Enable production source maps for index.js, fix CSS sourcemaps (#27291) (#27295)
Backport #27291 by @silverwind
Previously, the production build never output sourcemaps. Now we emit
one file for `index.js` because it is the most likely one where we need
to be able to better debug reported issues like
https://github.com/go-gitea/gitea/issues/27213. This will currently
increase the binary size of gitea by around 700kB which is what the
gzipped source map file has.
Also, I fixed the CSS sourcemap generation which was broken since the
introduction of lightningcss.
The chinese docs are machine-translated, please correct accordingly.
Co-authored-by: silverwind <me@silverwind.io>
Now that we have the `/assets` directory, we can put`licenses.txt`
directly into it instead of incorrect `/js` path which was previously
only done to avoid reserving a username.
---------
Co-authored-by: Giteabot <teabot@gitea.io>
Use case-insensitive regex for all webpack assets (#26867)
Previously, only some of these regex had the `i` flag and while we can
likely ensure case for our files, these regexes are also used for
third-party files, so it's better to always match insensitively.
Fixes https://github.com/go-gitea/gitea/issues/26439.
The minification result is not ideal with esbuild, but it's better than
failing competely.
Co-authored-by: Giteabot <teabot@gitea.io>
Move public asset files to the proper directory (#25907)
Move `public/*` to `public/assets/*`
Some old PRs (like #15219) introduced inconsistent directory system.
For example: why the local directory "public" is accessed by
`http://site/assets`? How to serve the ".well-known" files properly in
the public directory?
For convention rules, the "public" directory is widely used for the
website's root directory. It shouldn't be an exception for Gitea.
So, this PR makes the things consistent:
* `http://site/assets/foo` means `{CustomPath}/public/assets/foo`.
* `{CustomPath}/public/.well-known` and `{CustomPath}/public/robots.txt`
can be used in the future.
This PR is also a prerequisite for a clear solution for:
* #21942
* #25892
* discourse.gitea.io: [.well-known path serving custom files behind
proxy?](https://discourse.gitea.io/t/well-known-path-serving-custom-files-behind-proxy/5445/1)
This PR is breaking for users who have custom "public" files (CSS/JS).
After getting approvals, I will update the documents.
----
## ⚠️ BREAKING ⚠️
If you have files in your "custom/public/" folder, please move them to
"custom/public/assets/".
---------
Co-authored-by: 6543 <6543@obermui.de>
Co-authored-by: Giteabot <teabot@gitea.io>
Replace esbuild's rudimentary CSS minifier with [Lightning
CSS](https://github.com/parcel-bundler/lightningcss), which results in
around 4% smaller CSS bundle size, index.css goes from 608K to 584K.
The module is quite new but has active use in the Parcel bundler, and as
of yesterday is also under experimental support in Vite, so I trust it
that bugs will be sorted out quickly.
Before: `assets by path css/*.css 959 KiB`
After: `assets by path css/*.css 933 KiB`
I did give this a cursory test and everything seems to be in order.
Update JS dependencies, remove space after emoji completion (#25266)
- Update all JS dependencies
- Enable stylint
[`media-feature-name-value-no-unknown`](https://stylelint.io/user-guide/rules/media-feature-name-value-no-unknown)
- Make use of new features in webpack and text-expander-element
- Tested Swagger and Mermaid
To explain the `text-expander-element` change: Before this version, the
element added a unavoidable space after emoji completion. Now that
https://github.com/github/text-expander-element/pull/36 is in, we gain
control over this space and I opted to remove it for emoji completion
and retain it for `@` mentions.
---------
Co-authored-by: Giteabot <teabot@gitea.io>
Enable all webpack sourcemaps in dev build, disable all in prod build (#25089)
- Enable all source maps in dev build
- Disable all source maps in prod build
- Provide `ENABLE_SOURCEMAP` env var to override it.
I think the strange error seen in
https://github.com/go-gitea/gitea/issues/24784 is sourcemap related, so
if we enable/disable them all, it might go away. But it's most
definitely a Safari bug.
With all sourcemaps disabled, binary size goes down by around 1-2 MB,
with all enabled it goes up by around 12MB. If +12MB is acceptable, we
could also always enable them by default as fully source maps do have
some debugging benefits.
It's been disabled by default since 1.17
(https://github.com/go-gitea/gitea/pull/18914), and it never really
delivered any benefit except being another cache layer that has its own
unsolved invalidation issues. HTTP cache works, we don't need two cache
layers at the browser for assets.
## :warning: BREAKING
You can remove the config `[ui].USE_SERVICE_WORKER` from your `app.ini`
now.
