Paraphrasing Goldstein
If we've $N$ number of particles and there exist holonomic constraints, expressed in $k$ equations, then we may use these equations to eliminate $k$ of the $3 N$ coordinates, and we are left with $3 N-k$ independent coordinates. We introduce $3 N-k$ generalized coordinates $q_{1}, q_{2}, \ldots, q_{3 N-k}$ in terms of which the old coordinates $\mathbf{r}_{1}, \mathbf{r}_{2}, \ldots, \mathbf{r}_{N}$ are expressed by equations of the form $$ \begin{aligned} \mathbf{r} &=\mathbf{r}_{1}\left(q_{1}, q_{2}, \ldots q_{3 N-k}, t\right) \\ & \vdots \\ \mathbf{r}_{N} &=\mathbf{r}_{N}\left(q_{1}, q_{2}, \ldots, q_{3 N-k}, t\right) \end{aligned} $$ It is always assumed that we can also transform back from the $\left(q_{l}\right)$ to the $\left(\mathbf{r}_{l}\right)$ set, i.e., that above equations combined with the $k$ equations of constraint can be inverted to obtain any $q_{i}$ as a function of the $\left(\mathbf{r}_{l}\right)$ variable and time.
My doubt: The above set of $N$ vector equations are $3N$ scalar equations. If we want to find all the $q_i$ we can solve $3N-k$ equations from the above set of equations and be done. In other words it's given that there are already $3N$ equations in $3N−k$ variables, which we can solve without the k constraints.
Why then does the author say that he also needs the $k$ equations of constraint?