hack so epydoc doesn't show __test__
__version__ = '1.9.4'

import numpy as np
from datetime import datetime
from numpy import ma
try:
    import pyproj
except ImportError:
    try:
        from mpl_toolkits.basemap import pyproj
    except:
        raise ImportError("either pyproj or basemap required")
from ncepgrib2 import Grib2Decode
import_array()

cdef extern from "stdlib.h":
    ctypedef long size_t
    void *malloc(size_t size)
    void free(void *ptr)

cdef extern from "stdio.h":
    ctypedef struct FILE
    FILE *fopen(char *path, char *mode)
    int	fclose(FILE *)
    size_t fwrite(void *ptr, size_t size, size_t nitems, FILE *stream)
    void rewind (FILE *)

cdef extern from "Python.h":
    object PyBytes_FromStringAndSize(char *s, size_t size)
default_encoding = 'ascii'

cdef extern from "numpy/arrayobject.h":
    ctypedef int npy_intp 
    ctypedef extern class numpy.ndarray [object PyArrayObject]:
        cdef char *data
        cdef int nd
        cdef npy_intp *dimensions
        cdef npy_intp *strides
        cdef object base
        cdef int flags
    npy_intp PyArray_SIZE(ndarray arr)
    npy_intp PyArray_ISCONTIGUOUS(ndarray arr)
    npy_intp PyArray_ISALIGNED(ndarray arr)
    void import_array()

cdef extern from "grib_api.h":
    ctypedef struct grib_handle
    ctypedef struct grib_index
    ctypedef struct grib_keys_iterator
    ctypedef struct grib_context
    cdef enum:
        GRIB_TYPE_UNDEFINED
        GRIB_TYPE_LONG
        GRIB_TYPE_DOUBLE
        GRIB_TYPE_STRING
        GRIB_TYPE_BYTES 
        GRIB_TYPE_SECTION 
        GRIB_TYPE_LABEL 
        GRIB_TYPE_MISSING 
        GRIB_KEYS_ITERATOR_ALL_KEYS            
        GRIB_KEYS_ITERATOR_SKIP_READ_ONLY         
        GRIB_KEYS_ITERATOR_SKIP_OPTIONAL          
        GRIB_KEYS_ITERATOR_SKIP_EDITION_SPECIFIC  
        GRIB_KEYS_ITERATOR_SKIP_CODED             
        GRIB_KEYS_ITERATOR_SKIP_COMPUTED         
        GRIB_KEYS_ITERATOR_SKIP_FUNCTION         
        GRIB_KEYS_ITERATOR_SKIP_DUPLICATES       
        GRIB_MISSING_LONG 
        GRIB_MISSING_DOUBLE
    int grib_get_size(grib_handle *h, char *name, size_t *size)
    int grib_get_native_type(grib_handle *h, char *name, int *type)
    int grib_get_long(grib_handle *h, char *name, long *ival)
    int grib_set_long(grib_handle *h, char *name, long val)
    int grib_get_long_array(grib_handle *h, char *name, long *ival, size_t *size)
    int grib_set_long_array(grib_handle *h, char *name, long *ival, size_t size)
    int grib_get_double(grib_handle *h, char *name, double *dval)
    int grib_set_double(grib_handle *h, char *name, double dval)
    int grib_get_double_array(grib_handle *h, char *name, double *dval, size_t *size)
    int grib_set_double_array(grib_handle *h, char *name, double *dval, size_t size)
    int grib_get_string(grib_handle *h, char *name, char *mesg, size_t *size)
    int grib_set_string(grib_handle *h, char *name, char *mesg, size_t *size)
    grib_keys_iterator* grib_keys_iterator_new(grib_handle* h,unsigned long filter_flags, char* name_space)
    int grib_keys_iterator_next(grib_keys_iterator *kiter)
    char* grib_keys_iterator_get_name(grib_keys_iterator *kiter)
    int grib_handle_delete(grib_handle* h)
    grib_handle* grib_handle_new_from_file(grib_context* c, FILE* f, int* error)        
    char* grib_get_error_message(int code)
    int grib_keys_iterator_delete( grib_keys_iterator* kiter)
    void grib_multi_support_on(grib_context* c)
    void grib_multi_support_off(grib_context* c)
    int grib_get_message(grib_handle* h ,  void** message,size_t *message_length)
    int grib_get_message_copy(grib_handle* h ,  void* message,size_t *message_length)
    long grib_get_api_version()
    grib_handle* grib_handle_clone(grib_handle* h)
    grib_index* grib_index_new_from_file(grib_context* c,\
                            char* filename,char* keys,int *err)
    int grib_index_get_size(grib_index* index,char* key,size_t* size)
    int grib_index_get_long(grib_index* index,char* key,\
                        long* values,size_t *size)
    int grib_index_get_double(grib_index* index,char* key,\
                              double* values,size_t *size)
    int grib_index_get_string(grib_index* index,char* key,\
                          char** values,size_t *size)
    int grib_index_select_long(grib_index* index,char* key,long value)
    int grib_index_select_double(grib_index* index,char* key,double value)
    int grib_index_select_string(grib_index* index,char* key,char* value)
    grib_handle* grib_handle_new_from_index(grib_index* index,int *err)
    void grib_index_delete(grib_index* index)
    int grib_is_missing(grib_handle* h, char* key, int* err)
    grib_handle* grib_handle_new_from_message(grib_context * c, void * data,\
                             size_t data_len)
    grib_handle* grib_handle_new_from_message_copy(grib_context * c, void * data,\
                             size_t data_len)
    int grib_julian_to_datetime(double jd, long *year, long *month, long *day, long *hour, long *minute, long *second)
    int grib_datetime_to_julian(long year, long month, long day, long hour, long minute, long second, double *jd)
    int grib_get_gaussian_latitudes(long truncation,double* latitudes)
    int grib_index_write(grib_index *index, char *filename)
    grib_index* grib_index_read(grib_context* c, char* filename,int *err)


