CPU Supplement Reorg Proposal

[Joel Sherrill]

Hi,

Last week, I reached a conclusion to a problem that has long
bothered me.  The sizing and timing information in the CPU
supplements is always out of date.  Plus the sizing information
is based on a methodology that was developed over 10 years ago
and I do not believe it reflects what users today really want.
The old code sizing information was based on arbitrarily pulling
individual functions out of the RTEMS library with an "ld -r".
It did not account for the BSP, file system, system call, or malloc
family infrastructure.  And it varied based upon RTEMS configure
options, RTEMS version, compiler version, and compiler flags.

So I am proposing that this information be eliminated from the
supplements.  Along with this I am proposing that the remaining
information in the supplements be merged into a single document
with one chapter per architecture supported.  I have a prototype
of this and the merged document is not a whole lot longer than
a single CPU supplement.

This will reduce the documentation for a port to a single file
of 6-10 pages in formatted length.

All the sizing information I have given out in the past five
years is based upon "Application Profiles" which are fully
linked baseline samples for hello world, a TCP/IP applciation,
HTTPD, FTPD, and HTTPD/FTPD.  All are sized with and without
POSIX enabled and the measure is for a fully linked executable
on a particular board.  This is a more reasonable and useful
measurement.  I think all of the applications used are already
in the RTEMS source tree so it would be a matter of capturing
the standard configurations and reporting format.

Comments.

--joel