On 10/24/2010 05:26 PM, Pádraig Brady wrote:
> On a related note I was wondering if we should fall back
> to increasing the buffer by +=BUFSIZ rather than *=2
> when we get ENOMEM?
I wouldn't bother, as it would lead to O(N**2) behavior,
which could well be worse than the problem that it'd cure.
(Currently it's not *=2, by the way, it's more like *=1.5.)
I found a couple of minor problems in the patch. This comment needs to
be updated, to match the new behavior:
/* For a regular file, allocate a buffer that has exactly the right
size....
since the buffer is now 1 byte "too large". Also, this line:
if (SIZE_MAX <= alloc_off)
should be changed to:
if (SIZE_MAX - 2 < alloc_off)
in order to detect the overflow that now occurs when
st.st_size - ftello (stream) == SIZE_MAX - 1. Using
it "SIZE_MAX - 2 <" rather than "SIZE_MAX - 1 <=" will
make the relationship between this "2" and the "2" in
"alloc_off + 2" more obvious to the reader.
