;+
;FUNCTION:   interp(y,x,u)
;PURPOSE:
;  Linearly Interpolates vectors with an irregular grid.
;  INTERP is functionally the same as INTERPOL, however it is typically
;  much faster for most applications.
;USAGE:
;  result = interp(y,x,u)
;INPUTS:
;       Y:      The input vector can be any type except string.
;
;       X:      The absicissae values for Y.  This vector must have same # of
;               elements as Y.  The values MUST be monotonically ascending
;               or descending.
;
;       U:      The absicissae values for the result.  The result will have
;               the same number of elements as U.  U does not need to be
;               monotonic.
;KEYWORDS:
;  NO_CHECK_MONOTONIC:   set this keyword to skip the check for monotonic data.
;  INDEX:  Set to named variable to return the index of the closest x less than u.
;      (same dimensons as u)
;  NO_EXTRAPOLATE:  Set this keyword to prevent extrapolation.
;  LAST_VALUE:  Set this keyword to return the last value of y array:  y[index]    (no interpolation performed)
;  INTERP_THRESHOLD:  Set to minimum allowed gap size.
;
;CREATED BY:	Davin Larson  4-30-96
;FILE:  interp.pro
; $LastChangedBy: dmitchell $
; $LastChangedDate: 2024-06-04 16:43:03 -0700 (Tue, 04 Jun 2024) $
; $LastChangedRevision: 32686 $
; $URL: svn+ssh://thmsvn@ambrosia.ssl.berkeley.edu/repos/spdsoft/tags/spedas_6_1/general/misc/interp.pro $
; $Id: interp.pro 32686 2024-06-04 23:43:03Z dmitchell $
;-
function interp,y,x,u,index=i,no_check_monotonic=ch_mon,no_extrapolate=no_extrap,interp_threshold=int_th, $
  ignore_nan=ignore_nan,last_value=last_value,missing = missing,xlog=xlog,ylog=ylog

  ;on_error,2


  if keyword_set(xlog) || keyword_set(ylog) then begin
    new_y = keyword_set(ylog) ? alog(y) : y
    new_x = keyword_set(xlog) ? alog(x) : x
    new_u = keyword_set(xlog) ? alog(u) : u
    rval = interp( new_y,  new_x, new_u,index=i,no_extrap=no_extrap,interp_thresh=int_th,ignore_nan=ignore_nan,last_value=last_value)
    if keyword_set(ylog) then rval =  exp(rval)
    return, rval
  endif

  if n_params() eq 2 then begin
    nx = n_elements(y)
    return,interp(y,findgen(nx),findgen(x)/(x-1)*(nx-1),index=i,no_extrap=no_extrap,interp_thresh=int_th,ignore_nan=ignore_nan,last_value=last_value,xlog=xlog,ylog=ylog)
  endif

  if array_equal(x,u) then begin   ; quick check, improves speed in many cases
    dprint,dlevel=3,'Same time steps, no need to interpolate'
    i = lindgen(n_elements(u))     ; need to define index in case interp_threshold is set (DLM, 2024-06-04)
    return,y
  endif
  ndimy = size(/n_dimension,y)
  ndimx = size(/n_dimension,x)
  if ndimy eq 2 then begin
    dimy= size(/dimension,y)
    dimv = dimy
    dimv[0]=n_elements(u)
    nv = make_array(dimv,type=size(/type,y))
    for j=0l,dimy[1]-1 do begin
      xx = (ndimx eq 2) ? x[*,j] : x
      nv[*,j] = interp(y[*,j],xx,u,no_extrapolate=no_extrap,interp_threshold=int_th,no_check_mono=ch_mon,index=i,ignore_nan=ignore_nan,last_value=last_value,xlog=xlog,ylog=ylog)
    endfor
    return,nv
  endif

  ;check for invalid x values:

  nx = n_elements(x)

  good = finite(x)
  if keyword_set(ignore_nan) then good = good and finite(y)
  good = where(good,c )
  if c lt 1 then begin
    ;   message,/info,'Not enough valid data points to interpolate.'
    return,replicate(!values.f_nan,n_elements(u))
  endif


  ; insure that all x points are valid
  if c ne nx then return, interp(y[good],x[good],u,index=i,no_extrap=no_extrap,interp_thresh=int_th,xlog=xlog,ylog=ylog)

  ; insure that x is monotonically increasing
  if x[0] gt x[nx-1] then return,interp(reverse(y),reverse(x),u,index=i,interp_thresh=int_th,no_extrap=no_extrap,last_value=last_value,xlog=xlog,ylog=ylog)


  if not keyword_set(ch_mon) then begin
    dx = x-shift(x,1)
    dx[0] = 0
    bad = where(dx lt 0,c)
    if c ne 0 then dprint,dlevel=3,'Warning: Data not monotonic!'
  endif

  if keyword_set(int_th) then begin
    w = where( finite(y) ,c )
    if c eq 0 then w=[0]
    nv = interp(y[w],x[w],u,index=i,no_extrap=no_extrap,xlog=xlog,ylog=ylog)
    dx = (x[w])[i+1] - (x[w])[i]
    w = where(dx gt int_th,c)
    if c ne 0 then nv[w]=!values.f_nan
    return, nv
  endif

  mn = long(u)  &  mn[*] = 0l
  mx = long(u)  &  mx[*] = nx-1

  repeat begin           ; This loop should execute approximately log2(nx) times
    i = (mx+mn)/2
    tst = x[i] lt u
    ntst = tst eq 0
    mn =  tst*i + ntst*mn
    mx = ntst*i +  tst*mx
  endrep  until max(mx-mn) le 1
  i = (mx+mn)/2
  if keyword_set(last_value) then begin
    nv = y[i]
  endif else begin
    if array_equal(u,x[i]) then begin   ;Smart to check if u eq x[i] here to avoid interpolating with nans
      dprint,dlevel=3,'No need to interpolate!'
      return,y[i]
    endif
    nv = y[i] + (u-x[i])*(y[i+1]-y[i])/(x[i+1]-x[i])
  endelse



  if keyword_set(no_extrap) then begin
    mxmx = minmax(x)
    w = where( (u lt mxmx[0]) or (u gt mxmx[1]) , nbad)
    if nbad gt 0 then nv[w] = !values.f_nan
  endif

  return,nv
end