[`worker-loader`](https://github.com/webpack-contrib/worker-loader) is
deprecated since webpack 5 which can load workers without it now, so
remove it. I think it was already dysfunctional because the regex does
not match our current worker scripts:
```
web_src/js/features/eventsource.sharedworker.js
web_src/js/features/serviceworker.js
web_src/js/serviceworker.js
```
I did confirm that eventsource worker still loads via simple
`console.log` inside the script.
Co-authored-by: delvh <leon@kske.dev>
Fix regression from https://github.com/go-gitea/gitea/pull/23481.
The conditional on the CSS import was being stripped away by webpack's
`css-loader`, resulting in the dark theme always loading. The old syntax
with `@import` nested inside `@media` also did not work as `css-loader`
(rightfully) ignores such non-standard `@import` syntax that was
previously supported by Less.
Unfortunately, we have to re-introduce postcss to the CSS pipeline to
fix this and I loaded only the minimal plugins to make it work.
There is one variant of the fix that does work without postcss, which is
to exclude the file from transpilation but I did not consider it as it
would have meant the `@import` was being done without a version suffix
in the URL, which would have caused cache issue.
Related: https://github.com/webpack-contrib/css-loader/issues/1503
---------
Co-authored-by: John Olheiser <john.olheiser@gmail.com>
Ran most of the Less files through the Less compiler and Prettier and
then followed up with a round of manual fixes.
The Less compiler had unfortunately stripped all `//` style comments
that I had to restore (It did preserve `/* */` comments). Other fixes
include duplicate selector removal which were revealed after the
transpilation and which weren't caught by stylelint before but now are.
Fixes: https://github.com/go-gitea/gitea/issues/15565
Follow #23394
There were many bad smells in old code. This PR only moves the code into
Vue SFC, doesn't touch the unrelated logic.
update: after
5f23218c85
, there should be no usage of the vue-rumtime-compiler anymore
(hopefully), so I think this PR could close #19851
---------
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Introduce customized HTML elements, fix incorrect AppUrl usages in templates (#22861)
This PR follows:
* #21986
* #22831
This PR also introduce customized HTML elements, which would also help
problems like:
* #17760
* #21429
* #21440
With customized HTML elements, there won't be any load-search-replace
operations, and it can avoid page flicking (which @silverwind cares a
lot).
Browser support:
https://developer.mozilla.org/en-US/docs/Web/API/Window/customElements
# FAQ
## Why the component has the prefix?
As usual, I would strongly suggest to add prefixes for our own/private
names. The dedicated prefix will avoid conflicts in the future, and it
makes it easier to introduce various 3rd components, like GitHub's
`relative-time` component. If there is no prefix, it's impossible to
introduce another public component with the same name in the future.
## Why the `custcomp.js` is loaded before HTML body? The `index.js` is
after HTML body.
Customized components must be registered before the content loading.
Otherwise there would be still some flicking.
`custcomp.js` should have its own dependencies and should be very light,
so it won't affect the page loading time too much.
## Why use `data-url` attribute but not use the `textContent`?
According to the standard, the `connectedCallback` occurs on the
tag-opening moment. The element's children are not ready yet.
## Why not use `{{.GuessCurrentOrigin $.ctx ...}}` to let backend decide
the absolute URL?
It's difficult for backend to guess the correct protocol(scheme)
correctly with zero configuration. Generating the absolute URL from
frontend can guarantee that the URL is 100% correct -- since the user is
visiting it.
# Screenshot
<details>
</details>
- Update all JS dependencies to latest version
- Enable unicorn/prefer-node-protocol and autofix issues
- Regenerate SVGs
- Add some comments to eslint rules
- Tested build, Mermaid and Katex rendering
#19999 introduced a indirect dependency with a license that was not on
our allowlist yet which produced this warning during webpack:
````
WARNING in License: citeproc@2.4.62 has disallowed license CPAL-1.0 OR AGPL-1.0
````
I've added both licenses to the allowed list and made it so webpack will
now abort on such license errors so that we don't miss those next time.
Co-authored-by: John Olheiser <john.olheiser@gmail.com>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
- Update all JS dependencies and playwright image
- Add new eslint rules, enable a few more, fix issues
- Regenerate SVGs
- Tested Vue and Swagger
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
This PR adds mathematical rendering with KaTeX.
The first step is to add a Goldmark extension that detects the latex
(and tex) mathematics delimiters.
The second step to make this extension only run if math support is
enabled.
The second step is to then add KaTeX CSS and JS to the head which will
load after the dom is rendered.