missingvalue_int = GRIB_MISSING_LONG
#this doesn't work, since defined constants are assumed to be integers
#missingvalue_float = GRIB_MISSING_DOUBLE
missingvalue_float = -1.e100 # value given in grib_api.h version 1.90
def _get_grib_api_version():
    div = lambda v,d: (v/d,v%d)
    v = grib_get_api_version()
    v,revision = div(v,100)
    v,minor = div(v,100)
    major = v
    return "%d.%d.%d" % (major,minor,revision)
grib_api_version = _get_grib_api_version()

def gaulats(object nlats):
    """
    gaulats(nlats)

    Returns nlats gaussian latitudes, in degrees, oriented from
    north to south.  nlats must be even."""
    cdef ndarray lats
    if nlats%2:
        raise ValueError('nlats must be even')
    lats = np.empty(nlats, np.float64)
    grib_get_gaussian_latitudes(<long>nlats/2, <double *>lats.data)
    return lats

# dict for forecast time units (Code Table 4.4).
_ftimedict = {}
_ftimedict[0]='mins'
_ftimedict[1]='hrs'
_ftimedict[2]='days'
_ftimedict[3]='months'
_ftimedict[4]='yrs'
_ftimedict[5]='decades'
_ftimedict[6]='30 yr periods'
_ftimedict[7]='centuries'
_ftimedict[10]='3 hr periods'
_ftimedict[11]='6 hr periods'
_ftimedict[12]='12 hr periods'
_ftimedict[13]='secs'

# turn on support for multi-field grib messages.
grib_multi_support_on(NULL)

cdef class open(object):
    """ 
    open(filename)
    
    returns GRIB file iterator object given GRIB filename. When iterated, returns
    instances of the L{gribmessage} class. Behaves much like a python file
    object, with L{seek}, L{tell}, L{read} and L{close} methods, 
    except that offsets are measured in grib messages instead of bytes.
    Additional methods include L{rewind} (like C{seek(0)}), L{message}
    (like C{seek(N-1)}; followed by C{readline()}), and L{select} (filters
    messages based on specified conditions).  The C{__call__} method forwards
    to L{select}, and instances can be sliced with C{__getitem__} (returning
    lists of L{gribmessage} instances).  The position of the iterator is not
    altered by slicing with C{__getitem__}.
     
    @ivar messages: The total number of grib messages in the file.

    @ivar messagenumber: The grib message number that the iterator currently
    points to (the value returned by L{tell}).

    @ivar name: The GRIB file which the instance represents."""
    cdef FILE *_fd
    cdef grib_handle *_gh
    cdef public object name, messagenumber, messages, closed
    def __cinit__(self, filename):
        # initialize C level objects.
        cdef grib_handle *gh
        cdef FILE *_fd
        bytestr = _strencode(filename)
        self._fd = fopen(bytestr, "rb") 
        if self._fd == NULL:
            raise IOError("could not open %s", filename)
        self._gh = NULL
    def __init__(self, filename):
        cdef int err
        cdef grib_handle *gh
        # initalize Python level objects
        self.name = filename
        self.closed = False
        self.messagenumber = 0
        # count number of messages in file.
        nmsgs = 0
        while 1:
            gh = grib_handle_new_from_file(NULL, self._fd, &err)
            err = grib_handle_delete(gh)
            if gh == NULL: break
            nmsgs = nmsgs + 1
        rewind(self._fd)
        self.messages = nmsgs 
    def __iter__(self):
        return self
    def __next__(self):
        cdef grib_handle* gh 
        cdef int err
        if self.messagenumber == self.messages:
            raise StopIteration
        if self._gh is not NULL:
            err = grib_handle_delete(self._gh)
            if err:
                raise RuntimeError(grib_get_error_message(err))
        gh = grib_handle_new_from_file(NULL, self._fd, &err)
        if err:
            raise RuntimeError(grib_get_error_message(err))
        if gh == NULL:
            raise StopIteration
        else:
            self._gh = gh
            self.messagenumber = self.messagenumber + 1
        return _create_gribmessage(self._gh, self.messagenumber)
    def __getitem__(self, key):
        if type(key) == slice:
            # for a slice, return a list of grib messages.
            beg, end, inc = key.indices(self.messages)
            msg = self.tell()
            grbs = [self.message(n+1) for n in xrange(beg,end,inc)]
            self.seek(msg) # put iterator back in original position
            return grbs
        elif type(key) == int or type(key) == long:
            # for an integer, return a single grib message.
            msg = self.tell()
            grb = self.message(key)
            self.seek(msg) # put iterator back in original position
            return grb
        else:
            raise KeyError('key must be an integer message number or a slice')
    def __call__(self, **kwargs):
        """same as L{select}"""
        return self.select(**kwargs)
    def __enter__(self):
        return self
    def __exit__(self,atype,value,traceback):
        self.close()
    def tell(self):
        """returns position of iterator (grib message number, 0 means iterator
        is positioned at beginning of file)."""
        return self.messagenumber
    def seek(self, msg, from_what=0):
        """
        seek(N,from_what=0)
        
