| 1 | package calc |
| 2 | |
| 3 | // fn_lookup.go implements VLOOKUP, INDEX and MATCH. All three need a |
| 4 | // range argument's shape (its width and height as written), not the |
| 5 | // flattened, boundary-free []Value evalCallArgs hands an ordinary Func — |
| 6 | // see registry.go's nodeFunc doc comment for why. All three register as |
| 7 | // nodeFuncs and evaluate their own arguments directly against env, using |
| 8 | // the same position-as-id CellKey convention refKey uses elsewhere in |
| 9 | // this package (RowID = RowPos+1, ColID = ColPos+1; see eval.go). |
| 10 | |
| 11 | func init() { |
| 12 | registerNode("VLOOKUP", fnVLookup) |
| 13 | registerNode("INDEX", fnIndex) |
| 14 | registerNode("MATCH", fnMatch) |
| 15 | } |
| 16 | |
| 17 | // boundedRangeBounds is rangeBounds (eval.go) plus the same cap |
| 18 | // rangeCells, evalCallArg and evalAggArgs all enforce on a Range |
| 19 | // (maxRangeCells; see eval.go's rangeCellCount). VLOOKUP, INDEX and MATCH |
| 20 | // evaluate their range argument's shape directly against env rather than |
| 21 | // through rangeCells — they need the range's four bounds to compute an |
| 22 | // offset or walk cells directly, not a flattened []CellKey (see this |
| 23 | // file's package doc comment) — so without this check they would be the |
| 24 | // one path into a Range that never consults the cap at all: unbounded |
| 25 | // synchronous iteration over a pathological reference (decodeRef accepts |
| 26 | // any int row), and — because collectRefs also uses rangeCells and so |
| 27 | // never wires an over-cap range's cells as precedents — a stale-cache |
| 28 | // hole where these functions could read live cells the graph never |
| 29 | // tracked. ok is false when the range is over the cap; callers must |
| 30 | // resolve to #REF! rather than iterate, the same answer evalCallArg gives |
| 31 | // an ordinary function for the same over-cap range. |
| 32 | func boundedRangeBounds(rng *Range) (r0, r1, c0, c1 int, ok bool) { |
| 33 | r0, r1, c0, c1 = rangeBounds(rng) |
| 34 | _, ok = rangeCellCount(r0, r1, c0, c1) |
| 35 | return |
| 36 | } |
| 37 | |
| 38 | func cellValueAt(env Env, row, col int) Value { |
| 39 | return env.CellValue(CellKey{RowID: uint64(row + 1), ColID: uint64(col + 1)}) |
| 40 | } |
| 41 | |
| 42 | // fnVLookup implements VLOOKUP(lookup_value, table_array, col_index_num, |
| 43 | // [range_lookup]). range_lookup defaults to TRUE (approximate match, |
| 44 | // requiring the first column sorted ascending — Excel's own precondition; |
| 45 | // an unsorted first column gives an unspecified answer in Excel too); |
| 46 | // FALSE requires an exact match. table_array must be a literal range — a |
| 47 | // scalar or a call result has no shape to look up a column within. |
| 48 | func fnVLookup(argNodes []Node, env Env) Value { |
| 49 | if len(argNodes) < 3 || len(argNodes) > 4 { |
| 50 | return errValue(ErrValue) |
| 51 | } |
| 52 | lookup := Eval(argNodes[0], env) |
| 53 | if lookup.Kind == KindErr { |
| 54 | return lookup |
| 55 | } |
| 56 | rng, ok := argNodes[1].(*Range) |
| 57 | if !ok { |
| 58 | return errValue(ErrValue) |
| 59 | } |
| 60 | ciV := Eval(argNodes[2], env) |
