1
* [ ] dag doesn't draw between-lines correctly when they cross stream load boundaries
2
3
* [ ] https://github.com/GitoxideLabs/gitoxide instead of libgit2
4
5
* [ ] We might want to make an über-NIF including connecting to Comrak/ammonia/etc. ourselves (or a separate NIF for that), since overriding stuff in MDEx/Autumn is hard. See markdown.ex.
6
7
* [ ] show dag nav on mobile with popout
8
9
* [ ] use read replica where appropriate
10
* [ ] "hooks" also point HEAD to tags when received, oops
11
* [ ] autumn and mdex NIFs are both around 80MB; the latter includes the former. dedupe.
12
13
* [ ] relative URIs: ugc- prefix, #xyz shouldn't navigate away, etc.
14
* [ ] pressing browser forward/back should trigger fancy blob highlights, not
15
just the automatic :target ones
16
* [ ] some patch links really need to scroll back to top
17
18
* [ ] https://talya.nossa.ee ?
19
20
### short term
21
22
- [x] blob
23
- [x] images
24
- [ ] pdfs, etc.
25
- [ ] less polysaturated, leaner design
26
- [ ] do we use Binary explicitly to not have to pass Unk back as a linked list? @fun
27
- [ ] get our tests and coverage fixed again. @pls
28
- [ ] message to end-user when following a redirect (via git)
29
30
### features
31
32
- [ ] blob permalink
33
- [ ] responsive (mobile) styles
34
- [ ] override locale user option
35
- [ ] diff 2+ parent commit
36
- [ ] passkey support
37
- jfc https://shadowfacts.net/2023/phoenix-passkeys/
38
- [ ] default branch: choose from existing branches; non-UTF-8 support
39
40
### housekeeping
41
42
- [ ] tab indices on e.g. crear repositorio
43
- [ ] IPv6 check
44
- [ ] redirect to primary host if not on it
45
- [ ] reverse where the structs are defined (between Git2/Nossa.Storage*)
46
- [ ] rename RepositorySelf* to RepositorySettings*?
47
- [ ] extract CGI into its own application; stream request/response in/out
48
- [ ] consider erlexec, or write our own driver, so we can get stderr out of
49
the way
50
- [ ] gettext once-over in hard-to-reach spots
51
- [ ] verify typescript in compile step
52
- [ ] override elixir/erlang used in autumn/mdex to keep all together
53
- there's some wip of this in zmlqmwzq
54
55
### graceful degradation
56
57
- [ ] "method" links don't work with noscript since they don't do the right
58
thing
59
- [ ] can't dismiss flash
60
- [ ] javascriptless sidebar changes? (then they must go in the URL)
61
62
### production ops
63
64
- [ ] caching: nginx, varnish? fragment?, git object? how much does bandwidth
65
cost at ovh? bunny? integrate caching, backup into nix module.
66
- right now the pawiwo visibly reloads on every page
67
- [ ] DIY anubis? (for easy deploy to other Phoenix/etc. strength in diversity @fun
68
is a real consideration in this fight.)
69
- [ ] backup restore procedure/test
70
- [ ] feature that restores from backup as we make it. ez!
71
- [ ] admin diagnostics page (combine into/with live dashboard somehow?)
72
- [ ] per repo: audit log, pushes, etc. (git hooks?)
73
- [ ] LRU-ish app-level caching: memcached? redis? ets? mnesia?
74
75
### future
76
77
- [ ] avatar upload
78
- [ ] federation?
79
- identity: auth to other nossa instance
80
- auth another instance for avatars
81
- federated MR-type deal?
82
- don't want to redo a whole code review thing here, idt? check out gerrit
83
etc.
84
- see also simon tatham's thing on git; pull request is just "please check
85
again"
86
- opinionated about rebase/jj norms
87
- at the very least we could show interdiffs between pushes etc.
88
- [ ] merge queue? what do i need at work?
89
- [ ] Log in with GitHub/GitLab/Forgxejo? make it easy to work across, to not be
90
completely hecked by network effects?
91
92
93
## archivo
94
95
- [x] change "settings" top right to "account"
96
- [x] repository delete
97
- [x] repository storage
98
- [x] push keys
99
- [x] proper logging (insert some in e.g. storage creation failure)
100
- [x] some kind of log aggregation/interface on kala
101
- [x] test coverage back up
102
- [x] do some more!
