Selecting a constant question: A summary

Dann Corbit wrote:

First a comment:

At CONNX Solutions Inc., we believe sincerely that we should do whatever
is necessary to make our customers prosper. This means creation of
excellent tools and being responsive to customer needs. Secondly, we
believe that we should treat the customers the way that we want to be
treated.

I think that the PostgreSQL group has managed the first objective, but
not the second. Of course, that is only an opinion, but I think that
success hinges on both factors. Our objective in this issue has also
been to improve PostgreSQL so that it can become more useful to the end
users and not to denigrate the work of the engineers that have toiled on
it. I will also admit that frustration has caused our tone to become
sharp at times. This is clearly a mistake on our part and for this, I
apologize.

Woah, now this is interesting. This morning, I read this whole thread,
wondering what in the world could possibly be taking so long ;).

I will admit that many of us in the community tend to try to provide a
solution without actually understanding the problem. I think it kind of
comes with the territory, a lot of times it seems like FOSS is all about
the work around versus the solution because the solution takes longer.

Perception is a powerful thing. Personally, I didn't see any of the
community doing anything but trying their best to help you with the
problem you were experiencing.

What I did see, is a lot of tenseness from your side, to what basically
amounts to "free" support. Remember that we are here, at no cost to you.

Lastly, the PostgreSQL community doesn't have customer. You have
customer, CMD has customers, the PostgreSQL community does not.

The best correlation I can give you is this. We (the community) are all
a team. You are part of that team. We are not your vendor.

Sincerely,

Joshua D. Drake

--

=== The PostgreSQL Company: Command Prompt, Inc. ===
Sales/Support: +1.503.667.4564 || 24x7/Emergency: +1.800.492.2240
Providing the most comprehensive PostgreSQL solutions since 1997
http://www.commandprompt.com/

Donate to the PostgreSQL Project: http://www.postgresql.org/about/donate
PostgreSQL Replication: http://www.commandprompt.com/products/

Dan,

Secondly, we
believe that we should treat the customers the way that we want to be
treated.
I think that the PostgreSQL group has managed the first objective, but
not the second.

I just read this whole thread, and I feel that the sort of comment above is
completely unjustified, and counterproductive to your goal of adding a
feature to PostgreSQL which will make your driver work better. You'll be a
lot more likely to persuade people in the community to work with you if
you're not trying to convince them to change the project culture at the same
time.

You are on the developer mailing list for an open, community-based open source
project and *not* a commercial company. Therefore we do not have "customers"
and your paradigm is wrong. The PostgreSQL developers *are* treating you
exactly has they expect to be treated; as a developer, meaning that you argue
things out and defend your desire for a change. If you read anybody else's
discussion on this list you'll see that's how everyone else interacts.

If anything you've gotten more than your fair share of attention ... 40+ posts
from 1/2 dozen senior developers in less than 48 hours!

If you would prefer a more formal customer-vendor relationship, then I suggest
that you sign up as a customer of EnterpriseDB, Red Hat, Sun, Command Prompt,
SRA etc. or similar.

Now, that aside:

According to SQL/CLI and ODBC 3.5, we should bind the length of a
character column.

This is a much better approach. Standards are always nice.

But it would be very nice if the database could provide a good estimate
for us so that PostgreSQL could work like all of the other database
systems. Code full of kludges is harder to maintain.

Do you have any information about how binding works in other databases? A
clear roadmap would make it easier for eventual developer implementation, and
obviously this is a solved problem elsewhere.

And so I hope that we can get off on a better foot this time. If the
answer is "No, the priority for this sort of thing is low, and we do not
consider it important for our customers."

Again, we don't have "customers". So your desire to implement a change in
behavior is dependant on:
1. Getting this list to agree on the specification;
2. Convincing an *individual* PostgreSQL developer or contributing company
that this issue is in their high priority interest to fix,
OR
Fixing it yourself and submitting the patch to PostgreSQL.org.

Then we will have to work
around it. Hopefully, at least, it will get put into a queue of future
enhancements.

Getting it on the TODO list is a good first step. However, that doesn't get
it implemented until it becomes some other developer's problem as well.

--
Josh Berkus
PostgreSQL @ Sun
San Francisco

"Dann Corbit" <DCorbit@connx.com> writes:

In the case of a SELECT query that selects a fixed constant of any sort,
it would be a definite improvement for PostgreSQL to give some sort of
upper maximum.

What's the point? You keep reminding us that your code is middleware
that can't assume anything much about the queries you're dealing with.
Therefore, I see no real value in fixing up one corner case. Your
argument about space allocation falls to the ground unless we can
provide a guaranteed, and usefully tight, upper bound on the column
width in *every* situation. If we cannot (which we can't), you're still
going to need those client-side "kluges".