| 61 | if ciV.Kind == KindErr { |
| 62 | return ciV |
| 63 | } |
| 64 | ciF, ek := toNumber(ciV) |
| 65 | if ek != ErrNone { |
| 66 | return errValue(ek) |
| 67 | } |
| 68 | ci := int(ciF) |
| 69 | |
| 70 | approx := true |
| 71 | if len(argNodes) == 4 { |
| 72 | rlV := Eval(argNodes[3], env) |
| 73 | if rlV.Kind == KindErr { |
| 74 | return rlV |
| 75 | } |
| 76 | b, ek := toBool(rlV) |
| 77 | if ek != ErrNone { |
| 78 | return errValue(ek) |
| 79 | } |
| 80 | approx = b |
| 81 | } |
| 82 | |
| 83 | r0, r1, c0, c1, withinCap := boundedRangeBounds(rng) |
| 84 | if !withinCap { |
| 85 | return errValue(ErrRef) |
| 86 | } |
| 87 | cols := c1 - c0 + 1 |
| 88 | if ci < 1 || ci > cols { |
| 89 | return errValue(ErrRef) |
| 90 | } |
| 91 | resultCol := c0 + ci - 1 |
| 92 | |
| 93 | if !approx { |
| 94 | for row := r0; row <= r1; row++ { |
| 95 | if compareValues(cellValueAt(env, row, c0), lookup) == 0 { |
| 96 | return cellValueAt(env, row, resultCol) |
| 97 | } |
| 98 | } |
| 99 | return errValue(ErrNA) |
| 100 | } |
| 101 | |
| 102 | found := false |
| 103 | bestRow := 0 |
| 104 | for row := r0; row <= r1; row++ { |
| 105 | if compareValues(cellValueAt(env, row, c0), lookup) <= 0 { |
| 106 | bestRow = row |
| 107 | found = true |
| 108 | continue |
| 109 | } |
| 110 | break |
| 111 | } |
| 112 | if !found { |
| 113 | return errValue(ErrNA) |
| 114 | } |
| 115 | return cellValueAt(env, bestRow, resultCol) |
| 116 | } |
| 117 | |
| 118 | // fnIndex implements INDEX(array, row_num, [col_num]); col_num defaults |
| 119 | // to 1. This engine has no array/spill Value kind, so row_num=0 or |
| 120 | // col_num=0 ("return the whole row/column") is unsupported and returns |
| 121 | // #VALUE! rather than silently returning one arbitrary cell of it. |
| 122 | func fnIndex(argNodes []Node, env Env) Value { |
| 123 | if len(argNodes) < 2 || len(argNodes) > 3 { |
| 124 | return errValue(ErrValue) |
| 125 | } |
| 126 | rowV := Eval(argNodes[1], env) |
| 127 | if rowV.Kind == KindErr { |
| 128 | return rowV |
| 129 | } |
| 130 | rowF, ek := toNumber(rowV) |
| 131 | if ek != ErrNone { |
| 132 | return errValue(ek) |
| 133 | } |
| 134 | rowN := int(rowF) |
| 135 | |
| 136 | colN := 1 |
| 137 | if len(argNodes) == 3 { |
| 138 | colV := Eval(argNodes[2], env) |
| 139 | if colV.Kind == KindErr { |
| 140 | return colV |
| 141 | } |
| 142 | colF, ek := toNumber(colV) |
| 143 | if ek != ErrNone { |
| 144 | return errValue(ek) |
| 145 | } |
| 146 | colN = int(colF) |
| 147 | } |
| 148 | if rowN < 1 || colN < 1 { |
| 149 | return errValue(ErrValue) |
| 150 | } |
| 151 | |
| 152 | rng, ok := argNodes[0].(*Range) |
| 153 | if !ok { |
| 154 | if rowN != 1 || colN != 1 { |
| 155 | return errValue(ErrRef) |
| 156 | } |
| 157 | return Eval(argNodes[0], env) |
| 158 | } |
| 159 | r0, r1, c0, c1, withinCap := boundedRangeBounds(rng) |
| 160 | if !withinCap { |
| 161 | return errValue(ErrRef) |
| 162 | } |
| 163 | if r0+rowN-1 > r1 || c0+colN-1 > c1 { |
| 164 | return errValue(ErrRef) |
| 165 | } |
| 166 | return cellValueAt(env, r0+rowN-1, c0+colN-1) |
| 167 | } |
| 168 | |
| 169 | // fnMatch implements MATCH(lookup_value, lookup_array, [match_type]). |
