Re: regclass and format('%I') - Mailing list pgsql-general

It honestly seems far more reasonable to me that %s and %I should do
the exact same thing with regclass. My reasoning is as follows:

‘%I’ formats a something such that it is a valid identifier,

regclass is already a valid identifier,

therefore, do nothing.

Another line of reasoning:

If you format with ‘%s’ you are saying: I don’t care whether it’s a
valid identifier or literal or whatever, just put the string there,

but when we sub a regclass into a string, we want it to be a valid identifier,

therefore we should write ‘%I’ when subbing it, so as not to confuse
our readers,

therefore ‘%I’ should do nothing.

On 13 March 2015 at 12:42, David G. Johnston <david.g.johnston@gmail.com> wrote:
> On Fri, Mar 13, 2015 at 12:18 PM, Jason Dusek <jason.dusek@gmail.com> wrote:
>>
>> Hi All,
>>
>> The difference in how format handles `regclass` and `name` seems like an
>> inconsistency:
>>
>>     WITH conversions(casts, format, result) AS (
>>     VALUES (ARRAY['name']::regtype[],             '%I', format('%I',
>> name('select'))),
>>            (ARRAY['name']::regtype[],             '%L', format('%L',
>> name('select'))),
>>            (ARRAY['name']::regtype[],             '%s', format('%s',
>> name('select'))),
>>            (ARRAY['regclass']::regtype[],         '%I', format('%I',
>> regclass('select'))),
>>            (ARRAY['regclass']::regtype[],         '%L', format('%L',
>> regclass('select'))),
>>            (ARRAY['regclass']::regtype[],         '%s', format('%s',
>> regclass('select'))),
>>            (ARRAY['regclass', 'name']::regtype[], '%I', format('%I',
>> name(regclass('select')))),
>>            (ARRAY['regclass', 'name']::regtype[], '%L', format('%L',
>> name(regclass('select')))),
>>            (ARRAY['regclass', 'name']::regtype[], '%s', format('%s',
>> name(regclass('select'))))
>>     ) SELECT * FROM conversions;
>>           casts      | format |    result
>>     -----------------+--------+--------------
>>      {name}          | %I     | "select"
>>      {name}          | %L     | 'select'
>>      {name}          | %s     | select
>>      {regclass}      | %I     | """select"""
>>      {regclass}      | %L     | '"select"'
>>      {regclass}      | %s     | "select"
>>      {regclass,name} | %I     | """select"""
>>      {regclass,name} | %L     | '"select"'
>>      {regclass,name} | %s     | "select"
>>
>> My assumption is that they both represent valid SQL identifiers, so it
>> stands
>> to reason that `%I` should result in a valid identifier for both of them
>> (or
>> neither one).
>
>
> All three of the %I results are valid identifiers.
>
> regclass performs the same conversion that %I performs.  But since the
> output of the regclass conversion is a valid identifier, with double-quotes,
> the %I adds another pair of double-quotes and doubles-up the existing pair
> thus leaving you with 6.
>
> <select> is a reserved word and thus can only be used as an identifier if it
> is surrounded in double-quotes.  name() doesn't care (not that it is
> user-documented that I can find) about making its value usable as an
> identifier so when its output goes through %I you get the expected value.
>
> If you are going to use regclass you want to use %s to insert the result
> into your string; not %I.
>
> David J.
>