Add insensitive search for documents #545
Conversation
CkuT
changed the title
|
I think the search by tag or correspondant is now broken if there are any accents. The PR is still in WIP. |
There was a problem hiding this comment.
It's a good addition, but there were a few things that I think need to be fixed before we accept the merge. I'm sorry that I'm not as readily available to this project as I once was, so if you do make the requested changes and I don't properly respond in a timely manner to toggle approve, feel free to email me directly: paperless at <my-github-username> dot org.
src/documents/admin.py
Outdated
| remove_tag_from_selected, | ||
| set_correspondent_on_selected | ||
| ) | ||
| from documents.actions import (add_tag_to_selected, |
There was a problem hiding this comment.
In the interests of consistency, please don't reformat imports. If anything, imports should conform to an isort configuration of:
isort -m 3 --dont-skip __init__.py --virtual-env ${virtualenv}/.. --quiet ${file_name}|
|
||
| from django.db import migrations, models | ||
|
|
||
| from paperless.utils import slugify as slugifyOCR |
There was a problem hiding this comment.
While it can be tempting to import stuff from modules into a migration to keep your code DRY, this will later bite us in the ass should we decide to rename/remove this function. It will break all migrations for all time as a result.
Instead, if you have logic you wish to make available in a migration, copy it verbatim into the migration.
src/paperless/utils.py
Outdated
| import unicodedata | ||
|
|
||
|
|
||
| def slugify(content): |
There was a problem hiding this comment.
Can we come up with a more appropriate name for this than slugify()? Something like make_searchable() would be more apt as this doesn't turn the content into a slug at all.
| doc.searchable_content = slugifyOCR(doc.content) | ||
| doc.save() | ||
|
|
||
| def casefold_backwards(apps, schema_editor): |
There was a problem hiding this comment.
Protip: you don't need to create an empty backwards method. You can just do migrations.RunPython(casefold_forwards, migrations.RunPython.noop) below instead.
| @@ -21,3 +21,28 @@ def test_file_deletion(self): | |||
| mock_unlink.assert_any_call(file_path) | |||
| mock_unlink.assert_any_call(thumb_path) | |||
| self.assertEqual(mock_unlink.call_count, 2) | |||
|
|
|||
| @@ -266,6 +281,13 @@ def __str__(self): | |||
| return "{}: {}".format(created, self.correspondent or self.title) | |||
| return str(created) | |||
|
|
|||
| def save(self, *args, **kwargs): | |||
There was a problem hiding this comment.
Thank you for doing this in .save() and not in a signal. This is much more appropriate.
CkuT
force-pushed
the
insensitive-search
branch
3 times, most recently
from
de68565
to
51b0dc0
Compare
CkuT
changed the title
CkuT
force-pushed
the
insensitive-search
branch
from
51b0dc0
to
4ac538b
Compare
CkuT
force-pushed
the
insensitive-search
branch
from
4ac538b
to
77b0f65
Compare
|
Hey ! Does someone have any time to review this ? :) |
| def make_searchable(content): | ||
| return ( | ||
| unicodedata.normalize("NFKD", content.casefold()) | ||
| .encode("ASCII", "ignore") |
There was a problem hiding this comment.
Hm, I'm not sure if this is the right thing to do. In light of your change only searching the new searchable fields, this straight up breaks search for languages that share no characters with the ASCII codepage, right? Judging from the official Python docs, "ignore" would simply drop non-ASCII characters, resulting in a potentially empty search-term.
As far as I can tell not ignoring these characters here would break the entire feature, your Zürich Weiß example would, after case-folding first, produce zu\xcc\x88rich weiss, which obviously cannot be searched for with zurich weiss.
Maybe the simplest fix is to search both the new ASCII-fied fields and the regular fields?
This PR adds an insensitive-search on document title and content. This should close #115.
As it was pointed out by @danielquinn,
str.casefold()is a good start, but not enough: it does not remove accents. For the latter, we need aNFKDunicode normalization (see https://unicode.org/reports/tr15/).I'm open to any improvements !