        advance iterator N grib messages from beginning of file 
        (if C{from_what=0}), from current position (if C{from_what=1})
        or from the end of file (if C{from_what=2})."""
        if from_what not in [0,1,2]:
            raise ValueError('from_what keyword arg to seek must be 0,1 or 2')
        if msg == 0:
            if from_what == 0:
                self.rewind()
            elif from_what == 1:
                return
            elif from_what == 2:
                self.message(self.messages)
        else:
            if from_what == 0:
                self.message(msg)
            elif from_what == 1:
                self.message(self.messagenumber+msg)
            elif from_what == 2:
                self.message(self.messages+msg)
    def readline(self):
        """
        readline()

        read one entire grib message from the file.
        Returns a L{gribmessage} instance, or None if an EOF is encountered."""
        try:
            if hasattr(self,'next'):
                grb = self.next()
            else:
                grb = next(self)
        except StopIteration:
            grb = None
        return grb
    def read(self,msgs=None):
        """
        read(N=None)
        
        read N messages from current position, returning grib messages instances in a
        list.  If N=None, all the messages to the end of the file are read.
        C{pygrib.open(f).read()} is equivalent to C{list(pygrib.open(f))},
        both return a list containing L{gribmessage} instances for all the
        grib messages in the file C{f}.
        """
        if msgs is None:
            grbs = self._advance(self.messages-self.messagenumber,return_msgs=True)
        else:
            grbs = self._advance(msgs,return_msgs=True)
        return grbs
    def close(self):
        """
        close()

        close GRIB file, deallocate C structures associated with class instance"""
        # doesn't have __dealloc__, user must explicitly call close.
        # having both causes segfaults.
        cdef int err
        fclose(self._fd)
        if self._gh != NULL:
            err = grib_handle_delete(self._gh)
            if err:
                raise RuntimeError(grib_get_error_message(err))
        self.closed = True
    def rewind(self):
        """rewind iterator (same as seek(0))"""
        # before rewinding, move iterator to end of file
        # to make sure it is not left in a funky state
        # (such as in the middle of a multi-part message, issue 54)
        while 1:
            gh = grib_handle_new_from_file(NULL, self._fd, &err)
            err = grib_handle_delete(gh)
            if gh == NULL: break
        rewind(self._fd)
        self.messagenumber = 0
    def message(self, N):
        """
        message(N)
        
        retrieve N'th message in iterator.
        same as seek(N-1) followed by readline()."""
        if N < 1:
            raise IOError('grb message numbers start at 1')
        # if iterator positioned past message N, reposition at beginning.
        if self.messagenumber >= N:
            self.rewind()
        # move iterator forward to message N.
        self._advance(N-self.messagenumber)
        return _create_gribmessage(self._gh, self.messagenumber)
    def select(self, **kwargs):
        """
select(**kwargs)

return a list of L{gribmessage} instances from iterator filtered by kwargs.
If keyword is a container object, each grib message
in the iterator is searched for membership in the container.
If keyword is a callable (has a _call__ method), each grib
message in the iterator is tested using the callable (which should
return a boolean).
If keyword is not a container object or a callable, each 
grib message in the iterator is tested for equality.

