[6] | 1 | module m_converters |
---|
| 2 | |
---|
| 3 | use m_debug |
---|
| 4 | |
---|
| 5 | private |
---|
| 6 | ! |
---|
| 7 | ! Takes a string and turns it into useful data structures, |
---|
| 8 | ! such as numerical arrays. |
---|
| 9 | ! |
---|
| 10 | ! NOTE: The string must contain *homogeneous* data, i.e.: all real numbers, |
---|
| 11 | ! all integers, etc. |
---|
| 12 | ! |
---|
| 13 | public :: build_data_array |
---|
| 14 | |
---|
| 15 | interface build_data_array |
---|
| 16 | module procedure build_data_array_real_sp, & |
---|
| 17 | build_data_array_real_dp, & |
---|
| 18 | build_data_array_integer |
---|
| 19 | end interface |
---|
| 20 | private :: build_data_array_real_sp |
---|
| 21 | private :: build_data_array_real_dp |
---|
| 22 | private :: build_data_array_integer |
---|
| 23 | |
---|
| 24 | private :: token_analysis, is_separator, is_CR_or_LF |
---|
| 25 | |
---|
| 26 | CONTAINS |
---|
| 27 | |
---|
| 28 | !--------------------------------------------------------------- |
---|
| 29 | subroutine build_data_array_real_dp(str,x,n) |
---|
| 30 | integer, parameter :: dp = selected_real_kind(14) |
---|
| 31 | ! |
---|
| 32 | character(len=*), intent(in) :: str |
---|
| 33 | real(kind=dp), dimension(:), intent(inout) :: x |
---|
| 34 | integer, intent(inout) :: n |
---|
| 35 | |
---|
| 36 | integer :: ntokens, status, last_pos |
---|
| 37 | character(len=len(str)) :: s |
---|
| 38 | |
---|
| 39 | s = str |
---|
| 40 | call token_analysis(s,ntokens,last_pos) |
---|
| 41 | if (debug) print *, "ntokens, last_pos ", ntokens, last_pos |
---|
| 42 | if (debug) print *, s |
---|
| 43 | if ((n + ntokens) > size(x)) STOP "data array full" |
---|
| 44 | read(unit=s(1:last_pos),fmt=*,iostat=status) x(n+1:n+ntokens) |
---|
| 45 | if (status /= 0) STOP "real conversion error" |
---|
| 46 | n = n + ntokens |
---|
| 47 | |
---|
| 48 | end subroutine build_data_array_real_dp |
---|
| 49 | !--------------------------------------------------------------- |
---|
| 50 | |
---|
| 51 | subroutine build_data_array_real_sp(str,x,n) |
---|
| 52 | integer, parameter :: sp = selected_real_kind(6) |
---|
| 53 | ! |
---|
| 54 | character(len=*), intent(in) :: str |
---|
| 55 | real(kind=sp), dimension(:), intent(inout) :: x |
---|
| 56 | integer, intent(inout) :: n |
---|
| 57 | |
---|
| 58 | integer :: ntokens, status, last_pos |
---|
| 59 | character(len=len(str)) :: s |
---|
| 60 | |
---|
| 61 | s = str |
---|
| 62 | call token_analysis(s,ntokens,last_pos) |
---|
| 63 | if (debug) print *, "ntokens, last_pos ", ntokens, last_pos |
---|
| 64 | if (debug) print *, s |
---|
| 65 | if ((n + ntokens) > size(x)) STOP "data array full" |
---|
| 66 | read(unit=s(1:last_pos),fmt=*,iostat=status) x(n+1:n+ntokens) |
---|
| 67 | if (status /= 0) STOP "real conversion error" |
---|
| 68 | n = n + ntokens |
---|
| 69 | |
---|
| 70 | end subroutine build_data_array_real_sp |
---|
| 71 | |
---|
| 72 | !--------------------------------------------------------------- |
---|
| 73 | subroutine build_data_array_integer(str,x,n) |
---|
| 74 | integer, parameter :: sp = selected_real_kind(14) |
---|
| 75 | ! |
---|
| 76 | character(len=*), intent(in) :: str |
---|
| 77 | integer, dimension(:), intent(inout) :: x |
---|
| 78 | integer, intent(inout) :: n |
---|
| 79 | |
---|
| 80 | integer :: ntokens, status, last_pos |
---|
| 81 | character(len=len(str)) :: s |
---|
| 82 | |
---|
| 83 | s = str |
---|
| 84 | call token_analysis(s,ntokens,last_pos) |
---|
| 85 | if (debug) print *, "ntokens, last_pos ", ntokens, last_pos |
---|
