Gel sqla #536
Gel sqla #536
Conversation
vpetrovykh
force-pushed
the
gel-sqla
branch
2 times, most recently
from
0885a10 to
fd5217c
Compare
|
Could you post some generated code? |
vpetrovykh
force-pushed
the
gel-sqla
branch
3 times, most recently
from
8c0f82e to
aa57a94
Compare
vpetrovykh
force-pushed
the
gel-sqla
branch
3 times, most recently
from
6aa78ae to
61c9c75
Compare
Create separate intermediate objects to represent links with link properties (because ORMs tend to view them as such). Don't allow crazy names (not usual type of identifiers). Multi properties behave more like plain multi links than anything else (because they have a separate table with the property value in it), so they should be reflected like that establishing a relationship. Reflect Gel modules into Python modules. Add tests that setup a Gel database and then generate the SQLAlchemy models from it. The individual tests use a SQLAlchemy session to access the database using postgres protocol.
| for bklinks in bk.values(): | ||
| if len(bklinks) > 1: | ||
| # We have a collision, so each backlink in it must now be | ||
| # disambiguated. | ||
| for link in bklinks: | ||
| origsrc = get_sql_name(link['target']['name']) | ||
| lname = link['name'] | ||
| link['name'] = f'{lname}_from_{origsrc}' | ||
| # Also update the original source of the link with the | ||
| # special backlink name. | ||
| source = type_map[link['target']['name']] | ||
| fwname = lname.replace('backlink_via_', '', 1) | ||
| link['fwname'] = fwname | ||
| source['backlink_renames'][fwname] = link['name'] |
There was a problem hiding this comment.
Could we always generate the disambiguated backlink, and then only generate the short name when it is unambiguous, or does that not work?
There was a problem hiding this comment.
We mostly expect not too many collisions so I tend to treat that as a the default. Backlink collisions appear when everything is linked to everything, which is not necessarily a common schema since that creates a lot of links to maintain.
Also, detecting unambiguous backlinks is a little more annoying that the other way around. Collisions are shown by identical names, but unambiguous long names would have to be checked by looking up their corresponding forward link name, seems more roundabout.
| # have psycopg2 installed, as long as we run this in the test | ||
| # environments that have this, it is fine since we're not expecting | ||
| # different functionality based on flavours of psycopg2. | ||
| if importlib.util.find_spec("psycopg2") is None: |
There was a problem hiding this comment.
I think it would be cleaner to do
try:
import psychopg2
except ImportError:
# handle it
There was a problem hiding this comment.
I wanted to avoid the import since I don't even need anything from psycopg2 directly, not using it anywhere myself. Didn't want to leave the impression that I needed the import myself and if it's not used, maybe it's OK to "remove this dependency check" making it look like an outdated comment or dead code. importlib seemed like a more overt and unambiguous check.
Run `gel-orm --help` for more details.
More tests coming...