Example usage:

>>> import pygrib
>>> grbs = pygrib.open('sampledata/gfs.grb')
>>> selected_grbs=grbs.select(shortName='gh',typeOfLevel='isobaricInhPa',level=10)
>>> for grb in selected_grbs: grb
26:Geopotential height:gpm (instant):regular_ll:isobaricInhPa:level 10 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
>>> # the __call__ method does the same thing
>>> selected_grbs=grbs(shortName='gh',typeOfLevel='isobaricInhPa',level=10)
>>> for grb in selected_grbs: grb
26:Geopotential height:gpm (instant):regular_ll:isobaricInhPa:level 10 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
>>> # to select multiple specific key values, use containers (e.g. sequences)
>>> selected_grbs=grbs(shortName=['u','v'],typeOfLevel='isobaricInhPa',level=[10,50])
>>> for grb in selected_grbs: grb
193:u-component of wind:m s**-1 (instant):regular_ll:isobaricInhPa:level 50 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
194:v-component of wind:m s**-1 (instant):regular_ll:isobaricInhPa:level 50 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
199:u-component of wind:m s**-1 (instant):regular_ll:isobaricInhPa:level 10 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
200:v-component of wind:m s**-1 (instant):regular_ll:isobaricInhPa:level 10 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
>>> # to select key values based on a conditional expression, use a function
>>> selected_grbs=grbs(shortName='gh',level=lambda l: l < 500 and l >= 300)
>>> for grb in selected_grbs: grb
14:Geopotential height:gpm (instant):regular_ll:isobaricInhPa:level 45000 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
15:Geopotential height:gpm (instant):regular_ll:isobaricInhPa:level 40000 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
16:Geopotential height:gpm (instant):regular_ll:isobaricInhPa:level 35000 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
17:Geopotential height:gpm (instant):regular_ll:isobaricInhPa:level 30000 Pa:fcst time 72 hrs:from 200412091200:lo res cntl fcst
"""
        msgnum = self.tell()
        self.rewind() # always search from beginning
        grbs = [grb for grb in self if _find(grb, **kwargs)]
        self.seek(msgnum) # leave iterator in original position.
        return grbs
    def _advance(self,nmsgs,return_msgs=False):
        """advance iterator n messages from current position.
        if return_msgs==True, grib message instances are returned
        in a list"""
        cdef int err
        if nmsgs < 0: 
            raise ValueError('nmsgs must be >= 0 in _advance')
        if return_msgs: grbs=[]
        for n in range(self.messagenumber,self.messagenumber+nmsgs):
            err = grib_handle_delete(self._gh)
            if err:
                raise RuntimeError(grib_get_error_message(err))
            self._gh = grib_handle_new_from_file(NULL, self._fd, &err)
            if err:
                raise RuntimeError(grib_get_error_message(err))
            if self._gh == NULL:
                raise IOError('not that many messages in file')
            self.messagenumber = self.messagenumber + 1
            if return_msgs: grbs.append(_create_gribmessage(self._gh, self.messagenumber))
        if return_msgs: return grbs

# keep track of python gribmessage attributes so they can be
# distinguished from grib keys.
_private_atts =\
['_gh','fcstimeunits','expand_reduced','projparams','messagenumber','_all_keys','_ro_keys']

def julian_to_datetime(object jd):
    """
    julian_to_datetime(julday)
    
    convert Julian day number to python datetime instance.

    Used to create validDate and analDate attributes from
    julianDay and forecastTime keys."""
    cdef double julday
    cdef long year, month, day, hour, minute, second
    cdef int err
    julday = jd
    err = grib_julian_to_datetime(julday, &year, &month, &day, &hour, &minute, &second)
    if err:
        raise RuntimeError(grib_get_error_message(err))
    return datetime(year, month, day, hour, minute, second)

def datetime_to_julian(object d):
    """
    datetime_to_julian(date)
    
    convert python datetime instance to Julian day number."""
    cdef double julday
    cdef int err
    cdef long year, month, day, hour, minute, second
    year = d.year; month = d.month; day = d.day; hour = d.hour
    minute = d.minute; second = d.second
    err = grib_datetime_to_julian(year,month,day,hour,minute,second,&julday)
    if err:
        raise RuntimeError(grib_get_error_message(err))
    return julday

cdef _create_gribmessage(grib_handle *gh, object messagenumber):
    """factory function for creating gribmessage instances"""
    cdef gribmessage grb  = gribmessage.__new__(gribmessage)
    grb.messagenumber = messagenumber
    grb.expand_reduced = True
    grb._gh = grib_handle_clone(gh)
    grb._all_keys = grb.keys()
    grb._ro_keys  = grb._read_only_keys()
    grb._set_projparams() # set projection parameter dict.
    return setdates(grb)

def fromstring(gribstring):
    """
    fromstring(string)

    Create a gribmessage instance from a python bytes object
    representing a binary grib message (the reverse of L{gribmessage.tostring}).
    """
    cdef char* gribstr
    cdef grib_handle * gh
    cdef gribmessage grb
    gribstr = gribstring
    gh = grib_handle_new_from_message_copy(NULL, <void *>gribstr, len(gribstring))
    grb  = gribmessage.__new__(gribmessage)
    grb.messagenumber = 1
    grb.expand_reduced = True
    grb._gh = gh
    grb._all_keys = grb.keys()
    grb._ro_keys  = grb._read_only_keys()
    grb._set_projparams() # set projection parameter dict.
    return setdates(grb)

def setdates(gribmessage grb):
    """
    setdates(grb)
    