Fix #3445
Signed-off-by: Andrew Thornton <art27@cantab.net>
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
This removes the JS dependency in the checks pipeline. JSON output is
different because the previous JS did indent the license data
differently and a JSON key was changed, but the end result is the same
as it gets re-indented by wepack.
Co-authored-by: 6543 <6543@obermui.de>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: zeripath <art27@cantab.net>
`make go-licenses` will generate `assets/go-licenses.json` which is then included in the webpack build.
This step depends on both go and node being present, so unfortunately, I could not automate the generation by hooking it up to `tidy` as that target is triggered on CI where we do not have a docker image with both go an node.
It should be ran from time to time, ideally after each go mod update.
Enable contenthash in filename for dynamic assets (#20813)
This should solve the main problem of dynamic assets getting stale after
a version upgrade. Everything not affected will use query-string based
cache busting, which includes files loaded via HTML or worker scripts.
- Update all JS dependencies minus vue ones
- Remove workaround for case-insensitive attribute selector
- Add new linter rules and fix issues
- Tested SVG display and swagger
- Remove __dirname, use file URLs instead
- Upgrade fabric dependency
- Use fs/promises syntax, this breaks node 12 but we require 14 already
The change in public/img/favicon.svg is not caused by the fabric
upgrade, but it seems it was not properly generated when introduced.
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Migrate from U2F to Webauthn
Co-authored-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: 6543 <6543@obermui.de>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Fix CSS specificity issue with easymde's css (#18201)
* Fix CSS specificity issue with easymde's css
PR #18069 introduced a regression in certain overwritten editor styles
because the dynamic loading of easymde.min.css causes its's style to
apply after our supposed override styles.
Solve this by bundling the styles into index.css. We should later aim to
completely replace easymde.min.css completely with our own styles so there
are no more conflicts.
* Update web_src/js/features/comp/EasyMDE.js
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
* Update JS dependencies
- Upgrade to eslint 8 and add new plugin rules
- Adapt to various API changes
- Rebuild SVGs
* fix webpack warning on license
* order options alphabetically
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
It's about a 30% speedup in webpack build time with neglible differences
in the output size. We do lose the ability for CSS source maps, but I
rarely have a use for them anyways.
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
Co-authored-by: 6543 <6543@obermui.de>
- Limit SVG images to 600px width
- Adjust size of view toggle buttons to match other buttons
- Make Edit/Delete buttons easier to click
Had to create a separate CSS file because the less parser can not parse
CSS4 case-insensitive attribute selectors which are widely supported by
browsers.
Fixes: https://github.com/go-gitea/gitea/issues/15515
Fix missing icons and colorpicker when mounted on suburl (#15501)
* Fix missing icons and colorpicker when mounted on suburl
Signed-off-by: Andrew Thornton <art27@cantab.net>
* as per silverwind
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: Lauris BH <lauris@nix.lv>
It produces odd rgba values which also seem to cause issues in monaco's
color parser where the scoll shadow went red for some reason.
Regression by: https://github.com/go-gitea/gitea/pull/15333
- Add basic frontend unit testing infrastructure using jest in ESM mode
- Rename 'make test' to 'make test-backend'
- Introduce 'make test-frontend' and 'make test' that runs both
- Bump Node.js requirement to v12. v10 will be EOL in less than a month.
- Convert all build-related JS files to ESM.
I opted to run frontend tests run as part of the compliance pipeline because
they complete fast and are not platform-specific like the golang tests.
* Vendor node mods as cache; fix esbuild/fomantic offline build
* Fix --exclude; use bsdtar for consistent globbing
* Fall back to GNU tar; forward-compatible for APT 2.0
* Avoid having extd. attrs with bsdtar
* Dependency and misc. optimizations
* Remove extra code after esbuild-loader update
Co-authored-by: Mike L <cl.jeremy@qq.com>
- Upgrade webpack to 5.28 to enable publicPath option
- Use asset modules in place of deprecated file-loader
Co-authored-by: zeripath <art27@cantab.net>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: Lauris BH <lauris@nix.lv>
* Update JS dependencies
- Update all JS dependencies
- Adapt webpack config for version 5
- Update to Less 4.0, adapting usage of removed mixin syntax
- Enable new ESLint rules and fix discovered issues
* update license-webpack-plugin to fix missing licenses
* update license-webpack-plugin once more to get webpack into the license output
* switch to license-checker-webpack-plugin again for performance
* update deps again
Co-authored-by: Lauris BH <lauris@nix.lv>
Giteabot
silverwind
wxiaoguang
techknowlogick
Percy Ma
André Jaenisch
zeripath
Lunny Xiao
Kindyroo
