Re: seq_file and exporting dynamically allocated data

[viro@parcelfarce.linux.theplanet.co.uk]

On Mon, Nov 17, 2003 at 01:03:39AM -0800, William Lee Irwin III wrote:
> On Mon, Nov 17, 2003 at 12:48:04AM -0800, William Lee Irwin III wrote:
> > There's a retry loop where the buffer size is doubled each iteration
> > that looks to me like automagic sizing in the code for seq_read(). I
> > can't say I've actually tried to rely on getting more than a page
> > at a time through seq_read(), though.
> 
> Woops. This looks like it only makes sure there's enough to get the
> first ->show() into the buffer; I see that it later gives up when
> m->count == m->size once the first ->show() has enough bufferspace to
> complete later on. So if all ->show() operations to do are less than
> PAGE_SIZE, it'll never hand back more than a page at a time, and may
> hand back short reads prior to EOF, which doesn't bode well for my
> short read == EOF idea (maybe not a great assumption in general).

That's "broken assumption", thank you very much.  There is one and only one
indication of EOF - read() returning 0.  Any application that treats short
read as EOF is broken.  As in "apply Stevens to offender's cranium until
something gives; if it's the book, replace with fresh copy and repeat".