import numpy as np
from astropy.io import fits
import matplotlib.pyplot as plt
import pickle
from astropy.wcs import WCS
import tqdm
from multiprocessing.pool import ThreadPool
from chan_psf import solve_for_locations, solve_for_locations_eintp, solve_for_rates

psfe = np.array([1.8, 1.9, 3.0, 4.0, 6.0, 7.0, 8.0, 9.0])

def prepare_psf(evt):
    """
    find all unique psf for observation and load in single 3d data cuve
    return data cube with events slices indexes
    """
    u, ui = np.unique(evt["psf_cube"], return_inverse=True)
    data = np.array([np.load(p[3:])[:, ::-1,::-1].copy() for p in u])
    return data, ui

def select_xychunksize(wcs, halfpsfsize=36./3600.):
    """
    get wcs and find wcs pixel size of psf reach
    """
    sizex = int(abs(halfpsfsize/wcs.wcs.cdelt[1])) + 2
    sizey = int(abs(halfpsfsize/wcs.wcs.cdelt[0])) + 2
    return sizex, sizey

def read_wcs(h):
    """
    read events wcs header
    """
    w = WCS(naxis=2)
    w.wcs.ctype = [h["TCTYP11"], h["TCTYP12"]]
    w.wcs.crval = [h["TCRVL11"], h["TCRVL12"]]
    w.wcs.cdelt = [h["TCDLT11"], h["TCDLT12"]]
    w.wcs.crpix = [h["TCRPX11"], h["TCRPX12"]]
    w = WCS(w.to_header())
    return w


def create_neighboring_blocks(x, y, sizex, sizey):
    """
    schematically all sky is splitted on squares, which are approximatelly ~ 10 times greater then the psf
    events for corresponding square are joined :: squer + diluttaion of psf reach

    coordinate system with events and all required coefficiets are fed to psf solver
    current psf size is 25*0.5 arcsec (with up to \sqrt(2) factor in case of worst rolls -> 36''

    """
    """
    event list already contains x and y for each event
    """
    iix = (x//sizex + 0.5).astype(int)
    iiy = (y//sizey + 0.5).astype(int)
    isx, isy = np.mgrid[-1:2:1, -1:2:1]
    oidx = np.repeat(np.arange(x.size), 9)
    xyu, iixy, xyc = np.unique(np.array([np.repeat(iix, 9) + np.tile(isx.ravel(), x.size), 
                                         np.repeat(iiy, 9)+  np.tile(isy.ravel(), x.size)]), axis=1, return_counts=True, return_inverse=True)

    sord = np.argsort(iixy)
    return oidx[sord], xyu, xyc


def lkls_for_rates(evt, expv, wcs, srcx, srcy, rates):
    sizex, sizey = select_xychunksize(wcs)
    x, y = evt["x"].astype(float), evt["y"].astype(float)
    mask = np.logical_and.reduce([x > srcx - sizex//2, y > srcy - sizey//2, x < srcx + sizex//2, y < srcy + sizey//2], axis=0)
    print("mask sum", srcx, srcy, mask.sum())
    eloc = evt[mask]
    pickle.dump(eloc, open("eloc.pkl", "wb"))
    psfdata, ui = prepare_psf(eloc)
    xe, ye = np.copy(x[mask]), np.copy(y[mask])
    eidx = np.maximum(np.searchsorted(psfe*1e3, eloc["ENERGY"]) - 1, 0)
    ee = np.maximum((eloc["ENERGY"]/1000. - psfe[eidx])/(psfe[eidx + 1] - psfe[eidx]), 0.).astype(float) + eidx
    pk = np.copy(eloc["src_spec"]/eloc["bkg_spec"]).astype(float)
    roll = np.copy(np.deg2rad(eloc["roll_pnt"])).astype(float)
    #"OOOOOOOdddO", &psfi, &eidx, &x, &y, &roll, &pk, &rates, &xc, &yc, &eloc, &smat
    #                       O   O   O   O   O    O    O      d     d     d      O"
    print(ui, ee, xe, ye, roll, pk)
    lkls = solve_for_rates(ui, ee, xe, ye, roll, pk, rates, srcx, srcy, expv, psfdata)
    return lkls


if __name__ == "__main__":
    p1 = fits.open("test.fits")
    ewcs = read_wcs(p1[1].header)
    wcs = WCS(fits.getheader("eR_spec_asp_0.fits.gz", 0))
    xc, yc = 4290, 4147
    xc, yc = 4643, 4223.7
    #xc, yc = 4147,4290 
    xc, yc = ewcs.all_world2pix(wcs.all_pix2world([[xc, yc],], 0), 0).T
    print(xc, yc)
    eloc = 0.025 #0.0283
    #rates = np.array([4.2/eloc,]) #np.logspace(-0.5, 0.5, 129)*4.2/eloc
    rates = np.logspace(-0.5, 0.5, 129)*1352/eloc #*4.2/eloc
    lkls = lkls_for_rates(p1[1].data, eloc, ewcs, xc, yc, rates)
    plt.plot(rates, lkls)
    plt.axvline(rates[64])
    plt.show()