Optimizes the filterValidXMLChars function by changing the loop variable type from int to uint to eliminate bound checks during compilation, resulting in a ~4% performance improvement.
- Changes loop variable i from int to uint to remove compiler-generated bound checks
- Adjusts type conversions accordingly to maintain correctness
```
goos: linux
goarch: arm64
pkg: miniflux.app/v2/internal/reader/parser
│ old.txt │ new.txt │
│ sec/op │ sec/op vs base │
Parse-8 40.91m ± 3% 39.30m ± 2% -3.94% (p=0.000 n=50)
```
- Implement a better/simpler polyfill for web browsers that don't supported
trusted types yet
- Use two separate policies: one to create HTML, another to create/use script
urls
- Instead of having the policy live in the top-level scope, they're now
declared at the lowest possible scope, right before they're used, making them
inaccessible outside of it. This puts their usage completely out of reach of
an attacker unable to gain some control outside of those two (small) scopes,
and thus removes the need to tighten the policies.
- Remove the now-unused tt.js file
This has been tested on Firefox (doesn't support trusted types) and on Chromium
(does support trusted types).
Replaces usage of the word "bookmark" with "star"/"starred" in order to be more
consistent with the UI and database models, and to reduce confusion with
"bookmarklet" and integration features.
This is in preparation of future work on read-it-later features.
Which are also not called "bookmarks" to prevent any further confusion.
https://github.com/orgs/miniflux/discussions/3719
Related-to: https://github.com/miniflux/v2/pull/2219
In general, duration is used as time unit representation.
At some places when int is returned, there's no documentation which unit is used.
So just convert to time.Duration ASAP.
- b.body can never be of type error, so let's remove it from the switch-case
construct.
- there is no need to use defer when the only return statement is two lines
after.
Instead of blindly compiling all the common/ templates for every view/ ones,
let's be explicit about the dependencies. This should significantly decrease
the resident memory consumption, as ParseTemplate is responsible for ~10M of
the current 11M of heap memory on my instance, so any win there is interesting.
This will also allow better factorization of templates, now that everything is
explicit. Another side-effect is that it'll make testing easier, as we now have
a comprehensive list of views/ templates affected by a change in a file in
common/
Some Miniflux clients expect a specific version format.
For example, Flux News converts the string version to an integer.
Using `Development Version` will break some clients.
The length of a tsvector (lexemes + positions) must be less than 1 megabyte.
We don't need to index the entire content, and we need to keep a buffer for the positions.
- Add `content IS NOT NULL` to avoid clearing the same entries over and over
- Create `DeleteRemovedEntriesEnclosures` and `ClearRemovedEntriesContent` that report individual counts
