This is the mail archive of the
archer@sourceware.org
mailing list for the Archer project.
Froggy planning
- From: Chris Moller <cmoller at redhat dot com>
- To: Project Archer <archer at sourceware dot org>
- Date: Tue, 29 Jul 2008 22:30:34 -0400
- Subject: Froggy planning
I'm far enough along in froggy to start planning for the ultimate user
interface, of which I see, at the moment, three possibilities:
The first is a very low level i/f consisting ioctl() operations to the
froggy pseudo-file--if you look in froggy-test.c, that's what's there.
There are two up-sides of using ioctl()s: it's slightly more efficient
than the alternatives, and it's obviously more flexible. The down-sides
are that it's harder to code the ioctl()s and, if you do it wrong, you
and your app can get very, very, confused. (That's because the data
returned to the app from the module is in a streamed header/data format
where the length of the header depends on its type and the length of the
data is specified in the header--obviously if you don't do the right
things with the lengths, things will come unstuck in a hurry. (I could,
I suppose, write something that tries to resync out-of-sync streams, but
that adds overhead and complexity.))
Option two is to provide a higher level i/f that wraps the ioctl()s in
fcn calls in a library (libfroggy or libutrace or something.) That's a
good bit safer, of course, mostly because the packet lengths are buried,
and the app writer don't have to deal with _IOW macros and the like.
The down side is a little loss of efficiency--there shouldn't be much,
if any, loss of flexibility--pretty much anything can be added to the API.
A third possibilty comes to mind that's a kind of a superset of option
two but, in addition, hides the response thread, or at least more of the
messy details in it. Option two would probably provide a wrapper around
a blocking read that would return complete, partially decoded, packets
and guarantee packet synchronisation, but the loop that fcn is in would
be the responsibility of the app writer. Option three might
conceptually handle the loop as well, probably with registered callbacks
to do whatever needs to be done with any given packet type.
Or any other option people want to propose...
Let me know.
--
Chris Moller
I know that you believe you understand what you think I said, but
I'm not sure you realize that what you heard is not what I meant.
-- Robert McCloskey
Attachment:
signature.asc
Description: OpenPGP digital signature