[bind10-dev] Enums as array indices and manual vs. auto-numbering

[JINMEI Tatuya / 神明達哉]

At Thu, 25 Aug 2011 21:06:21 +0900,
Yoshitaka Aharen <aharen at jprs.co.jp> wrote:

> +1 to the opinion. If there's no rationale behind numbering enums
> manually, I prefer to assign numbers automatically. I also believe that
> it is safer.
> 
> Google Style Guide shows an example which omits assigning numbers:
> http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml?showone=Enumerator_Names#Enumerator_Names

Thanks for the reference, but this doesn't seem to be related to
whether to prefer specifying the enum values explicitly.

On the main point, I have a mixture feeling as I mentioned in #1068
(which I believe is the trigger of this thread):
http://bind10.isc.org/ticket/1068#comment:8

<quote>
Personally I have a mixture feeling about this. On one
hand I think your point is valid. On the other hand, I don't like to
depend too much on implicit assumptions even if they are guaranteed by
the language spec - just like it's error prone to enumerate everything
by hand (I agree on that), I think it's also the case code relying on
implicit assumptions is error prone. Someone who doesn't understand
the rationale may, perhaps for a bogus reason, want to change the
initial value of the enum to non 0:

    enum DeleteRecordParams {
        DEL_NAME = 1, ///< The owner name of the record (a domain name)
                      ///< It begins with 1 as it seems to make more sense
        DEL_TYPE, ///< The RRType of the record (A/NS/TXT etc.)
        DEL_RDATA ///< Full text representation of the record's RDATA
    };

And then the code will suddenly start failing in a strange way.
In this specific it would be easy to detect and fix it via unittests,
but there may be more subtle cases. We could also add a note that the
use of the default definitions has a reason and should not be changed
in comments, but commenting something is also an error-prone human
involved practice.
</quote>

So, I personally don't think that the use of implicit assignments is
as safe as it might look.  But this is not a strong opinion.  I don't
necessarily disagree with switching the style if others prefer that
way.

So my vote would be -0.1 (mostly neutral, but with a slight preference
on *not* relying on implicit assumptions).

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.