103
- [x] work out why jj fails to push!
104
- [x] fix up this /blah/!url nonsense
105
- [x] git store tests
106
- [x] git tests async: use per-runner store
107
- [x] last controller tests
108
- [x] don't rename storage: use only numeric ids in storage
109
- [x] figure out why our distribution metrics ("phoenix.*") don't make it to
110
grafana (or appear in /metrics at all)
111
- [x] try ETS
112
- [x] pawiwo icon
113
- [x] clean up RepositoryGitControllerTest
114
- [x] Anubis/Nepenthes/?
115
- See https://lobste.rs/s/dmuad3/mitigating_sourcehut_s_partial_outage --- we
116
will need to be prepared; robots.txt, etc., and above all usage stats/graphs
117
so we can see it happen.
118
- [x] https://git.sr.ht/robots.txt is a good start, though of course it isn't
119
saving them. Still, no reason to start from zero.
120
- [x] start of gettext support
121
- [x] main browsing interface, ui polish/consistency
122
- [x] navigate tree
123
- [x] zindex fix for icons overlaying nav bit
124
- [x] select commit
125
- need to resolve a rev instead of a branch now
126
- [x] include breadcrumbs in Diff view
127
- [x] wherever "path=" is sneaking into our URLs, get it out (breadcrumbs)
128
- [x] liveview for navigating within
129
- [x] extract inline classes into css (w/ @apply)
130
- [x] dark mode styles
131
- [x] actual colour system
132
- [x] pawiwo svg for colours
133
- [x] ok dark mode again
134
- [x] safari, chromium
135
- [x] repository redirects
136
- [x] tailwind v4
137
https://elixirforum.com/t/mix-tailwind-4-0-0-beta-1-support/67636/17
138
- [x] more post-liveview cleanup of repository controller; check over
139
auth-related stuff.
140
- [x] then back to (re)design
141
- [x] work out whether diff view will be within the same liveview, a different
142
one, etc.? - it may be overkill; navigating the tree responsively is nice,
143
but the huge diffs might make browsers sick. gotta test!
144
- [x] decide on the repository view design; do we keep the commits there? do
145
we make the sidebar collapsable a la ~~yuta~~ hexdocs etc.? something
146
else entirely?
147
- what do i want to see when i hit a root of a repo? what are the various
148
reasons i end up on one of those on github/srht/etc.?
149
* linked to as a project's "home page", searched and the first result (or
150
apparent best result for an official download or similar)
151
* directly navigated to because i want to browse the source of something
152
i don't have a local clone of (and i haven't yet hit the threshold
153
where i need to use tools to look thru; source aware views can sometimes
154
abrogate the need but often (e.g. github's views + semantic index)
155
they're an untenable degree of both meh design usability (in e.g., links
156
are all too small, everything bunched up on the side, often starting to
157
scroll horizontally by this point; not a priority) and flawed impl (in
158
e.g. elixir outlining. is this just from elixir-ls? tree-sitter?))
159
* often want a specific version (i.e. something i'm running/using),
160
sometimes just browse main/master if cbf/not relevant/more interested
161
in design than exact
162
* summary of this point: "whatever would be unnecessary if i always had
163
a clone of everything". i don't, but i'm in the position of improving
164
the situation for when you don't -- what actually helps?
165
* sometimes i do this even in the presence of a clone on disk, because
166
i'm *using* that one somehow, but i cbf creating a new worktree.
167
similar result. a browser window (and the way you can open lots of
168
them, to various paths/versions/etc) is a much smaller cognitive unit
169
to manage than a worktree.
170
* how often do i want to see the whole commit list straight up front like
171
this? (day separations would give this a lot more context, for what
172
it is.)
173
* almost always would prefer the *graph*; i almost never prefer the
174
straight log style locally. (think about jj log view.)
175
* there are three axes: commit (temporal), path (spatial), view (fn).
176
github exposes commit consistently, path inconsistently (compare blob
177
view to commit history restricted to a subpath), view inconsistently
178
("Commits" page vs 2.5 tree pages (root, non-root w/ README, non-root;
179
sidebar tree only visible in non-root) vs blob (very aligned with
180
non-root tree). root really is the odd one out, since it's a landing
181
page.)
