First pass at an xdotool-based command to edit the current text input in
emacs
Change-Id: I1e04612478292fe83083d197d481e034a9fce97f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9971
Reviewed-by: grfn <grfn@gws.fyi>
Autosubmit: grfn <grfn@gws.fyi>
Tested-by: BuildkiteCI
This can apparently work around some of the CPU throttling bugs on
~modern~ computers.
Change-Id: I807ece85d3eba53857a1cb1e73a33f7924538e96
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9895
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Expose `deps` separately, add a direnv with PATH_add for it to bring
tooling into $PATH.
Change-Id: I432cd2b082cad89e08bef78dc4653e10e137cd6b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9842
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Avoid having to re-enter the shell whenever the config is changed.
Change-Id: Ib9f6bb4075e29acaeb4863d64c017695ca85b60b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9841
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
This add the EC2 box config to the repo.
Change-Id: Id7a888a2cfbf1454cd9f9465018df377e14b4e9f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9836
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
This adds a deploy-archeology script.
I tried getting morph to work first, but passing it a
depot.ops.nixos.nixosFor seems to be very hard - the NixOS module system
doesn't like the arguments it's called with.
Replace morph with a 3 line bash script, which assumes your ssh_config
contains config for an `archeology` host.
Change-Id: I2bf694c60ded39c201efbbb899f3b5512aa4d0f2
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9835
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: edef <edef@edef.eu>
And of course I managed to move the cache creation into the handlers,
instead of doing it before starting the webserver.
And now I managed to create a hopeless mess of callbacks, but oh well.
Change-Id: I73c3aeced71923c7372496286a279e326b20c388
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9813
Reviewed-by: Profpatsch <mail@profpatsch.de>
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
I’ve been wanting to experiment with this stuff for a while,
abstracting away a handler type.
The existentials for parser and body took a bit of mucking about, but
in the end hiding the variable behind a `Body` constructor did the
trick.
Now every handler has its own cache, which means we can start caching
arbitrary results.
Change-Id: If57230c47f97ef4c548683f2c2f27660817a31f2
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9812
Autosubmit: Profpatsch <mail@profpatsch.de>
Reviewed-by: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Now this is getting cool. After 5 minutes we will ask the backend
again (which takes like 3 seconds), but then we compare the old cached
result with the new result and only send it back to the client iff it
changed.
So the client will still have to wait for the roundtrip time, but
doesn’t have to pay for the content. Plus, it gets some info that
upstream hasn’t been updated.
Change-Id: I6dba40321949da5da6a16b2e799d939573c77ba7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9811
Autosubmit: Profpatsch <mail@profpatsch.de>
Reviewed-by: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
This is a dumb experiment to see how hard it is to respect cache
headers; turns out, medium hard but doable.
Sets the correct expiry time according to the cache, plus respects
`If-Modified-Since` which is a tiny bit harder.
Change-Id: I9e6166af0fa254df2beb0f3919187b91a407487b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9810
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Reviewed-by: Profpatsch <mail@profpatsch.de>
Okay, so I guess you also have to seq the cache and everything in
between the IORef and the data.
Change-Id: I4c79c99afbd09e83e9d7a01d58b31b36862e4d11
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9807
Reviewed-by: Profpatsch <mail@profpatsch.de>
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Otherwise the table might potentially hold onto data from the website
request, it’s hard to say.
Change-Id: I786478bd1ce2d9775b3d0b57565d79666ef8a96f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9806
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Reviewed-by: Profpatsch <mail@profpatsch.de>
Back at my bullshit.
Mostly copied the setup from whatcd-resolver.
Change-Id: I9edd4387ee73c18816b1692d5338735536cce70f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9803
Reviewed-by: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Autosubmit: Profpatsch <mail@profpatsch.de>
We can cross-reference all of these to schema.org, it should work for
most of the fields.
Change-Id: I38d8dbc7e964764886ddd156c4148bcf3ee376f3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9744
Autosubmit: Profpatsch <mail@profpatsch.de>
Reviewed-by: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Ideally we can figure out how to search for single songs by grepping
through musicbrainz. For this we kinda need the jsonld results, so
this is a first step which visualizes the structure and makes it
easy-ish to lazily traverse it.
Change-Id: Ieca21674dee8e8c2dacbab4f2f15ccbe067da647
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9743
Reviewed-by: Profpatsch <mail@profpatsch.de>
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
We want to use this quite generic parser type for other things as
well.