    set fcstimeunits, analDate and validDate attributes using
    julianDay, forecastTime and indicatorOfUnitOfTimeRange.
    Called automatically when gribmessage instance created,
    but can be called manually to update keys if one of 
    them is modified after instance creation.
    """
    grb.fcstimeunits = ""
    if grb.has_key('indicatorOfUnitOfTimeRange') and\
       grb.indicatorOfUnitOfTimeRange in _ftimedict:
        grb.fcstimeunits = _ftimedict[grb.indicatorOfUnitOfTimeRange]
    if grb.has_key('forecastTime'):
        if grb.has_key('stepRange'):
            # this is a hack to work around grib_api bug
            # sometimes stepUnits and indicatorOfUnitOfTimeRange 
            # are inconsistent.
            grb.stepUnits = grb.indicatorOfUnitOfTimeRange
            ftime = grb['stepRange'] # computed key, uses stepUnits
            # if it's a range, use the end of the range to define validDate
            try: 
                ftime = float(ftime.split('-')[1])
            except:
                ftime = grb.forecastTime
        else:
            ftime = grb.forecastTime
    else:
        ftime = 0
    if grb.has_key('julianDay'):
        grb.analDate =\
        julian_to_datetime(grb.julianDay)
        if grb.fcstimeunits == 'hrs':
            grb.validDate =\
            julian_to_datetime(grb.julianDay+ftime/24.)
        elif grb.fcstimeunits == 'mins':
            grb.validDate =\
            julian_to_datetime(grb.julianDay+ftime/1440.)
        elif grb.fcstimeunits == 'days':
            grb.validDate =\
            julian_to_datetime(grb.julianDay+ftime)
        elif grb.fcstimeunits == 'secs':
            grb.validDate =\
            julian_to_datetime(grb.julianDay+ftime/86400.)
        elif grb.fcstimeunits == '3 hr periods':
            grb.validDate =\
            julian_to_datetime(grb.julianDay+ftime/8.)
        elif grb.fcstimeunits == '6 hr periods':
            grb.validDate =\
            julian_to_datetime(grb.julianDay+ftime/4.)
        elif grb.fcstimeunits == '12 hr periods':
            grb.validDate =\
            julian_to_datetime(grb.julianDay+ftime/2.)
    return grb

def reload(gribmessage grb):
    """
    reload(grb)

    Recreate gribmessage object, updating all the keys to be consistent
    with each other.  For example, if the forecastTime key is changed,
    recreating the gribmessage object with this function will cause
    the analDate and verifDate keys to be updated accordingly.

    Equivalent to fromstring(grb.tostring())"""
    return fromstring(grb.tostring())

cdef class gribmessage(object):
    """
    Grib message object.

    Each grib message has attributes corresponding to grib message
    keys for 
    U{GRIB1 <http://www.ecmwf.int/publications/manuals/d/gribapi/fm92/grib1>}
    and
    U{GRIB2 <http://www.ecmwf.int/publications/manuals/d/gribapi/fm92/grib2/>}.
    Parameter names are
    are given by the C{name}, C{shortName} and C{paramID}
    U{keys <http://www.ecmwf.int/publications/manuals/d/gribapi/param/>}.
    pygrib also defines some special attributes which are defined below
    under the heading B{Instance Variables}.

    @ivar messagenumber: The grib message number in the file.

    @ivar projparams: A dictionary containing proj4 key/value pairs describing 
    the grid.  Set to C{None} for unsupported grid types.

    @ivar expand_reduced:  If True (default), reduced lat/lon and gaussian grids
    will be expanded to regular grids when data is accessed via "values" key. If
    False, data is kept on unstructured reduced grid, and is returned in a 1-d
    array.

    @ivar fcstimeunits:  A string representing the forecast time units
    (an empty string if not defined).

    @ivar analDate:  A python datetime instance describing the analysis date
    and time for the forecast. Only set if forecastTime and julianDay keys
    exist.

