The SQL standard sorts the SQLSTATE table in a first step according to the textual representation of class (first two byte) and in a second step to the textual representation of subclass (next 3 byte) - keeping the text '(no subclass)' for subclass '000' on top. Obviously for some people this 'textual' sort order is relevant.
For my personal use I strongly prefer a pure 'numeric' sort order of SQLSTATE. Therefore I created a Perl program, which generates the SGML file out of errcode.txt and published it in https://www.postgresql.org/message-id/d69ce4f1-dd4e-61c1-070e-25f662e9c622%40purtz.de . This was a follow-up action after a discussion about illegal SQLSTATE values in PG, see: https://www.postgresql.org/message-id/11558.1513819695%40sss.pgh.pa.us.
PG documentation uses a numeric sort order for class and a semantic sort order for subclass.
Which one of the three is best? Here is my judgement: For technical oriented people the first approach is horrible because it sorts not only the subclass-text but also the class-text. This scrambles important and less important values to arbitrary positions. For me, sorting of the numeric value of class in a first step is essential. The third approach may be good for persons which work on the improvement of the PG source code. But they can refer to errcodes.txt. The primary readers of our documentation are users which work in roles as DBA or application developer. I doubt that they are interested in the semantic order of subclass. But I do not know whether they prefer numeric to text or text to numeric translation.
Jürgen Purtz
On 27.04.2018 17:16, Tom Lane wrote:
Bruce Momjian <bruce@momjian.us> writes:
On Fri, Apr 27, 2018 at 10:08:12AM +0000, PG Doc comments form wrote:
Would it be possible to present the codes in ascending sequence?
Uh, I am guessing this is the order listed in the SQL standard. How
would you order them differently? Have the letters only at the end?
There was some discussion of this a couple months ago, which I'm too lazy
to search the archives for right now, but it trailed off without any
agreement on what to do.
There are at least three different ordering principles that might be
applied here: numeric by code (well, perhaps more like "ASCII sort
order"); alphabetical by exception name (but probably still keeping the
000 category codes at the top); or semantic grouping (ie try to keep
related errors together). It looks to me like all three of these have
been applied in different places ;-).
One idea that might help is to present two tables with different sort
orders, say strict numeric and strict alphabetical. (If we did that,
I'd be inclined to leave errcodes.txt alone and put the sorting
responsibility on the script that converts it to SGML.)
One question that has to be asked is what we think the use-case for this
table is at all. Different use-cases result in different ideas about the
best ordering.
regards, tom lane