Re: [Netnice-daemon] netniced Linux port

[Takashi O. <ta...@cs...>]

hi,

Sergei Matusevich wrote:
> 
> Okay, here's another tiny patch and now we have netniced
> compiled for Linux! I have also managed to compile it with
> gcc -Wall, mostly by commenting out some variables that
> were declared but never used in the code. Please let me
> know if we need this patch as well.

thanks for the patch, again. i'll apply it shortly and commit.
regarding the aesthetic changes, i think we may defer, since
we'll need cleaning of the code anyway.

> In addition, I have some general remarks about our code.
> First of all, in our header files, we never use common
> #ifndef/#define technique to prevent errors when the header
> is included twice in some source (directly or indirectly).
> I think we should add add such wrappers to each .h file -
> please see the classes.h patch attached for an example.

i definitely have no objection about this. just take the current
code as a prototype.


> I am also extremely confused with String, Integer, Boolean,
> Process, ProcessGroup etc. typedefs defined in the
> classes.h. It is very hard to tell from the code, whether,
> say, SocketGroup is an object or a pointer. Things get even
> more obscure if we decide to pass SocketGroup by reference
> or store it in an STL container. I would suggest to get rid
> of these typedefs completely in future.

i think they are for automatic document generation by doc++. the problem
is this; although the code is written in C++, the user needed to have a
class library documentation (for the Java-like scripting language in the
former version).

	http://www.netnice.org/documents/doc/index.html

i used a lot of typedefs, just to fill this gap. if you have a good
wayaround for this problem, i'm flexible also about this, as long as
we provide a complete set of the class description automatically generated
from the source code at a reasonable cost.


> I have also noticed that some netniced classes, e.g.
> process, socket, processgroup, socketgroup, have a lot of
> things in common in both interface and implementation.
> Maybe it make sense to move the common functionality and/or
> interface to some generic base class and inherit from it.

for the same reason as above, automatic document generation, 
i think the inheritance approach makes things more complicated.
but, to write better common routines (e.g. for procfs access) is
a good idea. i believe that netnice/netnice-common/src/netniced/TODO
has a lot of suggestions for improvement, which can be a good starting
point.

thanks!


-- taka

p.s.

btw, do you think the package structure, to support various platforms
with a package, reasonable enough? for example, we have platform-dependent
directories and a common directory (netnice-common), with some others
(doc and contrib). do you like the idea? or, you have better plan?