    @ivar validDate:  A python datetime instance describing the valid date
    and time for the forecast. Only set if forecastTime and julianDay keys
    exist, and fcstimeunits is defined. If forecast time
    is a range, then C{validDate} corresponds to the end of the range."""
    cdef grib_handle *_gh
    cdef public messagenumber, projparams, validDate, analDate,\
    expand_reduced, _ro_keys, _all_keys, fcstimeunits
    def __init__(self):
        # calling "__new__()" will not call "__init__()" !
        raise TypeError("This class cannot be instantiated from Python")
    def __dealloc__(self):
        # finalization (inverse of __cinit__): needed to allow garbage collector to free memory.
        cdef int err
        err = grib_handle_delete(self._gh)
    def __getattr__(self, item):
        # allow gribmessage keys to accessed like attributes.
        # this is tried after looking for item in self.__dict__.keys().
        try:
            return self.__getitem__(item)
        except KeyError:
            raise AttributeError(item)
    def __setattr__(self, name, value):
        # allow gribmessage keys to be set like attributes.
        if name not in _private_atts:
            # these are grib message keys
            self[name] = value
        else:
            # these are python attributes.
            self.__dict__[name]=value
    def __repr__(self):
        """prints a short inventory of the grib message"""
        inventory = []
        if self.valid_key('name'):
            if self['name'] != 'unknown':
                inventory.append(repr(self.messagenumber)+':'+self['name'])
            elif self.valid_key('parameterName'):
                inventory.append(repr(self.messagenumber)+':'+self['parameterName'])
        if self.valid_key('units'):
            if self['units'] != 'unknown':
                inventory.append(':'+self['units'])
            elif self.valid_key('parameterUnits'):
                inventory.append(':'+self['parameterUnits'])
        if self.valid_key('stepType'):
            inventory.append(' ('+self['stepType']+')')
        if self.valid_key('typeOfGrid') or self.valid_key('gridType'):
            if self.valid_key('typeOfGrid'):
               inventory.append(':'+self['typeOfGrid'])
            else:
               inventory.append(':'+self['gridType'])
        if self.valid_key('typeOfLevel'):
            inventory.append(':'+self['typeOfLevel'])
        if self.valid_key('topLevel') and self.valid_key('bottomLevel'):
            toplev = None; botlev = None
            levunits = 'unknown'
            if self.valid_key('unitsOfFirstFixedSurface'):
                levunits = self['unitsOfFirstFixedSurface']
            if self.valid_key('typeOfFirstFixedSurface') and self['typeOfFirstFixedSurface'] != 255:
                toplev = self['topLevel']
                if self.valid_key('scaledValueOfFirstFixedSurface') and\
                   self.valid_key('scaleFactorOfFirstFixedSurface'):
                   if self['scaleFactorOfFirstFixedSurface']:
                       toplev = self['scaledValueOfFirstFixedSurface']/\
                                np.power(10.0,self['scaleFactorOfFirstFixedSurface'])
                   else:
                       toplev = self['scaledValueOfFirstFixedSurface']
            if self.valid_key('typeOfSecondFixedSurface') and self['typeOfSecondFixedSurface'] != 255:
                botlev = self['bottomLevel']
                if self.valid_key('scaledValueOfSecondFixedSurface') and\
                   self.valid_key('scaleFactorOfSecondFixedSurface'):
                   if self['scaleFactorOfSecondFixedSurface']:
                       botlev = self['scaledValueOfSecondFixedSurface']/\
                                np.power(10.0,self['scaleFactorOfSecondFixedSurface'])
                   else:
                       botlev = self['scaledValueOfSecondFixedSurface']
            levstring = None
            if botlev is None or toplev == botlev:
                levstring = ':level %s' % toplev
            else:
                levstring = ':levels %s-%s' % (toplev,botlev)
            if levunits != 'unknown':
                levstring = levstring+' %s' % levunits
            if levstring is not None:
                inventory.append(levstring)
        elif self.valid_key('level'):
            inventory.append(':level %s' % self['level'])
        if self.has_key('stepRange'):
            ftime = self['stepRange'] # computed key, uses stepUnits
            if self.valid_key('stepType') and self['stepType'] != 'instant':
                inventory.append(':fcst time %s %s (%s)'%\
                    (ftime,self.fcstimeunits,self.stepType))
            else:
                inventory.append(':fcst time %s %s'% (ftime,self.fcstimeunits))
        elif self.valid_key('forecastTime'):
            ftime = repr(self['forecastTime'])
            inventory.append(':fcst time %s %s'% (ftime,self.fcstimeunits))
        if self.valid_key('dataDate') and self.valid_key('dataTime'):
            inventory.append(
            ':from '+repr(self['dataDate'])+'%04i' % self['dataTime'])
        #if self.valid_key('validityDate') and self.valid_key('validityTime'):
        #    inventory.append(
        #    ':valid '+repr(self['validityDate'])+repr(self['validityTime']))
        if self.valid_key('perturbationNumber') and\
           self.valid_key('typeOfEnsembleForecast'):
            ens_type = self['typeOfEnsembleForecast']
            pert_num = self['perturbationNumber']
            if ens_type == 0:
               inventory.append(":lo res cntl fcst")
            elif ens_type == 1:
               inventory.append(":hi res cntl fcst")
            elif ens_type == 2:
               inventory.append(":neg ens pert %d" % pert_num)
            elif ens_type == 3:
               inventory.append(":pos ens pert %d" % pert_num)
        if self.valid_key('derivedForecast'):
            if self['derivedForecast'] == 0:
                inventory.append(":ens mean")
            elif self['derivedForecast'] == 1:
                inventory.append(":weighted ens mean")
            elif self['derivedForecast'] == 2:
                inventory.append(":ens std dev")
            elif self['derivedForecast'] == 3:
                inventory.append(":normalized ens std dev")
            elif self['derivedForecast'] == 4:
                inventory.append(":ens spread")
            elif self['derivedForecast'] == 5:
                inventory.append(":ens large anomaly index")
            elif self['derivedForecast'] == 6:
                inventory.append(":ens mean of cluster")
        if self.valid_key('probabilityTypeName'):
            inventory.append(":"+self['probabilityTypeName'])
            lowerlim = None
            if self.valid_key('scaledValueOfLowerLimit') and\
               self.valid_key('scaleFactorOfLowerLimit'):
               if self['scaledValueOfLowerLimit'] and\
                  self['scaleFactorOfLowerLimit']: 
                   lowerlim = self['scaledValueOfLowerLimit']/\
                              np.power(10.0,self['scaleFactorOfLowerLimit'])
            upperlim = None
            if self.valid_key('scaledValueOfUpperLimit') and\
               self.valid_key('scaleFactorOfUpperLimit'):
               if self['scaledValueOfUpperLimit'] and\
                  self['scaleFactorOfUpperLimit']: 
                   upperlim = self['scaledValueOfUpperLimit']/\
                              np.power(10.0,self['scaleFactorOfUpperLimit'])
            if upperlim is not None and lowerlim is not None:
                inventory.append(" (%s-%s)" % (upperlim,lowerlim))
            elif upperlim is not None:
                inventory.append(" (> %s)" % upperlim)
            elif lowerlim is not None:
                inventory.append(" (< %s)" % lowerlim)
        return ''.join(inventory)
    def is_missing(self,key):
        """
        is_missing(key)

