Re: [Simulavr-devel] avrtest and simulavr vs simulavrxx

[Joerg Wunsch]

As Klaus Rudolph wrote:

> I looked in the core_avr_dump.core
> 
> I dislike it!

:)

I didn't really want you to like it.  Frankly, my idea is to:

. add it with the -C option for backwards compatibility, and
. go ahead implementing a more versatile format (like, and ELF based
  one) which can also be read back, say using option -E (--elf-coredump)

Obviously, reading back an old status would not be necessary at all for
the compat file, but might indeed be a nice addition for the -E one.

> 1)
> eeprom section missing

Well, could be added, but if it's only for compat, it would not be
needed (as it has never been there).  Should certainly be around there
in a new format file.

> 2)
> actually we have no register names in simulavrxx. It is not a big deal 
> to add them, but I think that is wasting the code a bit. If we have this 
> text file only as an intermediate step, we change a lot of code and drop 
> it later.

You're talking about IO registers, are you?

Having the names around is just good for human readers, and might also
look good e.g. inside a trace file (so humans can e.g. quickly find
out all spots where UDR is written etc. pp.)  Even for the backwards
compat case, I don't think anyone would care about more than
SPL/SPH/SREG, and these could be added at fixed locations.  Since GCC
relies on them being at fixed locations, there's not much need to
bother more here.  (Just display all other IO registers as Reserved
then if you don't know better.)

> 3)
> writing to an elf give all these informations and you can use it also 
> for other purpose without any scripting in between.

I didn't want to give the impression that I don't like the ELF coredump
idea.

> If users actually use a script to read this textfile, they also can use 
> a script to read an elf file. I think this could not be a problem.

The only problem is that users have to adjust their scripts then.  I
simply don't trust Eric's judgment that avr-libc's testsuite is really
the only user.  simulavr has been around for years.  I see four
different users alone who filed bug reports against it within the last
year, and only one out of these four was Dmitry Khmelkov (the maintainer
of the avr-libc test suite).  So that makes for at least three further
users *whom we've heard from*, and I guess adding a factor of at
least ten for the "black unseen masses" is usually a gross
underestimation.  You'll simply never hear something from the remainder
unless you severely break something...

> I vote for an elf file :-)

I vote for both, optionally the historic format, and an ELF file for
all new users.

-- 
cheers, J"org               .-.-.   --... ...--   -.. .  DL8DTL

http://www.sax.de/~joerg/                        NIC: JW11-RIPE
Never trust an operating system you don't have sources for. ;-)