Why doesn't explicit COLLATE override database collation?

I am on SQL Server 2008 R2 dev, server default collation is Cyrillic_General_CI_AS

Executing in SSMS

SELECT 'éÉâÂàÀëËçæà' COLLATE Latin1_General_CS_AS

or

SELECT 'éÉâÂàÀëËçæà' COLL开发者_Go百科ATE Latin1_General_CI_AI   

outputs

• eEaAaAeEc?a on(in ocntext of/use dbName) of database with default collation Cyrillic_General_CI_AS
• éÉâÂàÀëËçæà on database with default collation Latin1_General_CI_AS

Why?

Those character literals in your queries are first converted to varchar strings under whatever collation the database is set for, and then your collation cast takes effect.

If you want to pass such character literals and ensure all characters are faithfully represented, it's better to pass them as nvarchar literals:

create database CollTest collate Cyrillic_General_CI_AS
go
use CollTest
go
SELECT 'éÉâÂàÀëËçæà' COLLATE Latin1_General_CS_AS   
SELECT 'éÉâÂàÀëËçæà' COLLATE Latin1_General_CI_AI
go
SELECT N'éÉâÂàÀëËçæà' COLLATE Latin1_General_CS_AS   
SELECT N'éÉâÂàÀëËçæà' COLLATE Latin1_General_CI_AI
go

Output:

-----------
eEaAaAeEc?a

(1 row(s) affected)


-----------
eEaAaAeEc?a

(1 row(s) affected)


-----------
éÉâÂàÀëËçæà

(1 row(s) affected)


-----------
éÉâÂàÀëËçæà

(1 row(s) affected)

According to this character table Cyrillic_General_CI_AS doesn't contain the æ chacter. This would explain why you see a ? in it's place if you are truly using the defaults instead of what you have in your select statements.