| 170 | // lookup_array must be a single row or a single column — Excel's own |
| 171 | // restriction, since MATCH returns one position, not a (row, col) pair. |
| 172 | // match_type: 1 (default, largest value <= lookup, ascending order), |
| 173 | // 0 (exact match, any order), -1 (smallest value >= lookup, descending |
| 174 | // order). No match is #N/A. |
| 175 | func fnMatch(argNodes []Node, env Env) Value { |
| 176 | if len(argNodes) < 2 || len(argNodes) > 3 { |
| 177 | return errValue(ErrValue) |
| 178 | } |
| 179 | lookup := Eval(argNodes[0], env) |
| 180 | if lookup.Kind == KindErr { |
| 181 | return lookup |
| 182 | } |
| 183 | rng, ok := argNodes[1].(*Range) |
| 184 | if !ok { |
| 185 | return errValue(ErrValue) |
| 186 | } |
| 187 | matchType := 1 |
| 188 | if len(argNodes) == 3 { |
| 189 | mtV := Eval(argNodes[2], env) |
| 190 | if mtV.Kind == KindErr { |
| 191 | return mtV |
| 192 | } |
| 193 | mtF, ek := toNumber(mtV) |
| 194 | if ek != ErrNone { |
| 195 | return errValue(ek) |
| 196 | } |
| 197 | matchType = int(mtF) |
| 198 | } |
| 199 | |
| 200 | r0, r1, c0, c1, withinCap := boundedRangeBounds(rng) |
| 201 | if !withinCap { |
| 202 | return errValue(ErrRef) |
| 203 | } |
| 204 | if r0 != r1 && c0 != c1 { |
| 205 | return errValue(ErrNA) |
| 206 | } |
| 207 | |
| 208 | switch matchType { |
| 209 | case 0: |
| 210 | return matchExact(env, r0, r1, c0, c1, lookup) |
| 211 | case 1: |
| 212 | return matchNearest(env, r0, r1, c0, c1, lookup, false) |
| 213 | case -1: |
| 214 | return matchNearest(env, r0, r1, c0, c1, lookup, true) |
| 215 | } |
| 216 | return errValue(ErrValue) |
| 217 | } |
| 218 | |
| 219 | func matchExact(env Env, r0, r1, c0, c1 int, lookup Value) Value { |
| 220 | pos := 0 |
| 221 | for row := r0; row <= r1; row++ { |
| 222 | for col := c0; col <= c1; col++ { |
| 223 | pos++ |
| 224 | if compareValues(cellValueAt(env, row, col), lookup) == 0 { |
| 225 | return Value{Kind: KindNum, Num: float64(pos)} |
| 226 | } |
| 227 | } |
| 228 | } |
| 229 | return errValue(ErrNA) |
| 230 | } |
| 231 | |
| 232 | // matchNearest implements match_type 1 (descending=false: largest value |
| 233 | // <= lookup, ascending order) and match_type -1 (descending=true: |
| 234 | // smallest value >= lookup, descending order). Both stop at the first |
| 235 | // position that breaks the required ordering, returning the best |
| 236 | // position seen so far. |
| 237 | func matchNearest(env Env, r0, r1, c0, c1 int, lookup Value, descending bool) Value { |
| 238 | found := false |
| 239 | bestPos := 0 |
| 240 | pos := 0 |
| 241 | for row := r0; row <= r1; row++ { |
| 242 | for col := c0; col <= c1; col++ { |
| 243 | pos++ |
| 244 | c := compareValues(cellValueAt(env, row, col), lookup) |
| 245 | ok := c <= 0 |
| 246 | if descending { |
| 247 | ok = c >= 0 |
| 248 | } |
| 249 | if ok { |
| 250 | bestPos = pos |
| 251 | found = true |
| 252 | continue |
| 253 | } |
| 254 | if found { |
| 255 | return Value{Kind: KindNum, Num: float64(bestPos)} |
| 256 | } |
| 257 | return errValue(ErrNA) |
| 258 | } |
| 259 | } |
| 260 | if found { |
| 261 | return Value{Kind: KindNum, Num: float64(bestPos)} |
| 262 | } |
| 263 | return errValue(ErrNA) |
| 264 | } |
| 265 | |
amadan