Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

These should now let us reconstruct the commit object (or at least the
git cat-file equivalent) appview-side.

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: oppiliappan <me@oppi.li>

this also populates the script with the correct executable path for
the `knot` command.

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: oppiliappan <me@oppi.li>

performing rev-list and filetree in sequential occupy a large amount of
the client budget. we can parallelize all these expensive operations.

Signed-off-by: oppiliappan <me@oppi.li>

uses `git log --name-only` to speed things along. the performance
tradeoff here is that active repositories tend to load last-commit times
really quickly, but directories with one file that was modified 20k
commits ago will result in the entire log being walked through.

git-log command is parsed as it streams to avoid having to wait for the
entire command to complete, and as soon as the entire directory's
listing is hydrated, we cancel the command. most of the time, this
results in hydration of subdirectories also (this gets cached),
resulting in good experience when browsing a repo (typically moving down
directories gets faster with each click, as fewer files need hydration).

Signed-off-by: oppiliappan <me@oppi.li>

RepoIndex and RepoLog should no longer take longer on bigger
repositories. all endpoints are still backwards compatible.

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: oppiliappan <me@oppi.li>

replace with `knot guard` instead.

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: oppiliappan <me@oppi.li>

in profile pages, when expanding activities, it can cause a page to
begin overflowing. this settings reserves space for the scrollbar
upfront.

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

And rename resolver -> idresolver to be a bit more explicit.

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Improved logging and adds a 5s tolerance to the iat claim so PDSes with
a slightly skewed clock (or just aggressive timestamp checks) should
work.

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Rename FullyResolvedRepo -> ResolvedRepo

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

repoguard still uses an in-memory resolver.

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: oppiliappan <me@oppi.li>

trigger comparison on button click, this simplifes a variety of things:

- we can load a diff on page visit without javascript
- we can avoid modifying url using javascript and breaking back buttons
- we can avoid a lot of javascript code

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: Anirudh Oppiliappan <x@icyphox.sh>

Also move FormatPatch to types to avoid an import cycle.

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

- Disable caching of responses as suggested by documentation.
<https://git-scm.com/docs/http-protocol#_smart_server_response>
- Return plain-text errors so the git client will display them.
- Check Content-Type of upload-pack request.

v1-style requests are unaffected and v0 (dumb-protocol) requests remain
unsupported.

this gets noisy really quickly when the stack is rebased. now, the only
things that can cause a resubmit are: reauthoring commit messages or
modifying the patch itself.

speak more about tangled itself, link to discord, irc and source code.

- fix kotlinscript detection (there is an open patch upstream for this)
- fix gleam highlighting

the stacking working introduced some unnecessary changes here