* [2025-01-17 Fri 06:58 +0100] Ihor Radchenko <yantar92@posteo.net>
> Phil Estival <pe@7d.nz> writes:
> 
>>> Is there any specific reason why you are seemingly re-implementing what
>>> `sql-product-interactive' does? May we re-use it instead?
>>
>> Yes. When opening a new connection `sql-product-interactive' will
>> systematically ask for input in the mini-buffer prompt and fill the
>> required information, suggesting inputs from the ones previously given.
> 
> What about changing sql.el to provide the necessary flexibility?
> I'd prefer it better than rewriting parts of sql.el in Org mode.

Gladly. Here is a proposal for patch for `sql-product-interactive' in
sql.el. A specific case for sqlite is handled and this is not 100%
satisfying. I have the feeling this should rather be where a
conversion of nil params to empty string happens, but sql.el is
rather long and the verifications required are numerous... There may
also be other db that allow connection without one these parameters...

The previous configuration from this series of patch, with a variant
of `sql-product-interactive', would allow a nil database for a session
of sqlite, now it needs to mention an empty string for :database or to
set it as a default value in `sql-sqlite-login-params'.

Next is the patch for ob-sql that rely on the modification of
`sql-product-interactive' to keep session connection smooth.  This
substract an addition from the previous series of patch but I guess
the proper way to do will be to submit the diff from main after
these reconsiderations.

A side note: there are contradictory comments in sql.el.
- `sql-product-interactive'
      Do not call this function by yourself.  The environment must be
      initialized by an entry function specific for the SQL interpreter.
      See `sql-help' for a list of available entry functions.
- `sql-help'
      You can also use M-x sql-product-interactive to invoke the
      interpreter for the current ‘sql-product’.
-- 
Phil Estival