        returns True if value associated with key is equal
        to grib missing value flag (False otherwise)"""
        cdef int err,miss
        cdef char *name
        bytestr = _strencode(key)
        name = bytestr
        miss = grib_is_missing(self._gh, name, &err)
        if err:
            raise RuntimeError(grib_get_error_message(err))
        if miss:
            return True
        else:
            return False
    def keys(self):
        """
        keys()

        return keys associated with a grib message (a dictionary-like object)
        """
        cdef grib_keys_iterator* gi
        cdef int err, typ
        cdef char *name
        # use cached keys if they exist.
        if self._all_keys is not None: return self._all_keys
        # if not, get keys from grib file.
        gi = grib_keys_iterator_new(self._gh,\
                GRIB_KEYS_ITERATOR_ALL_KEYS, NULL)
        keys = []
        while grib_keys_iterator_next(gi):
            name = grib_keys_iterator_get_name(gi)
            key = name.decode('ascii')
            # ignore these keys.
            if key in ["zero","one","eight","eleven","false","thousand","file",
                       "localDir","7777","oneThousand"]:
                continue
            err = grib_get_native_type(self._gh, name, &typ)
            if err: # skip unreadable keys
                continue
            # keys with these types are ignored.
            if typ not in\
            [GRIB_TYPE_UNDEFINED,GRIB_TYPE_SECTION,GRIB_TYPE_BYTES,GRIB_TYPE_LABEL,GRIB_TYPE_MISSING]:
                keys.append(key)
        err = grib_keys_iterator_delete(gi)
        if err:
            raise RuntimeError(grib_get_error_message(err))
        return keys
    def _read_only_keys(self):
        """
        _read_only_keys()

        return read-only keys associated with a grib message (a dictionary-like object)
        """
        cdef grib_keys_iterator* gi
        cdef int err, typ
        cdef char *name
        if self._all_keys is None:
            self._all_keys = self.keys()
        gi = grib_keys_iterator_new(self._gh,\
                GRIB_KEYS_ITERATOR_SKIP_READ_ONLY, NULL)
        keys_noro = []
        while grib_keys_iterator_next(gi):
            name = grib_keys_iterator_get_name(gi)
            key = name.decode('ascii')
            keys_noro.append(key)
        err = grib_keys_iterator_delete(gi)
        if err:
            raise RuntimeError(grib_get_error_message(err))
        keys_ro = []
        for key in self._all_keys:
            if key not in keys_noro:
                keys_ro.append(key)
        return keys_ro
    def __setitem__(self, key, value):
        """
        change values associated with existing grib keys.
        """
        cdef int err, typ
        cdef size_t size
        cdef char *name
        cdef long longval
        cdef double doubleval
        cdef ndarray datarr
        cdef char *strdata
        if key in self._ro_keys:
            raise KeyError('key "%s" is read only' % key)
        if key not in self._all_keys:
            raise KeyError('can only modify existing grib keys (key "%s" not found)'
                    % key )
        bytestr = _strencode(key)
        name = bytestr
        err = grib_get_native_type(self._gh, name, &typ)
        if err:
            raise RuntimeError(grib_get_error_message(err))
        elif typ == GRIB_TYPE_LONG:
            # is value an array or a scalar?
            datarr = np.asarray(value, np.int)
            is_array = False
            if datarr.shape:
                is_array = True
            if not is_array: # scalar
                longval = value
                err = grib_set_long(self._gh, name, longval)
                if err:
                    raise RuntimeError(grib_get_error_message(err))
            else:
                if key == 'values':
                    datarr = self._unshape_mask(datarr)
                if not PyArray_ISCONTIGUOUS(datarr):
                    datarr = datarr.copy()
                size = datarr.size
                err = grib_set_long_array(self._gh, name, <long *>datarr.data, size)
                if err:
                    raise RuntimeError(grib_get_error_message(err))
        elif typ == GRIB_TYPE_DOUBLE:
            # is value an array or a scalar?
            datarr = np.asarray(value, np.float)
            is_array = False
            if datarr.shape:
                is_array = True
            if not is_array: # scalar
                doubleval = value
                err = grib_set_double(self._gh, name, doubleval)
                if err:
                    raise RuntimeError(grib_get_error_message(err))
            else:
                if key == 'values':
                    datarr = self._unshape_mask(datarr)
                if not PyArray_ISCONTIGUOUS(datarr):
                    datarr = datarr.copy()
                size = datarr.size
                err = grib_set_double_array(self._gh, name, <double *>datarr.data, size)
                if err:
                    raise RuntimeError(grib_get_error_message(err))
        elif typ == GRIB_TYPE_STRING:
            size=len(value)
            bytestr = _strencode(value)
            strdata = bytestr
            err = grib_set_string(self._gh, name, strdata, &size)
            if err:
                raise RuntimeError(grib_get_error_message(err))
        else:
            raise ValueError("unrecognized grib type % d" % typ)
    def __getitem__(self, key):
        """
        access values associated with grib keys.
        
