* basic changes
update version check
set ef_search for clip
* pgvector compatibility
Revert "pgvector compatibility"
This reverts commit 2b66a52aa4097dd27da58138c5288fd87cb9b24a.
pgvector compatibility: minimal edition
pgvector startup check
* update extension at startup
* wording
shortened vector extension variable name
* nightly docker
* fixed version checks
* update tests
add tests for updating extension
remove unnecessary check
* simplify `getRuntimeConfig`
* wording
* reindex on minor version update
* 0.2 upgrade testing
update prod compose
* acquire lock for init
* wip vector down on shutdown
* use upgrade helper
* update image tag
* refine restart check
check error message
* test reindex
testing
upstream fix
formatting
fixed reindexing
* use enum in signature
* fix tests
remove unused code
* add reindexing tests
* update to official 0.2
remove alpha from version name
* add warning test if restart required
* update test image to 0.2.0
* linting and test cleanup
* formatting
* update sql
* wording
* handle setting search path for new and existing databases
* handle new db in reindex check
* fix post-update reindexing
* get dim size
* formatting
* use vbase
* handle different db name
* update sql
* linting
* fix suggested env
* feat: search peoples
* fix: responsive design
* use existing count
* generate sql file
* fix: tests
* remove visible people
* fix: merge, hide...
* use component
* fix: linter
* chore: regenerate api
* fix: change name when searching for a face
* save search
* remove duplicate
* use enums for query parameters
* fix: increase to 20 for the local search
* use constants
* simplify
* fix: number of people more visible
* fix: merge
* fix: search
* fix: loading spinner position
* pr feedback
Fixes https://github.com/immich-app/immich/issues/5982.
There are basically three options:
1. Search `originalFileName` by dropping a file extension from the query
(if present). Lower fidelity but very easy - just a standard index &
equality.
2. Search `originalPath` by adding an index on `reverse(originalPath)`
and using `starts_with(reverse(query) + "/", reverse(originalPath)`. A
weird index & query but high fidelity.
3. Add a new generated column called `originalFileNameWithExtension` or
something. More storage, kinda jank.
TBH, I think (1) is good enough and easy to make better in the future.
For example, if I search "DSC_4242.jpg", I don't really think it matters
if "DSC_4242.mov" also shows up.
edit: There's a fourth approach that we discussed a bit in Discord and
decided we could switch to it in the future: using a GIN. The minor
issue is that Postgres doesn't tokenize paths in a useful (they're a
single token and it won't match against partial components). We can
solve that by tokenizing it ourselves. For example:
```
immich=# with vecs as (select to_tsvector('simple', array_to_string(string_to_array('upload/library/sushain/2015/2015-08-09/IMG_275.JPG', '/'), ' ')) as vec) select * from vecs where vec @@ phraseto_tsquery('simple', array_to_string(string_to_array('library/sushain', '/'), ' '));
vec
-------------------------------------------------------------------------------
'-08':6 '-09':7 '2015':4,5 'img_275.jpg':8 'library':2 'sushain':3 'upload':1
(1 row)
```
The query is also tokenized with the 'split-by-slash-join-with-space'
strategy. This strategy results in `IMG_275.JPG`, `2015`, `sushain` and
`library/sushain` matching. But, `08` and `IMG_275` do not match. The
former is because the token is `-08` and the latter because the
`img_275.jpg` token is matched against exactly.
