Hi All
Currently, in NPOIFS (NPOIFSFileSystem + POIFSFileSystem, but not
OPOIFSFileSystem), you can either load from a File or from a Stream.
With a Stream, everything gets buffered into memory
When loading from a File, you can load read-write or read-only. In
read-write mode, we use NIO to mmap the file so we can quickly + low
memory read + write.
The other case, File + read-only, isn't so ideal. We need to support
both people doing true read-only cases, eg reading or text extraction,
plus people doing read-template + write or read-change-saveas cases. To
support that, we buffer into memory.
Taking another look at the NIO stuff for bug #60670, I was wondering
about splitting those two cases out. For true read-only cases, we could
then mmap in read-only mode. For cases where you're loading a file and
want to change + save elsewhere with the original unchanged, I think we
could mmap in "private" mode, which looks to be copy-on-write, so only
changes would be buffered in memory.
Any thoughts? Is it worth making the change? Can anyone spot any
downsides to the change? Anyone think of any good enum names for the
different modes if we replace the read-only boolean with a 3-way enum?
(Changing all file-based access to mmap, rather than a mix of mmap +
in-memory buffers, should make it easier to tweak the logic to mmap
bigger chunks)
Cheers
Nick
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