Change-Id: I890b43c58e479bdf2d179a724280ef1d8748fafa
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9742
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Reviewed-by: Profpatsch <mail@profpatsch.de>
nixpkgs changed something in how it deals with configuration of the
package set itself when that is externally instantiated (like in
depot)
It seems like we can work around this mostly by just ... deleting some
code, as all instances of this were for allowing unfree code, which
we've already set on the top-level anyways.
* //users/sterni: fix nixpkgs config assertion to point at
pkgs.config
* //users/wpcarro: disable locate service, which is broken in nixpkgs
Change-Id: Iacf6f1c8fd5b5289e7265e155d74f8269a858ceb
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9541
Reviewed-by: sterni <sternenseemann@systemli.org>
Reviewed-by: wpcarro <wpcarro@gmail.com>
Reviewed-by: grfn <grfn@gws.fyi>
Tested-by: BuildkiteCI
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: tazjin <tazjin@tvl.su>
Without this line, accessing external APIs from Gonic (e.g. scrobbling
to Listenbrainz) does not work (unless a local resolver exists). This
is because in the pure Go DNS implementation used in gonic, only
/etc/resolv.conf can provide resolvers.
Change-Id: I26dd9a845b0a70c4cfb983c68da2b76b7e57dfec
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9502
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
* zramSwap: during reindexing, geesefs can be pushed to the boundaries
of what the new overaggressive OOM killer accepts, and I don't want
to bother configuring that thing instead.
* umount: geesefs dying unexpectedly leaves broken mounts around,
clean these up before proceeding
This bakes in the assumption that there's only ever one geesefs
service, but that assumption is baked in anyways.
Change-Id: Id85c2f5bc2312a7246ea20229eb36d2cc1bd82c0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9500
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
... this will make sense soon!
Change-Id: I1f8f32d655afdf868fff4bd09e1fea2943fd7558
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9496
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Can't be bothered to make something more automated for now.
Change-Id: Ie245af90c1a62a5988b97d16f86b6567e1ffafd0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9493
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Experiment of how to instrument a lib I’m using to trace instead of
log.
Now that we added MonadTracer to Transaction, we can drop the unlifted `inSpanT`.
Change-Id: Iea891a58cfb33a0837978611456c33aefcccf0d7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9491
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Reviewed-by: Profpatsch <mail@profpatsch.de>
For it to work, you need otel (e.g. jaeger) to run on port 4317.
Change-Id: I36f0493b9be26af256769ae5af8916029036a76e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9488
Tested-by: BuildkiteCI
Autosubmit: Profpatsch <mail@profpatsch.de>
Reviewed-by: Profpatsch <mail@profpatsch.de>
This is horrible, but I don’t think there’s a more pragmatic way to
set this up right now.
Change-Id: I166550292b4eb979864f7d69ea56a2827bbf21bd
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9487
Tested-by: BuildkiteCI
Autosubmit: Profpatsch <mail@profpatsch.de>
Reviewed-by: Profpatsch <mail@profpatsch.de>
Instead of inserting torrents and every tour group seperately, insert
the tour groups and then the torrents in one go (unzipped).
I finally found a good use for list comprehensions, flattening nested
lists.
Change-Id: I7dfc765ad058dff3afb3b03887141b334a4b1988
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9486
Tested-by: BuildkiteCI
Reviewed-by: Profpatsch <mail@profpatsch.de>
Autosubmit: Profpatsch <mail@profpatsch.de>
Still n+1, but now we got the IO where we want it to make the database
inserts efficient.
Change-Id: I2ee36ea41bf186cbeb5581b0df802bb3611769fe
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9485
Reviewed-by: Profpatsch <mail@profpatsch.de>
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Now we can move the I/O into a where block.
Change-Id: Ib5334948f3d11ca120ce0b7a46c67f8500fdab3a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9484
Reviewed-by: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Autosubmit: Profpatsch <mail@profpatsch.de>
Returning an I/O action was a good first approximation, but leads to a
n+1 query problem, making the whole shebang pretty slow after doing a
search.
Thus we need to split data & I/O, so we can be more clever in the next
commit.