| 86 | if (debug) print *, s |
---|
| 87 | if ((n + ntokens) > size(x)) STOP "data array full" |
---|
| 88 | read(unit=s(1:last_pos),fmt=*,iostat=status) x(n+1:n+ntokens) |
---|
| 89 | if (status /= 0) STOP "integer conversion error" |
---|
| 90 | n = n + ntokens |
---|
| 91 | |
---|
| 92 | end subroutine build_data_array_integer |
---|
| 93 | |
---|
| 94 | |
---|
| 95 | !================================================================== |
---|
| 96 | |
---|
| 97 | function is_separator(c) result(sep) |
---|
| 98 | character(len=1), intent(in) :: c |
---|
| 99 | logical :: sep |
---|
| 100 | |
---|
| 101 | sep = ((c == char(32)) .or. (c == char(10)) & |
---|
| 102 | .or. (c == char(9)) .or. (c == char(13))) |
---|
| 103 | |
---|
| 104 | end function is_separator |
---|
| 105 | !---------------------------------------------------------------- |
---|
| 106 | function is_CR_or_LF(c) result(res) |
---|
| 107 | character(len=1), intent(in) :: c |
---|
| 108 | logical :: res |
---|
| 109 | |
---|
| 110 | res = ((c == char(10)) .or. (c == char(13))) |
---|
| 111 | |
---|
| 112 | end function is_CR_or_LF |
---|
| 113 | |
---|
| 114 | !================================================================== |
---|
| 115 | |
---|
| 116 | subroutine token_analysis(str,ntokens,last_pos) |
---|
| 117 | ! |
---|
| 118 | character(len=*), intent(inout) :: str |
---|
| 119 | integer, intent(out) :: ntokens, last_pos |
---|
| 120 | ! |
---|
| 121 | ! |
---|
| 122 | ! Checks the contents of a string and finds the number of tokens it contains |
---|
| 123 | ! The standard separator is generalized whitespace (space, tab, CR, or LF) |
---|
| 124 | ! It also returns the last useful position in the string (excluding |
---|
| 125 | ! separator characters which are not blanks, and thus not caught by the |
---|
| 126 | ! (len_)trim fortran intrinsic). This is necessary to perform list-directed |
---|
| 127 | ! I/O in the string as an internal file. |
---|
| 128 | ! |
---|
| 129 | ! Also, replace on the fly CR and LF by blanks. This is necessary if |
---|
| 130 | ! str spans more than one record. In that case, internal reads only |
---|
| 131 | ! look at the first record. |
---|
| 132 | ! -- ** Compiler limits on size of internal record?? |
---|
| 133 | ! |
---|
| 134 | integer :: i, str_length |
---|
| 135 | logical :: in_token |
---|
| 136 | character(len=1) :: c |
---|
| 137 | |
---|
| 138 | in_token = .false. |
---|
| 139 | ntokens = 0 |
---|
| 140 | last_pos = 0 |
---|
| 141 | |
---|
| 142 | str_length = len_trim(str) |
---|
| 143 | !print *, "string length: ", str_length |
---|
| 144 | |
---|
| 145 | do i = 1, str_length |
---|
| 146 | c = str(i:i) |
---|
| 147 | |
---|
| 148 | if (in_token) then |
---|
| 149 | if (is_separator(c)) then |
---|
| 150 | in_token = .false. |
---|
| 151 | if (is_CR_or_LF(c)) str(i:i) = " " |
---|
| 152 | else |
---|
| 153 | last_pos = i |
---|
| 154 | endif |
---|
| 155 | |
---|
| 156 | else ! not in token |
---|
| 157 | |
---|
| 158 | if (is_separator(c)) then |
---|
| 159 | if (is_CR_or_LF(c)) str(i:i) = " " |
---|
| 160 | ! do nothing |
---|
| 161 | else |
---|
| 162 | in_token = .true. |
---|
| 163 | last_pos = i |
---|
| 164 | ntokens = ntokens + 1 |
---|
| 165 | endif |
---|
| 166 | endif |
---|
| 167 | enddo |
---|
| 168 | !print *, "ntokens, last_pos: ", ntokens, last_pos |
---|
| 169 | |
---|
| 170 | end subroutine token_analysis |
---|
| 171 | |
---|
| 172 | |
---|
| 173 | end module m_converters |
---|
| 174 | |
---|
| 175 | |
---|
| 176 | |
---|
| 177 | |
---|
| 178 | |
---|
| 179 | |
---|
| 180 | |
---|