Buffer for page-aligned I/O
This kind of buffer uses page-aligned bigarrays. Data can only be added to the end, or deleted at the beginning of the buffer.
The idea of this buffer is that input data is added to the last
page of the buffer only (with add_inplace
). If then all previous
input was already a multiple of the page size, it is ensured that
the new input is added at a page boundary. This kind of input operation
can often be accelerated by the OS ("zero copy network I/O").
create blocksize
: creates new buffer with this blocksize
, which must
be a whole multiple of the page size of the OS
The blocksize
is normally either
These cases are optimized, and the buffer is allocated in a shared pool.
Blits contents to another memory buffer
Adds a sub memory buffer to the end of the buffer
add_inplace b f
: Calls f m pos len
where m
is the last page
of the buffer, and pos
is the first free byte on the page, and
len
is the number of free bytes on the page. The function f
is
expected to store new data in m
from pos
to pos+n-1
and to
return n
. The number n
is also returned as final result.
It is ensured that f
is called with a value of len>=1
.
let (m,pos,len) = page_for_additions b
: Returns the last page
in m
, the first free byte on the page in pos
, and
the number of free bytes on the page in len
. (The same values
the function f
would have got as arguments in add_inplace
.)
advance b n
: Marks further n
bytes in the last page of the
buffer as used. These bytes are not modified in any way - it is
expected that the user calls page_for_additions
first, and sets
these n
bytes to new values directly.
let (m,pos,len) = page_for_consumption b
: Returns the first page
in m
, the first used byte on the page in pos
, and
the number of used bytes on the page in len
.