182
* big commit *graph* that stashes in side: file listing-style tree view
183
that then collapses into expanding tree sidebar: blob.
184
* then you can shift around points in the graph while retaining
185
visibility over the directory tree, no reloads/scrolls breaking
186
"immersion".
187
* does the graph stash over to a plain log? i should make sure while i'm
188
developing that i work with the browser the sizes i want to actually
189
be using it at. (i.e. push back against default to 2/3rds code, unless
190
i'm developing a usecase where i'd have the browser at 1/3rd).
191
* sometimes you do want to view just the log of a file or subtree.
192
kinda wanna apply that filter *to* the far-left graph, as a
193
saved-in-navigation thing. then you can nav to a commit (still
194
centred on the tree or file you started the filter from), see it in
195
context, move around, and then at some point you back up or clear
196
the 'path filter' on the DAG view.
197
* note that we want the current path to remain in the tree view
198
when navigating the DAG view even when we don't have a path filter
199
active. same with current blob, if any.
200
* this kind of removes that fake third axis, exposes the other two as
201
proper orthogonal ones. github flattening the DAG with no way to
202
recover is criminal.
203
* and then there's commit view, which is a whole other thing: viewing a
204
commit message & diff.
205
- [x] let's fix our URIs in gen i guess, ?path= is ugly
206
- [x] support submodules in tree
207
- [x] be explicit in which params are UTF-8 encoded & which aren't
208
- [x] global support for non-UTF-8; no error messages
209
- [x] navigate commits on left; focus current commit as navigating
210
- [x] compile-time the storage layer choice
211
- [x] vyxos push fails!
212
- [x] "last commit modified" links are incorrect for subtrees, there's a TODO in
213
git2.rs somewhere
214
- [x] decompose git2's monster "resolve_commit" into somewhat orthogonal ops
215
- [x] syntax highlighting
216
- [x] little live demo with channels of push response
217
- [x] reaffirm backup situation for kala
218
- [x] nix syntax highlighting egumi??
219
- [x] wow! nossa in docker/k8s!
220
- [x] ES: "Clave de empujar creada exitosamente" when creating a new repo!?
221
- [x] new repo => brocken
222
- [x] remove register links if register is disabled
223
- [x] install our own git hooks in repositories, set HEAD on push if unset
224
- [x] assign default branch on first push
225
- [x] this causes clone failures, the weird nix flake stuff we've been seeing, etc.
226
- [x] floating windows don't preserve state on livereload
227
- [x] let's fix the first thing: navigating back to a subtree shouldn't close an
228
open blob. we can keep our existing assigns, or just stream.
229
- [x] open new windows in specified order
230
- [x] default orientation based on aspect ratio
231
- [x] splith(3): need to have up/down too; splitv(3) needs to have left/right.
232
it's too hard to just do that with merges and unmerges.
233
- [x] skip floating windows when finding clientInDirection; similarly,
234
moveClient/mergeClient; their behaviour is very puzzling if it's not clear
235
what's happening
236
- [x] "hooks" are breaking on current live because they're symlinks into old
237
versions of nossa, since GC'd from the nix store!
238
* [x] shynet: add javascript call on live patch
239
* [x] what's nom? https://github.com/NixOS/nixpkgs/pull/430275#pullrequestreview-3089109039
240
* [x] set inline-block for repo breadcrumbs so the underline doesn't over-extend
241
* [x] NixOS tests to ensure the module stays good even after we move to deploying on k8s
242
* [x] per-repo deploy keys: R/O, R/W
243
* i want this for flux
244
* [x] everything except the UI
245
* [x] the UI
246
* [x] update the unique index to be per-repo since we've changed auth anyway
247
* [x] tree view isn't getting the right commits for files in subdirectories
248
since we're back to one-directory-at-a-time and need to redo how we search
249
for the last changed commit.
250
* [x] actually show commit details at top of tree/blob views
251
- [x] branch selector disappears on safari with javascript on
252
* [x] address our lack of sendmail in container; send mail to status.hostIP:25 (no auth)
253
- [x] dag infinite scroll/load to context
254
* [x] visual confirmation on "copy to clipboard" shouldn't be a modal, just temporarily change the icon
255