In my opinion, variable-length data is a fact of life and you should
endeavor to make your code deal with it gracefully. There are bits of
the SQL spec that assume fixed-width data specifications are useful,
but to be blunt that's all a holdover from 1960s 80-column-punch-card
thinking. It's no way to design a modern application.

BTW, the reason I'm resistant to even thinking about this is that
Postgres is designed as an extensible system. Trying to do what you
want is not a matter of fixing literal constants and concatenation
and one or two other places --- it's a matter of imposing a new and
potentially hard-to-meet requirement on every datatype under the sun,
including a lot of user-written code that we don't control and would
break by adding such a requirement. So it's not even likely that we'd
think very hard about making this work, let alone actually do it.

regards, tom lane

As Dann pointed out we were one of the first companies to port Postgres
to windows many many years ago (7.1 days), and part of that porting work
is in the current postgresql product.

As I pointed out in a prior post, for the ODBC specification at least
(probably others), a maximum upper bound on returned data *must* be
reported and determined ahead of time when using binding ... A technique
where the client application allocates memory for the data and supplies
a pointer to that memory location for the driver.

Postgres unlike other databases shifts the burden of determining this
maximum size to the client and/or driver. Our company specializes in
access to wide variety of databases, both relational and non relational,
including SQL Server, Oracle, DB2, Sybase, Informix, etc. Postgres
sticks out as the only database that we have encountered with this
behavior .. Which is why we posted the original message.

Also as Dann pointed out even if this issue was addressed, it does not
help us because every existing installation of Postgres has the metadata
bug, so we *have* to bandaid it at the client/driver level anyway.

At least we have a record of the issue, so the next time a developer in
the community runs across the same oddity hopefully they will find this
and won't be scratching their heads like we were for a bit :)

Thanks

lm

-----Original Message-----
From: Tom Lane [mailto:tgl@sss.pgh.pa.us]
Sent: Tuesday, June 12, 2007 3:11 PM
To: Dann Corbit
Cc: pgsql-hackers@postgresql.org; Larry McGhaw
Subject: Re: [HACKERS] Selecting a constant question: A summary

"Dann Corbit" <DCorbit@connx.com> writes:

In the case of a SELECT query that selects a fixed constant of any
sort, it would be a definite improvement for PostgreSQL to give some
sort of upper maximum.

What's the point? You keep reminding us that your code is middleware
that can't assume anything much about the queries you're dealing with.
Therefore, I see no real value in fixing up one corner case. Your
argument about space allocation falls to the ground unless we can
provide a guaranteed, and usefully tight, upper bound on the column
width in *every* situation. If we cannot (which we can't), you're still
going to need those client-side "kluges".

In my opinion, variable-length data is a fact of life and you should
endeavor to make your code deal with it gracefully. There are bits of
the SQL spec that assume fixed-width data specifications are useful, but
to be blunt that's all a holdover from 1960s 80-column-punch-card
thinking. It's no way to design a modern application.

BTW, the reason I'm resistant to even thinking about this is that
Postgres is designed as an extensible system. Trying to do what you
want is not a matter of fixing literal constants and concatenation and
one or two other places --- it's a matter of imposing a new and
potentially hard-to-meet requirement on every datatype under the sun,
including a lot of user-written code that we don't control and would
break by adding such a requirement. So it's not even likely that we'd
think very hard about making this work, let alone actually do it.

regards, tom lane

Tom,

What's the point? You keep reminding us that your code is middleware
that can't assume anything much about the queries you're dealing with.
Therefore, I see no real value in fixing up one corner case. Your
argument about space allocation falls to the ground unless we can
provide a guaranteed, and usefully tight, upper bound on the column
width in *every* situation. If we cannot (which we can't), you're still
going to need those client-side "kluges".

Hmmm? I thought that Dann was just talking about constants, and not column
results. Am I confused?

BTW, the reason I'm resistant to even thinking about this is that
Postgres is designed as an extensible system. Trying to do what you
want is not a matter of fixing literal constants and concatenation
and one or two other places --- it's a matter of imposing a new and
potentially hard-to-meet requirement on every datatype under the sun,
including a lot of user-written code that we don't control and would
break by adding such a requirement. So it's not even likely that we'd
think very hard about making this work, let alone actually do it.

I'd think it would be possible to do this in an abstract way ... having a
"DisplayLength()" call for each data type and value. That would require
casting the constant, though, or computing all uncast constants as text.

--
Josh Berkus
PostgreSQL @ Sun
San Francisco

On 6/12/07, Josh Berkus <josh@agliodbs.com> wrote:

Tom,

What's the point? You keep reminding us that your code is middleware
that can't assume anything much about the queries you're dealing with.
Therefore, I see no real value in fixing up one corner case. Your
argument about space allocation falls to the ground unless we can
provide a guaranteed, and usefully tight, upper bound on the column
width in *every* situation. If we cannot (which we can't), you're still
going to need those client-side "kluges".

