SQLiteDialect: Correct GUID to string conversion when BinaryGuid=False (regression) (fixes #2110)

[oskarb]

Attached are test cases for #2110 (and #2109), and a proposed fix for #2110.

The suggested fix teaches SQLiteDialect to detect the effective value of the BinaryGuid from the connection string and apply different expressions for the strguid function based on that. This fixes the first failure mentioned below.

In NH 5.2, without the fix, the result of the test cases are:

GH2110.ByCodeFixture(False).DbExplicitIdGuidToStringIsSameAsDotNet
Fails, because the expression given in the dialect for strguid is not valid with BinaryGuid=False. This is the regression caused by #1151. Fixed by this PR.

GH2110.ByCodeFixture(False).DbImplicitIdGuidToStringIsSameAsDotNet
Success, but only because NHibernate forgets to apply the strguid expression, opting for a straightforward cast(entity0_.Id as char) instead.

GH2110.ByCodeFixture(True).DbExplicitIdGuidToStringIsSameAsDotNet
Success. This is the case that was fixed by #1151.

GH2110.ByCodeFixture(True).DbImplicitIdGuidToStringIsSameAsDotNet
Fails (still), since the fix for #1151 is not applied here because NH doesn't apply the strguid method. Reported as #2109. Since it still fails, this case is ignored for now.

[hazzik]

Ok. This will go to 5.2.6.

[oskarb]

Rebased, and PR retargeted.

I also figured it would be better to expand the existing test cases for the original issue NH-3426 instead of effectively duplicating.

[oskarb]

It seems the test cases SelectGuidToStringImplicit and WhereGuidToStringImplicit that I added to showcase GH-2109 (affecting SQLite with BinaryGuid=true) also expose same problem on Firebird, Mysql and Oracle (any dialect where converting guid to string cannot be done using a regular cast in SQL).

[oskarb]

I've pushed an updated version with the following changes:

• Also look at named connection string.

• Introduce a dialect specific setting that can be used in cases where the automatic detection from connection string won't work.

• Disabled the failing test cases for Wrong GUID to string conversion when not explicitly referencing the id property #2109 for more dialects that suffer from that problem.

[hazzik]

Actually, I got into thinking... this applies not only to SQLite, but also to MySQL. The MySQL provider from Oracle has OldGuids=true/false connection string option. This kinda minor, so I would not bother with it now.

[oskarb]

This is intended to fix a regression I discovered, related to sqlite. Though I suppose there would be regressions for the other database where strguid was introduced/changed at the same time too.

I discovered this while working with unit tests that create the database on each run. Thinking about it, it's actually quite likely that I added BinaryGuid=false just to work around the problem with string conversions that has now been fixed in NH so maybe I don't need it anymore.

However, if someone have used BinaryGuid=false on a file-backed database that they can't easily just discard and recreate, they will be in more trouble.

[fredericDelaporte]

Looking into it again, I think this patch should be merged in its current state.

The MySql case is not so similar, because it is somewhat already handled by having two dialects: MySql5Dialect, working with binary(16) guids (so expecting OldGuids set as true), and MySql55Dialect, working with char(36) guids (so expecting OldGuids unset or false). And adapting the MySql dialect behavior with guids according to the OldGuids option would also require to override the guid type registration. So it seems better to handle that in another PR, not necessarily on 5.2.x.

[fredericDelaporte]

I just realize now the nhibernate-configuration.xsd has not been updated, so the new setting is anyway not usable in config file, because it will fail the validation...