1. Top
  2. legit
  3. Commits
legit

Fork of https://git.icyphox.sh/legit

Commits at 1fe217cff9d89738257162a063dd2de0ef6c621f

  1. 1fe217cf Fix double slash on repo tree page Shota FUJI authored at Shota FUJI comitted at
  2. 27b6f86d Add fuzz test for repo pages Shota FUJI authored at Shota FUJI comitted at
  3. c0aab936 Add tests for page routing Shota FUJI authored at Shota FUJI comitted at
  4. fde482b9 Fix unrecognized paths under repo page returns HTTP 500 Shota FUJI authored at Shota FUJI comitted at
  5. 74199381 Fix URLs including refs with slashes handled incorrectly Massive refactor. The net/http.ServeMux is still useless for this level of routing requirements. I initially tried to encode slash characters but they decode path components and they say it's not a bug. huh? Thankfully, URL structure of legit is tidy and organized so I could just write "mux" using if statement and switch statement. To be honest, I quit like this approach compared to sprinkling string patterns everywhere using struct holding mutex for weird usecase. I'm not including this in a patch release considering the amount of changes I made. As a test, I ran cmd/save-commit-pages against Zig repository cloned under demo/ directory before and after this patch. "diff" command showed nothing and exited with 0, so at least commit page is not regressed. Shota FUJI authored at Shota FUJI comitted at
  6. 2b8671d4 Load aside section before main section on commit page Commit page's body can be quite long depends on a commit. If it's sufficiently long, browser renders aside after long main content. The problem is that causes reflow thus unnecessary layout calculations. With this patch, aside allocates its space by appearing before the main content, so regardless of how long the main content, reflow does not happen. The only downside is order of tab navigation and screen reader pronouncing sections is different to other pages, due to HTML/CSS spec and legacy bullshit a.k.a., accessibility practices is total garbage. But this is worth the inconvenience given how inefficient and annoying the reflow is. Shota FUJI authored at Shota FUJI comitted at
  7. 1a250e6f Add "ui.diff.hideThresholdLines" option Rendering over 10,000 lines diff is tough for browsers. This option hides the element to ease layout calculation. This does not change the page load speed, though. Shota FUJI authored at Shota FUJI comitted at
  8. 74c5502f Add option to generated diff using system git go-diff is too slow. As legit parses the generated patch file, we can simply swap diff generator based on option. This is an opt-in config for backwards compatibility. The diff generated by git command is different (more intuitive hunk.) This is part of https://tangled.org/pocka.jp/legit/issues/25 Shota FUJI authored at Shota FUJI comitted at
  9. a430e231 Create internal cmd to test commit page regressions Shota FUJI authored at Shota FUJI comitted at
  10. 94dec475 Add load test for commit page with large diff The commit page is so slow on large-ish diffs. Shota FUJI authored at Shota FUJI comitted at
  11. cd6335e2 v1.1.0 Shota FUJI authored at Shota FUJI comitted at
  12. b6743137 Add trailing slash to always-307 redirecting links Links this patch touches require a trailing slash, and request without trailing slash will be 307 redirected to the one with a trailing slash. This patch eliminates those unnecessary roundtrip. Shota FUJI authored at Shota FUJI comitted at
  13. fa2cc136 Add staticDirRevision option / build flag Mainly for invalidating browser cache after upgrading legit. https://tangled.org/pocka.jp/legit/issues/22 Shota FUJI authored at Shota FUJI comitted at
  14. 1d9d7675 Fix 404/500 pages have broken style and are unhelpful https://tangled.org/pocka.jp/legit/issues/21 The initial plan was to just fix the broken style. As I editing the HTML, I started to think this is a great place to put bug report link. In the past, I put this kind of diagnosis info on an error page in my work and that greatly helped debugging issues. The info here is just a four-line text that is obvious in server log (if setup,) but end user attaching this in complaints or whatever is really helpful, trust me. UA is for grepping log lines. Even if it was generalized (for privacy reasons,) it's still useful and time-saver. Shota FUJI authored at Shota FUJI comitted at
  15. 67ae060c Update README Slightly outdated. Shota FUJI authored at Shota FUJI comitted at
  16. bd275169 Fix GET / is 404 when non-git repository exists in scanPath https://tangled.org/pocka.jp/legit/issues/20 Regression introduced in e31c3f8f4e4271038ef02560da68c6154b2e6121 Shota FUJI authored at Shota FUJI comitted at