Hmmm? I thought that Dann was just talking about constants, and not column
results. Am I confused?

BTW, the reason I'm resistant to even thinking about this is that
Postgres is designed as an extensible system. Trying to do what you
want is not a matter of fixing literal constants and concatenation
and one or two other places --- it's a matter of imposing a new and
potentially hard-to-meet requirement on every datatype under the sun,
including a lot of user-written code that we don't control and would
break by adding such a requirement. So it's not even likely that we'd
think very hard about making this work, let alone actually do it.

I'd think it would be possible to do this in an abstract way ... having a
"DisplayLength()" call for each data type and value. That would require
casting the constant, though, or computing all uncast constants as text.

The simplest formulation of this problem appears to be that constant
strings that are uncast are treated as type unknown. The connx guys
seem to think that they should be implicitly cast to char(n) where n
is the length of the string. Is that a reasonable description, or are
you guys looking for something more general?

If you're just talking about the strings, then here are the thoughts
I've gleaned from the preceding thread.

- This makes possible some performance tweaks for drivers
- It achieves spec compliance (albeit for a stupid part of the spec)
- Implicit casting of unknown to char(n) or anything else seems rather
sketchy to me, but I can't see any specific objection, except that...
- I don't know when the right time to do the cast is. And doing it too
early seems obviously wrong.
- This only helps in corner case of string constants that are
1. not already cast and
2. not manipulated in any way
And that seems like a very small corner case with little or no
practical use. I guess if you have some code that turns query output
into some flavor of pretty-print, it'd make sense to have a constant
column as output of a CASE statement or something.
- The corner case must already be correctly handled by the general
case for arbitrary sized text, or alternatively phrased: there is no
way to conform to the standard while supporting arbitrary sized text.
Unless you're willing to pay the cost of scanning twice, or
maintaining "biggest entry" data for each variable length column.
- I don't know how much effort it would require to implement this, nor
how much complexity it would add to the code base. Clearly both of
these would be non-zero values.

Given the above, I agree with Tom: this seems like corner case where
the returns are marginal at best, compared to the cost to implement
and maintain.

Is there something I'm getting wrong in this summary?

Andrew

Josh Berkus <josh@agliodbs.com> writes:

What's the point? You keep reminding us that your code is middleware
that can't assume anything much about the queries you're dealing with.

Hmmm? I thought that Dann was just talking about constants, and not column
results. Am I confused?

Well, the specific example he was on about was a constant, but I don't
think it does him any good for us to fix just that one case. He'll
still have to deal with columns of indeterminate width in a whole lot of
other cases. If there were a reasonable path for us to report a useful
width bound in *every* case, then I could see spending time on it ...
but there's not.

BTW, it would certainly be trivial to extend libpq to report the actual
max width of a column within an already-retrieved PGresult. This isn't
anything the client code can't compute for itself, of course, but libpq
could get it in somewhat fewer cycles. However, I'm under the
impression that Dann wants the number at statement prepare time, and
we simply haven't got the information then.

I'd think it would be possible to do this in an abstract way ... having a
"DisplayLength()" call for each data type and value. That would require
casting the constant, though, or computing all uncast constants as text.

No, the point is about predicting the max width of a column of a query
result in advance of actually running the query. After you've got the
values in hand, it's not a very interesting problem. Before, well,
consider these examples:

select repeat(text_col, int_col) from my_table;
select repeat(text_col, int_col * random()) from my_table;
select repeat(text_col, some_user_defined_function(int_col)) from my_table;

The problem's really not soluble unless you want to dumb Postgres down
to approximately the capabilities of SQL89 -- no user-defined functions,
let alone user-defined types, plus pull out a whole lot of the built-in
functions that don't have readily predictable result widths.

regards, tom lane

"Andrew Hammond" <andrew.george.hammond@gmail.com> writes:

- Implicit casting of unknown to char(n) or anything else seems rather
sketchy to me, but I can't see any specific objection, except that...
- I don't know when the right time to do the cast is. And doing it too
early seems obviously wrong.

Well, I don't see any reason that we'd consider an implicit cast to
char(N) without context to drive us in that direction. The system is
currently biased to prefer casts to text. You could make a reasonable
case for forcing a cast to text if the constant's type is still
unresolved at the end of parsing, and indeed people have proposed that
off and on just so that clients would have one less type to think about.
In itself it doesn't do anything for Dann's problem though, because
unspecified width is unspecified width.

I've been thinking lately about trying harder to unify the text and
varchar types; I'm not sure about details yet, except that text should
be *exactly* the same thing as unconstrained-width varchar, rather than
almost the same except we claim it's a different type. The reason I'd
been thinking about this was mainly to get rid of the complexity and
runtime overhead that comes from having RelabelType nodes all over the
place when someone uses varchar instead of text. But if we did that,
we could also arrange that unknown literals coerce to varchar(N) with
N equal to their actual width, rather than coercing to text, and not
create any weird corner-case behaviors in the type system.

