Re: Proposed changes to sys/stats

[Christopher Collins]

Thanks for the feedback, all.  I will commit the first three changes to
the develop branch, but leave proposal 4 unimplemented.

Chris

On Tue, Feb 16, 2016 at 01:35:16PM -0800, will sanfilippo wrote:
> +1 for 1, 2 and 3.
> 
> Not sure about 4. I am fine with using the macros and prefacing the element 
> names but not sure what other issues this might cause so I will abstain.
> 
> Will
> 
> > On Feb 16, 2016, at 1:13 PM, Sterling Hughes  wrote:
> > 
> > 
> >> 1. The STATS_SECT_START and STATS_SECT_END macros just define a struct;
> >> they don't create an instance.  Generally, these macros would be used
> >> in a header file so that source files can have access to the struct
> >> definition.
> >> 
> >> 2. The addition of a STATS_SECT_DECL macro.  This macro would be used in
> >> two places:
> >> * In source files to instantiate a stats struct.
> >> * In header files to expose an extern declaration of a stats
> >>   instance.
> >> 
> >> 3. As a consequence of the above two points: the names of struct
> >> instances are no longer auto-generated.  The user needs to specify the
> >> exact name.  All macros which derive the instance name from the struct
> >> name are changed: now they just accept the instance name directly.
> >> 
> > 
> > 
> > +1
> > 
> > 
> >> 4. Remove the "s" which prefaces the name of each stat field in a
> >> statistics struct.  By doing this, the STATS_SECT_VAR, STATS_INC, and
> >> STATS_INCN macros can be removed.
> >> 
> > 
> > -1
> > 
> > I think we want these macros used everywhere: they allow us more 
> > flexibility to refactor this, and there is a relatively well defined access 
> > mechanism for stats.
> > 
> > Sterling
> 

Re: Proposed changes to sys/stats

[Sterling Hughes]

1. The STATS_SECT_START and STATS_SECT_END macros just define a struct;
they don't create an instance.  Generally, these macros would be used
in a header file so that source files can have access to the struct
definition.

2. The addition of a STATS_SECT_DECL macro.  This macro would be used in
two places:
 * In source files to instantiate a stats struct.
 * In header files to expose an extern declaration of a stats
   instance.

3. As a consequence of the above two points: the names of struct
instances are no longer auto-generated.  The user needs to specify the
exact name.  All macros which derive the instance name from the struct
name are changed: now they just accept the instance name directly.




+1



4. Remove the "s" which prefaces the name of each stat field in a
statistics struct.  By doing this, the STATS_SECT_VAR, STATS_INC, and
STATS_INCN macros can be removed.



-1

I think we want these macros used everywhere: they allow us more 
flexibility to refactor this, and there is a relatively well defined 
access mechanism for stats.


Sterling