Re: [Monotone-devel] RFC: Restricting app_state to commands

[Markus Schiltknecht]

Hi,

Zack Weinberg wrote:
> Your point is well-taken - but I think you're not being ambitious
> enough.  

Huh? Did you see the n.v.m.experiment.encapsulation branch already?

> As you mention, app.signers and app.verifiers ought to be
> combined with the key_store - but probably so should app.agent; I
> haven't looked at the code in enough detail to be sure, but I think
> Justin just made it a separate object because that was the path of
> least resistance, not because it actually *needs* to be separate - and
> logically, it is part of the key store.

Agreed.

> Furthermore, if you look at the specific options and hooks that are
> being used, I'd argue that they ought to be properties of the
> key_store too.  With the exception of the get_passphrase hook (which
> we are killing, aren't we?) they are constant across the application
> run.

Sure. If we are going to split the app_state into multiple context 
objects, we need to sort out what to put where. See the 
README.encapsulation in that branch.

> > Anyway, I would also really like to get a better separation. I have
> > started with (partly) ripping out the database from app_state and
> > passing only that where possible.
>
> Cool.  I will look at your branch when I get a chance (probably not this 
> week).

:-) answers my question above... Anyway, I'd certainly like to get some 
review before going all the way to change calling conventions for a 
whole lot of objects and functions.

> > How do you count that stuff? Is there a script? Or a makefile target or
> > something?
>
> Build before and after (in two separate directories) with "make
> CXXFLAGS='-g -O2 -save-temps'", and then
>
> $ (cd before && wc -l *.ii) > A
> $ (cd before && wc -l *.s) > As
> $ (cd after && wc -l *.ii) > B
> $ (cd after && wc -l *.s) > Bs
> $ paste A B As Bs |  awk 'BEGIN{ OFS="\t"; print "file", ".ii", ".s" }
> { if ($3 - $1 != 0 || $7 - $5 != 0) { sub(/\.ii/, "", $2); print $2,
> $3 - $1, $7 - $5 } }'
>
> (I really wish coreutils' join(1) utility were more powerful, but this 
> works.)

Thanks for these instructions, very helpful! Especially the -save-temps 
option is new to me.

Regards

Markus