Re: Support for XLogRecPtr in expand_fmt_string?

On 07.07.2012 01:03, Peter Eisentraut wrote:
> On tis, 2012-07-03 at 14:52 -0400, Tom Lane wrote:
>> Peter Eisentraut<peter_e@gmx.net>  writes:
>>> On tis, 2012-07-03 at 19:35 +0200, Andres Freund wrote:
>>>> I wonder if we just should add a format code like %R or something similar as a
>>>> replacement for the %X/%X notion.
>>
>>> Maybe just print it as a single 64-bit value from now on.
>>
>> That'd be problematic also, because of the lack of standardization of
>> the format code for uint64.  We could write things like
>>     "message... " UINT64_FORMAT " ...more message"
>> but I wonder how well the translation tools would work with that;
>> and anyway it would at least double the translation effort for
>> messages containing such things.
>
> The existing uses of INT64_FORMAT and UINT64_FORMAT show how this is
> done:  You print the value in a temporary buffer and use %s in the final
> string.  It's not terribly pretty, but it's been done this way forever,
> including in xlog code, so there shouldn't be a reason to hesitate about
> the use for this particular case.

That's hardly any simpler than what we have now.

On 03.07.2012 21:09, Tom Lane wrote:> Andres Freund<andres@2ndquadrant.com>  writes:>> I wonder if we just should add a
formatcode like %R or something 
 
similar as a>> replacement for the %X/%X notion.>> Only if you can explain how to teach gcc what it means for elog
argument>match checking.  %m is a special case in that it matches up with a> longstanding glibc-ism that gcc knows
about. Adding format codes of our> own invention would be problematic.
 

One idea would be to use a macro, like this:

#define XLOGRECPTR_FMT_ARGS(recptr) (uint32) ((recptr) >> 32), (uint32) 
(recptr)

elog(LOG, "current WAL location is %X/%X", XLOGRECPTR_FMT_ARGS(RecPtr));

One downside is that at first glance, that elog() looks broken, because 
the number of arguments don't appear to match the format string.

--   Heikki Linnakangas  EnterpriseDB   http://www.enterprisedb.com