On Tue, 13 Nov 2018 09:10:30 +0100
Clément Lassieur <clement@lassieur.org> wrote:

> Hi Björn,
> 
> Björn Höfling <bjoern.hoefling@bjoernhoefling.de> writes:
> 
> > We are directly relying on the rowid here, there is no explicit
> > id-column.
> >
> > This could lead to unpredicted results and reorderings (6th Quirk in
> > document):
> >
> > https://www.sqlite.org/rowidtable.html
> >
> > We should add a column:
> >
> > id INTEGER PRIMARY KEY AUTOINCREMENT
> >
> > Problem is that this concept of AUTOINCREMENT does only work for
> > Primary Keys in Sqlite. So we need to degrade "derivation" to a
> > secondary key, i.e. make it non-null and unique:
> >
> > derivation    TEXT NOT NULL UNIQUE,
> >
> > Is there anything speaking against that?  
> 
> We only use that rowid to display a number at the left of every
> 'build' row.  I think it would make more sense to use the derivation
> name where we currently use the rowid.  It would also be more
> understandable for the users.
> 
> We don't even need rowid for sorting because we can sort with the
> timestamps.
> 
> The only issue is that we get further from hydra, but we are already
> pretty far away anyway.

I'm undecided about this, I don't understand enough of it yet. In some
sense I would prefer a numeric id to refer to. On the other hand the
derivation is the actual key, and if that is generated a second time,
the build is the same and will not be executed again.

Björn