I'm not very fond of the commit message, especially the fact it describes what is not done, but not really what is done, and I find Add deposit info to objects added to swh-storage from metadata-only deposits unclear. What (type of) objects are we talking of? How is the deposit info it "added" to them?

bump dep on swh.model + add a comment (as requested)

Make sure extid_get_from_target() is immune to murmur3 collisions

get rid of the OBJECT_TYPE_MAP_SHORT map, as requested

Add missing support for extid in the replayer

In D4807#132570, @olasd wrote:

I wonder if the TEST_OBJECTS addition will break the tests of swh.storage until D4698 lands.

Remove extid_by_extid in cassandre

Apply vlorentz and olasd's suggestions

forgot sql migration script

ExtID.target is now a CoreSWHID

Use CoreSWHID as target

All tests should now pass (with D4807)

Rebase, make ExtID a HashableObject, and remove hypothesis strategies for ExtID

Added test being:

rebase, rework to use new SWHID classes/model

superseeded by new SWHID classes

adapted to nes SWHID classes

again

blank, to fire jenkins

In D5172#131088, @ardumont wrote:

looks like it's doing the job ;)

I was wondering if it would not be clearer to alter the schema at the end of the file
(or in a subsequent file). Kinda what's done for indexes.

\if :dbflavor_with_path
alter table content_early_in_rev
add column loc bigint not null;  -- location of the content relative to the revision root directory

comment on column content_early_in_rev.loc is 'Location of content in revision';
...
\end

But that would split the reading so let's keep it that way
¯\_(ツ)_/¯

rebase

rebase

rebase

rebase

Rebase + better commit message + add a docstring

rebased

rebased

rebase + default to with_path=True as asked by @aeviso

rebase

The swh_storage_with_objects fixtures returns a (postgresal-based) storage with some objects in it (only a few of each type). The test that given storage do indeed have at least a content, a directory and a revision. I'll improve the commit message a bit.

In D5072#128933, @vlorentz wrote:

I'm not convinced by the flavor thing. Why not a single schema, but with empty values in loc when it's disabled?

(as mentioned on IRC): Since we are doing big works, I see no reason to keep the 'qualifiers' a schema-less dict-like attribute. I'd rather have them as attributes.

rebase

rebase

rebase

ad forgotten isorts (after proper config in pyproject.toml)

rebase and fix db initialization in tests

rebase and fix tests

add missing isort config in pyproject.toml