Adjusted .hgignore a bit more (added new targets for debug and profile mode) and
merged. Thanks!

Done and pushed.

Merged. Two things remain to be done so that Jendrik's scripts work again:

 1. Adapt build_all so that it creates all versions.

 2. Create a script called "downward" that calls the appropriate version of the
    dispatcher and has the same interface as the original downward executable
    (so that Jendrik's scripts keep working). I'm attaching one that should
    work; let me know what you think. We should then commit this and remove
    "downward" from .hgignore.

Sorry, brain fart. Forget the previous comment. :-)

The dispatch script looks buggy; maybe it skips over lines sometimes?
When I replace the "print 1" with "print max", I get an output of "8" on the
example I tested, but the correct value would have been "15".

Done and pushed

On 29.12.2010 19:24, Erez Karpas wrote:
> 
> Erez Karpas <batman@gmail.com> added the comment:
> 
> You're right.
> I'll change it to unsigned short/unsigned int,
> and adjust the gawk script.

I'd stick to unsigned short and (plain) int -- I think we assign from
state_var_t to int in many places, and assigning from unsigned int to
int can overflow. It's not an issue to have state vars limited at 2G, I
think. ;-)

You're right.
I'll change it to unsigned short/unsigned int,
and adjust the gawk script.

> if (max < 255) print 1; else if (max < 65524

Shouldn't the first be "<= 255"?

The second should be "<= 65535" if we used unsigned short, but since we use
short, it should be "<= 32767".

I don't like that you removed -g from the default build options.

-g is needed for debugging, and last time I checked it didn't slow down the code
at all (since the debug info is only metadata). I'm willing to be convinced to
turn this off if any benefit can be demonstrated, but otherwise it should be
kept on.

Pushed to my repository.
Make now accepts a STATE_VAR_BYTES parameter (1,2, or 4).
Each added .1 .2 or .4 to the objects and -1 -2 or -4 to the binary.

The ./build script should call
make STATE_VAR_BYTES=1
make STATE_VAR_BYTES=2
make STATE_VAR_BYTES=4

I also created a dispatch script, which takes the preprocessor output, and 
outputs 1 2 or 4 depending on the biggest variable domain size.
So the ./plan script should (after preprocessing) call:

/downward-`.dispatch output` OPTIONS

I did not change from -m32 to -m64 yet, but this should be done before the 
competition.

I'd like to also have an int option in there. The translator usually doesn't
create such large domains of course, but we occasionally get them from other
sources (e.g. Jendrik's fluent merging code could generate them). Not really
needed right now, but I don't think it should be significant extra work, right?

This could be added most cleanly to the Makefile by adding another group of
suffixes like the current "-debug", "-profile" and "" (for release); e.g.
"-char", "-short", "-int" or "-1", "-2", "-4". No preferences on whether this
should come before or after the "-debug"/"-profile" suffix.

1. Change state_var_t to unsigned char as default, with an option for short
2. Make this a compiler flag
3. Change Makefile to generate both versions
4. Write a dispatcher script