But at the end of the day this all would only solve Dann's problem for
the specific case of a SELECT with an undecorated literal constant in
its target list. He's still going to have to deal with unknown-width
columns in an enormous variety of cases, and so I completely fail to see
the point of changing the system's behavior for this one case.

regards, tom lane

Just a curiosity question: Why is the type of a literal '1' "unknown"
instead of varchar(1)?
Wouldn't varchar(1) cast properly to any use of the literal '1'?

What is the benefit of assuming it's an unknown?

"Chuck McDevitt" <cmcdevitt@greenplum.com> writes:

Just a curiosity question: Why is the type of a literal '1' "unknown"
instead of varchar(1)?

Because, for instance, it might be intended as an integer or float or
numeric value. Change the content a little, like '(1,2)' or '12:34',
and maybe it's a point or time value. There are plenty of contexts in
which the intended type of a literal is obviously not text/varchar.

We assign unknown initially as a way of flagging that the type
assignment is uncertain. Once we have a value that we think is varchar
(a table column for instance), the rules for deciding to cast it to a
different type get a lot more stringent.

regards, tom lane

I see... PostgreSQL wants to guess the datatype, given no clear
syntactic information, and perhaps a varchar(n) wouldn't be a valid cast
to some of the possible datatypes.

So, where x = '(1,2)' might be legal for comparing to x, but a field of
type varchar(5) might not be, as in where x = y, where y is type
varchar(5) containing '(1,2)'.

(Time values don't have this problem in pure ANSI SQL, since the literal
is TIME '12:34', but I can see for user types it might be ambiguous).

I find PostgreSQL's handling of this strange, as I come from systems
where 'xxx' is either a varchar or char type, in all contexts, and
implicit casts handle any needed conversions.
But now I understand why it does things this way.

Thanks.

-----Original Message-----
From: Tom Lane [mailto:tgl@sss.pgh.pa.us]
Sent: Tuesday, June 12, 2007 9:50 PM
To: Chuck McDevitt
Cc: Andrew Hammond; Josh Berkus; pgsql-hackers@postgresql.org; Dann
Corbit; Larry McGhaw
Subject: Re: [HACKERS] Selecting a constant question: A summary

"Chuck McDevitt" <cmcdevitt@greenplum.com> writes:

Just a curiosity question: Why is the type of a literal '1' "unknown"
instead of varchar(1)?

Because, for instance, it might be intended as an integer or float or
numeric value. Change the content a little, like '(1,2)' or '12:34',
and maybe it's a point or time value. There are plenty of contexts in
which the intended type of a literal is obviously not text/varchar.

We assign unknown initially as a way of flagging that the type
assignment is uncertain. Once we have a value that we think is varchar
(a table column for instance), the rules for deciding to cast it to a
different type get a lot more stringent.

regards, tom lane

On Wed, Jun 13, 2007 at 02:12:37AM -0400, Chuck McDevitt wrote:

So, where x = '(1,2)' might be legal for comparing to x, but a field of
type varchar(5) might not be, as in where x = y, where y is type
varchar(5) containing '(1,2)'.

Normally, just about every type can be converted to or from text. So if
postgres converted to varchar first you have problems with the
statement x='const' where x is type "foo". It's now ambiguous since you
either convert x to varchar or the constant to "foo". Instead, postgres
marks the constant as unknown and now it always gets converted because
nothing can convert to unknown.

Thus far this system has worked excellently, though not perfectly
obviously. It matches people's expectations well, which is the most
important part.

I find PostgreSQL's handling of this strange, as I come from systems
where 'xxx' is either a varchar or char type, in all contexts, and
implicit casts handle any needed conversions.
But now I understand why it does things this way.

User-defined types makes for lots of interesting choices, but they are
by far the most powerful feature of postgres and we don't want to mess
with that.

Have a nice day,
--
Martijn van Oosterhout <kleptog@svana.org> http://svana.org/kleptog/

"Chuck McDevitt" <cmcdevitt@greenplum.com> writes:

Just a curiosity question: Why is the type of a literal '1' "unknown"
instead of varchar(1)?

Even if it was assigned a text datatype it would be the unconstrainted "text"
not varchar(1). If we used varchar(1) then things like:

create table foo as select 'foo';

would create a table with varchar(3) which would then complain if you tried to
insert 'foobar' into. There doesn't seem to be enough evidence that the user
intended to constrain the input to just 'foo' in that case.

Of course right now you get a table with a column of type "unknown" which is
very unlikely to be what the user expects.

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com

For some Unicode character sets, element_width can be as much as 4

In UTF8 one char can be up to 6 bytes, so 4 is not correct in general.

Andreas