We run Dynamics GP, and anybody that's dealt with GP will be familiar with all of its "magic number" columns representing the various enum values inside the application. For example, a reporting query might look something like this:
...
WHERE sod.SOPTYPE = 2
AND iv.VCTNMTHD = 3
AND pod.POLNESTA IN (2, 3)
...which is about as far from self-documenting as possible.
At first I had the bright idea of creating a table called 'Enums', filling it with values, and making a scalar wrapper function so you could query like this:
...
WHERE sod.SOPTYPE = Enum('Sales Doc Type', 'Order')
AND iv.VCTNMTHD = Enum('Valuation Method', 'Average Perpetual')
AND pod.POLNESTA IN (Enum('PO Line State', 'Released'), Enum('PO Line State', 'Change Order'))
Of course, that sucked, because the optimizer couldn't see the underlying 'constant' value, and thus made some rather poor choices since it had to make cardinality guesses.
So it needs to be an actual constant value in the code, but I'm not sure if there are any nice features built into SSMS that would make this sort of thing easy. Could I (mis)use code snippets or templates in some way? If we could do some kind of lookup/insertion to make our code look something like this, that would be awesome:
...
WHERE sod.SOPTYPE = /*Order*/2
AND iv.VCTNMTHD = /*Average Perpetual*/3
AND pod.POLNESTA IN (/*Released*/2, /*Change Order*/3)
I'm also not opposed to checking out SSMS add-ins.
2 Answers 2
I would actually recommend against tying this lookup in any way to SSMS / SQLCMD -specific features. Doing so would require that the code could only ever be executed using just those two programs. You wouldn't even be able to put any such logic into a stored procedure, even if it was executed from SSMS or SQLCMD. You also can't use SQLCMD commands in SSRS, etc.
Some options (that are completely independent of the client tool used to execute them):
If SQL Server 2019 is an option, then it's possible that the new Scalar UDF Inlining feature might resolve the issue with your
Enum('Sales Doc Type', 'Order')
approach.
It's possible that an inline TVF might do the trick. You just might need to restructure those predicates in the
WHERE
clause to beINNER JOIN
s:FROM SOPxxxxx sod INNER JOIN dbo.Enum('Sales Doc Type', 'Order') sdt ON sod.SOPTYPE = sdt.Value
Of course, the
IN
list is a little trickier as I think that might require aLEFT JOIN
on each option and then aWHERE
predicate to ensure that both of those did not result in aNULL
:FROM dbo.POP10110 pod LEFT JOIN dbo.Enum('PO Line State', 'Released') pls_r ON pod.POLNESTA = pls_r.Value LEFT JOIN dbo.Enum('PO Line State', 'Change Order') pls_co ON pod.POLNESTA = pls_co.Value WHERE ( pls_r.Value IS NOT NULL OR pls_co.Value IS NOT NULL)
OR, using this same iTVF, you might be able to do non-correlated subqueries that might be more readable / manageable:
FROM SOPxxxxx sod WHERE sod.SOPTYPE = (SELECT sdt.Value FROM dbo.Enum('Sales Doc Type', 'Order') sdt)
And the
IN
list is at least looks better:FROM dbo.POP10110 pod WHERE pod.POLNESTA IN ( SELECT pls_r.Value FROM dbo.Enum('PO Line State', 'Released') pls_r UNION ALL SELECT pls_co.Value FROM dbo.Enum('PO Line State', 'Change Order') pls_co )
Or perhaps (not sure what the query optimizer will prefer):
FROM dbo.POP10110 pod WHERE pod.POLNESTA IN ( (SELECT pls_r.Value FROM dbo.Enum('PO Line State', 'Released') pls_r), (SELECT pls_co.Value FROM dbo.Enum('PO Line State', 'Change Order') pls_co) )
If SQLCLR is an option, then SQLCLR scalar UDFs that do not do any data access and are marked as
[SqlFunction(IsDeterministic = true)]
are constant folded into the execution plan (the "Query Processing Architecture Guide" documentation currently states that SQLCLR functions cannot be folded, but that is provably wrong, so I suspect that the documentation was merely not updated when SQL Server 2012 added that ability; I have submitted a documentation update to fix that: Fix "Constant Folding" info in "Query Processing Architecture Guide").The trick here is to read the data from the table without doing so in the SQLCLR scalar UDF so as to not prevent it from being foldable. In order to accomplish this, do the following:
- Create a static readonly dictionary at the class level.
- Create a static constructor for the class that uses a regular, external
SqlConnection
to connect to the local instance. This will select from the table and populate the static dictionary. This constructor will ensure that the dictionary is always populated, even if the assembly gets unloaded due to memory pressure,DBCC FREESYSTEMCACHE
, etc. Unfortunately, the internal "context" connection cannot be used here since there is no Sql Context when the static constructor is executed (but would be great if there was!). - Mark the assembly as
WITH PERMISSION_SET = EXTERNAL_ACCESS
. No need to specifyUNSAFE
because the dictionary is "readonly". And even though it is "readonly", it can still have items added and/or removed from it. This should not pose any problems as it's a single use collection that is always the same for all users. - Decorate the static method that is exposed to T-SQL to do the lookup with:
No need to specify[SqlFunction(IsDeterministic = true)]
DataAccess
orSystemDataAccess
properties as those are set toNone
by default.
This method will allow you to implement the
Enum()
approach that you had started doing. And, since you are working with a 3rd party app, you can keep things "clean" by installing the assembly into a separate database (and same would go for report procs, etc, I suppose). I have tested and verified that it still does the constant folding.
OR (this is really option 3b): Given that this is a 3rd party app, IF it is safe to assume that the lookup / enum values won't be changing much, if ever, then you don't really need to read them from a table. You can simple have the entire dictionary hard-coded in the .NET code. In this approach, you could then keep the assembly marked as
PERMISSION_SET = SAFE
as there would be no need for that externalSqlConnection
back to the current instance.
Within SSMS you can enable SQLCMD Mode and do things along the lines of the following:
-- Enable SQLCMD Mode (ALT+Q+M)
:SetVar cSomeNumber 42
SELECT $(cSomeNumber)
If you save the scripts to a file, these will operate the same when when you use with... wait for it... SQLCMD.exe
The ":SetVar" and "$(cSomeNumber)" are processed before the sql is sent to the engine, however. This means that this cannot be used in a different client (e.g. ADO.Net) and if you create any DBOs it will record the interperted value. So this:
-- Enable SQL Command Mode (ALT+Q+M)
:SetVar cSomeNumber 42
CREATE View Test AS SELECT $(cSomeNumber) as SomeNumber
will yield this:
-- Enable SQL Command Mode (ALT+Q+M)
CREATE View [dbo].[Test] AS SELECT 42 as SomeNumber
Which limits the over all utility.
-
Do I understand correctly that your idea of using SQLCMD variables to document those obscure constant values also involves choosing a more or less self-descriptive name for each constant (like e.g.
:SetVar sdtOrder 2
or:SetVar polsChangeOrder 3
, using the example given in the question)? Since the variables are assigned explicitly in the script, you are basically suggesting using them as named constants, correct?Andriy M– Andriy M2019年11月25日 09:14:13 +00:00Commented Nov 25, 2019 at 9:14 -
Yes, if you are saving the associated queries into script files (.sql). You could even extend the variable to include a comment: :SETVAR cTowels "42 /* Towels */"Graham– Graham2019年11月25日 17:09:35 +00:00Commented Nov 25, 2019 at 17:09
/*Inline Comments*/
into the code would help maintainability, and if I can also get automatic replacement when typing something likeValuationMethod:AveragePerpetual
, then that would be totally sweet.