← board

A 2-D array parameter loses its row length

Measured (before)

int m[3][4];                       /* filled with i*4+j */
int g(int a[][4]) { return a[1][2]; }

gcc: 6, pxx: 3 — that is m[0][3], i.e. the index flattened as 1 + 2 instead of 1*4 + 2. int g(int (*a)[4]) answered 6, and the two declarators name the same parameter type in C.

Root cause

A parameter's trailing brackets were consumed by a balanced-bracket skip whose only job was to reach the end of the declarator, and the parameter was then retyped as a bare pointer-to-element. SymPtrElemArrLen / SymPtrElemNDims — the channel int (*a)[4] already fills, and which a[i][j] reads to compute the stride — were left at 0. So the pointee was int, not int[4].

Fix

Record each bracket's bound while skipping it (a plain integer bound only; a VLA or macro bound stays unknown exactly as before), drop the first dimension — which is the one C decays away — and apply the rest through the same pptrarrlen / pptrndims / pptrdims channel the (*a)[N] form uses. One mechanism, reached by both spellings.

Result

test/carr2d_param_row_length.c — [][4], (*)[4], [3][4], [][3][4], a loop summing through the parameter, and the 1-D and char *[] forms that must stay untouched — returns 42 under both gcc and pxx.

Gate

make compiler/pascal26 + the test + tools/gate.sh quick — GREEN.

Log