  17. f4085ceb Add option to group repositories by category https://tangled.org/pocka.jp/legit/issues/17 Category grouping is convenient for telling project's state and/or kind (mirror, fork, demo.) My hosted instance serves 19 repositories yet it's messy and hard to see an overview. Shota FUJI authored at Shota FUJI comitted at
  18. 8e0d9c46 Read description from git config Align with gitweb's behavior. Shota FUJI authored at Shota FUJI comitted at
  19. db55fd92 Fix test repository shares git directory and worktree Possible, but not practical. Shota FUJI authored at Shota FUJI comitted at
  20. e31c3f8f Use os.Root for path checking os.Root uses system primitive rather than stringy securejoin. Ideally, git.Open should accept fs.FS (or *os.Root) but go-git in its current form (even v6) does not expose a function for "smart" open method that can open both bare and non-bare repositories. This patch also accidentally fixes following bugs: * checked out "description" file in a non-bare repository is treated as a description file. * ignored repository reveal using URL-encoded slashes (see the second test added in the parent commit) Load tests on a minimal setup (nix run .#testing) shows no differences before and after this patch. This is not a clean solution, but code is less duplicated and more correct and secure. I may revisit and refactor in near future. Shota FUJI authored at Shota FUJI comitted at
  21. eaeadc91 Add tests for path traversal Shota FUJI authored at Shota FUJI comitted at
  22. 43984419 Remove unused EditorConfig rule Shota FUJI authored at Shota FUJI comitted at
  23. e869ba63 Delete GitHub Actions workflow file Shota FUJI authored at Shota FUJI comitted at
  24. b92f7c89 Add relative time to commit timestamps' tooltip At first I tried to put these datetime formatting code into preferences file. But it didn't work well due to increased load time = content flashing. Shota FUJI authored at Shota FUJI comitted at
  25. 4059aa1e Override <select> UA styles Chromium (Linux) uses insanely low-contrast colors on dark mode. Shota FUJI authored at Shota FUJI comitted at
  26. 66cdbc5a Workaround degraded UX due to BFCache BFCache is the worst thing happened in the web. The technology itself is not bad, but how it's introduced (or who introduced) was the problem. A company that tries to advertise their hardware, or accurately say, deceive their hardware's customer to them believing their hardware is faster than competitors, started deplying BFCache widely to their browser that is only available one on the hardware. Because the intention, and there is no spec for back/forward behavior, they introduced the behavior to existing web pages without opt-out mechanism. Their defence is "this is good for users, and only situation this is not desirable is navigation after logout", which is completely bullshit. Client-side apps relies on established "script reload on navigation" were broken and had to add workaround or adapt to this breaking change. They didn't care as their goal is to create software monopoly using their proprietry app store, and web apps are "competitor" to those motherfuckers' products. I could have set "Cache-Control: no-store" but decided to "adapt" BFCache because: * disabling BFCache by using that header is side-effect, so "primary effect" takes place (I have a plan for cache design and this is not a good time to introduce the change) * those mfs can change heuristic or break web by another method, and this is more future-proof. Avoiding (fighting against) the cache is a hack, unfortunately. * whether showing cached content is good in this software is debatable, but considering there is no cache design yet, supporting BFCache is better than the prior state. The new code has consistent code style albeit little bit inefficient on storage access. I believe future changes naturally will follow this pattern and will put "sync" code into "pageshow" handler. Shota FUJI authored at Shota FUJI comitted at
  27. 22ff30a7 Add preferences for file tree The default is wasteful for my taste. Shota FUJI authored at Shota FUJI comitted at
  28. 9593ebb6 Add theme color preferences (JS) I hate the default blue convention. Shota FUJI authored at Shota FUJI comitted at
  29. f9ad53cb Move tab width selector into preferences dialog I'm going to add other options as well. Shota FUJI authored at Shota FUJI comitted at
  30. 99fddaf4 Set "color-scheme" HTML meta tag So UA applies their dark styles for unstyled elements. After this patch, tab width selector will use dark theme style provided by UA in dark mode. Shota FUJI authored at Shota FUJI comitted at