        The key "values" will return the data associated with the grib message.
        The data is returned as a numpy array, or if missing values or a bitmap
        are present, a numpy masked array.  Reduced lat/lon or gaussian grid
        data is automatically expanded to a regular grid using linear
        interpolation (nearest neighbor if an adjacent grid point is a missing
        value)."""
        cdef int err, typ
        cdef size_t size
        cdef char *name
        cdef long longval
        cdef double doubleval
        cdef ndarray datarr
        cdef char strdata[1024]
        bytestr = _strencode(key)
        name = bytestr
        usenceplib = key == 'values' and self.packingType.startswith('grid_complex')
        # this workaround only needed for grib_api < 1.9.16.
        if usenceplib:
            size = self.numberOfValues 
        else:
            err = grib_get_size(self._gh, name, &size)
            if err:
                raise RuntimeError(grib_get_error_message(err))
        # this workaround only needed for grib_api < 1.9.16.
        if usenceplib:
            typ = 2
            err = 0
        else:
            err = grib_get_native_type(self._gh, name, &typ)
        # force 'paramId' to be size 1 (it returns a size of 7,
        # which is a relic from earlier versions of grib_api in which
        # paramId was a string and not an integer)
        if key=='paramId' and typ == GRIB_TYPE_LONG: size=1
        if err:
            raise RuntimeError(grib_get_error_message(err))
        elif typ == GRIB_TYPE_LONG:
            if size == 1: # scalar
                err = grib_get_long(self._gh, name, &longval)
                if err:
                    raise RuntimeError(grib_get_error_message(err))
                return longval
            else: # array
                if self.has_key('jPointsAreConsecutive') and\
                   self['jPointsAreConsecutive']:
                    storageorder='F'
                else:
                    storageorder='C'
                datarr = np.zeros(size, np.int, order=storageorder)
                err = grib_get_long_array(self._gh, name, <long *>datarr.data, &size)
                if err:
                    raise RuntimeError(grib_get_error_message(err))
                if key == 'values':
                    return self._reshape_mask(datarr)
                else:
                    return datarr
        elif typ == GRIB_TYPE_DOUBLE:
            if size == 1: # scalar
                err = grib_get_double(self._gh, name, &doubleval)
                if err:
                    raise RuntimeError(grib_get_error_message(err))
                return doubleval
            else: # array
                if self.has_key('jPointsAreConsecutive') and\
                   self['jPointsAreConsecutive']:
                    storageorder='F'
                else:
                    storageorder='C'
                if usenceplib: 
                    # use ncep lib to decode data (workaround for grib_api
                    # bug with second-order complex packing).
                    grb = Grib2Decode(self.tostring(), gribmsg=True)
                    return grb.data()
                else:
                    datarr = np.zeros(size, np.double, order=storageorder)
                    err = grib_get_double_array(self._gh, name, <double *>datarr.data, &size)
                    if err:
                        raise RuntimeError(grib_get_error_message(err))
                    if key == 'values':
                        return self._reshape_mask(datarr)
                    else:
                        return datarr
        elif typ == GRIB_TYPE_STRING:
            size=1024 # grib_get_size returns 1 ?
            err = grib_get_string(self._gh, name, strdata, &size)
            if err:
                raise RuntimeError(grib_get_error_message(err))
            msg = strdata.decode(default_encoding)
            return msg.rstrip()
        else:
            raise ValueError("unrecognized grib type % d" % typ)
    def has_key(self,key):
        """
        has_key(key)

        tests whether a grib message object has a specified key.
        """
        return key in self._all_keys
    def valid_key(self,key):
        """
        valid_key(key)

        tests whether a grib…