Change-Id: Ieb2f8d5445f1258047da9b121b977c0b8d2dd7f8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9483
Reviewed-by: Profpatsch <mail@profpatsch.de>
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Telega tries to use some random completing-read functions, but I just
want it to use the default one, which I've configured the way I want.
Change-Id: I33da24f257e8ac0cae4a7d25ba6c5e4e656fcad5
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9478
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: tazjin <tazjin@tvl.su>
This is unmaintained and has been removed from nixpkgs. As an
alternative, 'eza' can be installed.
This is the last instance of 'exa' in depot, so the workaround from
the channel bump has been removed.
Change-Id: Id915260b6c969a6b2a8ecae49d3ce92285f09f71
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9330
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: wpcarro <wpcarro@gmail.com>
* run on port 9092 (transmission runs on 9091)
* run postgres on port 5431 instead of 5432 (to not interfere)
* only search for albums for now
* correctly handle missing torrent file in SELECT
Change-Id: I20125f7731c9b80a9e8ea05b726adfb1244a24bc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9335
Reviewed-by: Profpatsch <mail@profpatsch.de>
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
If more than one page is returned by the search, query all of them.
Also add an ON CONFLICT clause in case the torrent group already
exists, to update it.
This function is getting a bit unwieldy (plus it suffers from an n+1
problem), but ok.
Change-Id: Ib505a2be8286d658ae44a3fe124a4fb42d0fc0c6
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9334
Autosubmit: Profpatsch <mail@profpatsch.de>
Tested-by: BuildkiteCI
Reviewed-by: Profpatsch <mail@profpatsch.de>
Some of these I don't use, some of these have been replaced with
tree-sitter, some of these have moved into core.
Change-Id: Id0a8aac4189607a89db861bbc09b95c75c45dc0f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9310
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
The mechanism in nixpkgs does not work, but doing it myself is not
that difficult.
Change-Id: I91b12e9c1f5d381c9990fd46e02169f5275f88da
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9309
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Especially interested in the C++ mode, as I am regularly dealing with
extremely large C++ files where the existing mode is having trouble.
Change-Id: I8c17a35011a23b67043de5de58c4269c636de68b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9308
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
I want to experiment with the new tree-sitter modes in Emacs, maybe
they're good.
Change-Id: I706fc8a889fcfa61d0f11d50ecbc67a2899b5bbf
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9307
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Also expose both formats, then use it from
users/tazjin/presentations/tvix-eval-2023.
Change-Id: Id906e8aff5510a7a4f33336326472e86db18ea32
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9280
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: tazjin <tazjin@tvl.su>
Makes it possible for people in the audience to get to the link
collection reasonably quickly.
Later on that post can also have the talk itself added. For now it is
unlisted.
Change-Id: If9ae1d88d3c6f22bbd70b2b3cbea0b0d42895a93
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9208
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: Mark Shevchenko <markshevchenko@gmail.com>
Tested-by: BuildkiteCI
This demonstrates a Rust stdlib call that just causes runtime panics
on WASM, for explaining the problems with porting Tvixbolt.
Change-Id: Ief974f1bba509fdac4b9bc9f862ee8f4dfc5158e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9206
Tested-by: BuildkiteCI
Reviewed-by: Mark Shevchenko <markshevchenko@gmail.com>
Autosubmit: tazjin <tazjin@tvl.su>
refine doesn't work anymore, inspector seems to be a more modern
alternative.
it's impressive that they managed to write elisp code that broke.
Change-Id: I672de68abdc3d780f66769043afefd8d37438548
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9209
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
I spent way too much time fighting LaTeX to do roughly what I want
here again, but all the alternatives are even worse.
Change-Id: Ibe12a4ce175ceb73e9d6e276613dcd4827dd76c4
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9150
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: Mark Shevchenko <markshevchenko@gmail.com>
Tested-by: BuildkiteCI
We don’t strictly need servant-multipart, if all we need is to parse
some multipart forms. This removes some deps.
Change-Id: I218731fada056b9edfb3d01fc33880673d14473e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9187
Tested-by: BuildkiteCI
Reviewed-by: Profpatsch <mail@profpatsch.de>
This gives me the ability to override the Emacs per-machine easily.
Change-Id: Id480889c108833b0a11c377a9b1e946900c5aba1
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9166
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
I only wrote this as a hack to make my Emacs config run on gLinux, but
that isn't relevant anymore.
