Re: [PATCH 04/22] reftable/basics: handle allocation failures in `reftable_calloc()`

Patrick Steinhardt <ps@pks.im> writes:

Since this series is not about eradicating all avenues in reftable library code that can lead to die(), but only about dealing with allocation errors from the underlying malloc/realloc routines, I think the code posted is perfectly fine as-is as a part of this series, but since I noticed something, let me comment before I forget.

When st_mult() detects overflow, you'd still die(), wouldn't you?

We'd probably want a variant of st_mult() that lets us notice a condition that would yield too large a result, and code the above like so,

	size_t sz;
	void *p;

	if (st_mult_gently(nelem, elsize, &sz) ||
            !((p = reftable_malloc(sz))))
		return NULL;
	memset(p, 0, sz);
	return p;

or use the underlying helper ourselves, and say

	size_t sz;
	void *p;

	if (unsigned_mult_overflows(nelem, elsize)) ||
            !((sz = nelem * elsize, p = reftable_malloc(sz))))
		return NULL;
	memset(p, 0, sz);
	return p;

which lets us without an extra helper but after writing it myself, I find it a bit too wordy.

In a sense, it is on the borderline to handle st_mult() overflow in this function for a topic whose theme is about allocation failures.

From the point of view of callers of reftable_calloc(), whether the arguments they are feeding the function is too large to be multiplied or whether the request is too big for the underlying allocator to handle, the end result should be the same: they requested too large an allocation.

So I wouldn't complain that it is out of scope, if use of st_mult() that computes the allocation size is fixed as part of this series. But as I already said, I am also OK if we leave it to a separate series to tackle other potential callers of die().

Thanks.