Re: libpq: PQcmdStatus, PQcmdTuples signatures can be painlessly improved

,--- You/Peter (Tue, 10 Jan 2012 19:13:42 +0200) ----*
| On tis, 2011-12-13 at 07:55 -0500, Alex Goncharov wrote:
| >   char *PQcmdStatus(PGresult *res);
| >   char *PQcmdTuples(PGresult *res);
| > 
| > Unreasonable:
| > 
| >   a. What, these two can modify 'res' I pass in?..
| > 
| >   b. Oh, yes, because they return 'char *' pointing to
| >      'res->cmdStatus+n', so, a libpq user may write:
| > 
| >         char *s = PQcmdStatus(res);
| >         *s = 'x';
| > 
| >      and have 'res' modified.  (Would be the user's fault, of course.)
| > 
| Note that const PGresult * would only warn against changing the
| fields

It would not warn, it would err (the compilation should fail).

| of the PGresult struct.  It doesn't do anything about changing the data
| pointed to by pointers in the PGresult struct.  So what you are saying
| doesn't follow.

By this logic, passing 'const struct foo *' doesn't have any point and
value, for any function.  But we know that this is done (and thank you
for that) in many cases -- a good style, self-documentation and some
protection.

E.g. here:

,--- I/Alex (Tue, 13 Dec 2011 07:55:45 -0500) ----*
| Compare:
| 
|   int PQntuples(const PGresult *res)
| 
| Reasonable: doesn't modify 'res'.
`-------------------------------------------------*

BTW, I have not submitted the context differences, as suggested, only
because of extreme overload at work and the need to do a careful
caller and documentation analysis. I still hope to be able to do it in
a reasonably near future.

-- Alex -- alex-goncharov@comcast.net --