Change-Id: I19c49d500e0ec75bb85644d25a63b0b6c530aa62
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9165
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Instead of staying up-to-date with Emacs master, I'd like to be a bit
more laid back and stay with stable releases. Now that native
compilation is in this is not a big difference.
I tried to use `-pgtk`, but it broke EXWM.
Change-Id: I0789a1f73d0149bda912987e726de396545abce1
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9164
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Instead of producing a binary that gets called by Emacs, with
input/output serialisation, use a dynamic Emacs module that lets Emacs
more-or-less directly call the relevant GTK functions.
I'm doing this mostly as an experiment. Might be interesting to end up
with a dynamic module that I can dump some experimental code into that
improves my workflows.
To do this, I've exposed the emacs binary used by my Emacs
configuration in an additional `passthru` field. This ensures that the
module is linked against the right version of Emacs.
Change-Id: I1426994fe3455ed1b2a685c5a09705e29fa40950
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9163
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: tazjin <tazjin@tvl.su>
This built-in emacs library actually provides a data structure that
can work as an LRU list through the existing helper function to move
an element to the front of the ring if it already exists.
As a result, the code for workspace history moving becomes a lot less
brittle and complicated than it was before. No more carefully figuring
out when to modify state, just push it in the ring unless it's being
rotated already.
Change-Id: If354e0618fc5a6d7333776468eec077596cfe9df
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9162
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
I haven't used that since ... 2018 or so, time for it to go.
Change-Id: I5e1b729bd553940b98335e3d9c7ca5b134fdf692
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9161
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
This does nothing other than dump out the list of XDG apps to stdout
in JSON format. There are no options or anything.
This can be used for selection in app launchers (e.g. dmenu, something
based on completing-read in emacs, rofi, etc.).
I wrote this because I don't want to deal with having to do this in
Elisp. It's also unclear what logic actually hides behind under the
hood here, so why not just use the official library.
Change-Id: I16fed2c92760cadecc02c59a4e537a1fa247aff9
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9157
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
vertico and consult are more modern versions of interactive narrowing
helpers, as those implemented by ivy and its related packages.
The primary differences (and what I care about here) is that they are
more focused on integration with the core Emacs primitives, rather
than building an ecosystem around them.
For example:
* vertico enhances `completing-read' and friends, but does not attempt
to provide its own ecosystem of functions to *trigger* completions.
* vertico integrates with the default `completion-style' system,
meaning that I can continue to use things like prescient without
extra packages that integrate it with vertico
* consult does not rely on vertico or any other specific completion
framework (such as counsel/swiper do with ivy), and simply
implements its functions using completing-read
This reduces the overall amount of code in the dependency closure and
leads to a less special setup.
Functionality is basically equivalent, except for two things which
counsel came with that I will need to substitute:
* counsel-notmuch (actually this was a separate package, but I didn't
use it much anyways, so just ignoring it for now)
* counsel-linux-app (opening desktop shortcuts, this I will need to make)
As a side note, consult notes "This package is a part of GNU Emacs",
but it doesn't seem to be the case.
Change-Id: Ia046b763bf3d401b505e0f6393cfe1ccd6f41293
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9155
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
When entering an incorrect GPG key password, I don't want my whole
buffer layout ruined. A small error message in the modeline is enough.
Change-Id: I7318d685e74fa4e110a9bff30d0de9f7f18b2be4
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9149
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Especially the recursive minibuffers only cause a mess.
Change-Id: I6f7f790acd6a2f8cc4cec26c9cf97d5e53e77106
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9148
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Adds a completing-read function (defaulting to ivy for me, but it
doesn't matter) that offers a reliable alternative to standard
buffer-switching implementations.
In particular, this implementation retains a mapping of
buffer names to their buffer *objects*, so that the correct buffer is
selected even if some renaming took place during the selection.
I tried to account for a bunch of the common behaviours I could think
of:
* invisible buffers are ... invisible
* entering a buffer name manually creates that buffer, if there is no
match
* ... unless that buffer is an invisible buffer, in which case it is
selected and switched to
* the first element is always `(other-buffer (current-buffer))`,
because of the ordering of #'buffer-list
Yet, the entire code of my implementation is less than the *setup*
code of ivy-switch-buffers, so it's possible I missed something. Well,
I'll find out ...
Change-Id: I08be0da0863d06c9a930e5efaf916719655db90e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9147
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
