Re: [Planner Dev] MySQL database access support

[Jon Wilkins <jon wilkins winskil co uk>]

Richard Hult wrote:

> Hi,
>
> I would strongly recommend against this kind of solution, it is a
> nightmare to maintain. What I would do (and what the plan has been for a
> while but nobody has implemented it) is to rewrite the current database
> backend to not use any psql specific code.
>  
Since posting, I've noticed that the sql write code is actually far less 
pgsql specific than the sql read code -
so less re-writing would be needed - I think.

> > There was also a suggestion in the list that we should standardise all
> > database access to SQL95 -- essentially using the "lowest common
> > denominator" - but the postgresql code using cursors works so why change
> > it?
> >    
>
> The current code is kind of ugly and buggy so it needs to be cleaned up or
> rewritten anyway. And by adding mysql suppport we only get psql and mysql.
> By using a common subset we get potentially support for a lot of other
> providers through libgda.
>  
What I forgot to put in the previous post was that the mysql queries are 
as generic as possible -- the data format
conversion is the only "mysql" specific element (for expediancy at the 
moment and wasn't intended to stay). 
The idea was that the "mysql" code would be standard enough to connect 
to any database source that gda
suppored.  The db selection code in planner would still need to pass the 
database type to libplanner so that libplanner
can select the correct gda "provider".  (The db selection dialog may be 
able to dynamically build a list of installed
providers but I've not been looking at the gui side - perhaps the gui 
can code the provider string into the uri somehow,
so libplanner simply "wont know" but will just try to connect to the 
specified provider?).

If you're happy to lose the postgres-specific cursor queries then that 
makes it a lot simpler and definatley much more
maintainable & readable.  My approach was based on leaving alone the 
code that already worked (with pgsql).

I'll take the pgsql cursor queries out and this removes the need for the 
ugly branches in mrp-sql.c, but for the moment
I dont see a way round needing the gui to somehow pass the database type 
to libplanner so that the correct gda
provider is used.  I'll also check the bugzilla and see if there are db 
specific bugs that I can look at at the same time.

jon