Optimisations seem to cause *oauth2-redirect-uri* to be baked in if we
try to check the environment in DEFVAR.
Change-Id: Id53b880061239141936298f673ff8e745442af94
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13186
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
* //tvix/*-go: regenerate protobuf files
* //tvix/boot:
- Explicitly set compression method of mkBinaryCache which has
made this configurable and (at the same time) changed the default.
- Adjust to change of extension of the nar files from .nar.xz to
plain .xz.
Change-Id: Ie79ea8e0ac8fe04ae01f5558bffca93e9314f56d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13174
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Build //3p/lisp from pkgs proper, i.e. nixpkgs' nixos-unstable channel
instead of nixos-23.11 (yikes).
Basically, multiple package sets are attached to the different lisp
implementations now instead of having a “generic” lispPackages
set (which defaults to sbcl). We can just use that instead even though
it looks a bit weird having `srcOnly sbcl.pkgs.foo` everywhere when the
packages is not necessarily related to SBCL.
We could in theory create a source only package set by abusing how the
infrastructure works internally, but it's probably somewhat brittle:
callPackage (pkgs.path + "/pkgs/development/lisp-modules/imported.nix") {
build-asdf-system = { src, ... }: src;
}
Since we do a pretty hefty jump in package versions, many packages have
to be adapted to internal changes and restructuring:
- bordeaux-threads
- cffi
- cl-colors2 (which has been deprecated, but is still required by other
packages)
- cl-smtp
- cl-plus-ssl
- cl-prevalence
- hunchentoot (compiling the asd file no longer seemed to work)
- ironclad (fixes for SBCL compiler warnings caused a CCL compiler
warning)
- nibbles (revert the only commit to sbcl-opt/x86-vm.lisp that's new
compared to canon since it broke compilation for unknown reasons)
The following new packages had to be added as existing packages added
new dependencies:
- frugal-uuid, frugal-uuid/non-frugal
- trivial-clock
Change-Id: I8b94894df0357907cf2b27cf1e34a7e804b68e02
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13134
Autosubmit: sterni <sternenseemann@systemli.org>
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
I am already using this outside of depot where it isn't under VCS. It
makes sense to canonicalize it, also to ensure that it stays somewhat
generic.
Change-Id: I5595d98ab4198794c395feb4d3c08df1e2d01a36
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13184
Autosubmit: sterni <sternenseemann@systemli.org>
Reviewed-by: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Remember //nix/escapeExecline? Well, this is worse (for some possible
meanings of the word). Instead of taking a list and escaping/rendering
it to an execline script, Execline reimplements part of execlineb(1):
It takes a (nested) list where any enclosed list signifies an execline
block and produces the corresponding execline argv form as described in
execline-block(7). This means the result of Execline can directly be
executed using •SH without the need for execlineb(1).
Consequently, execlineb(1)'s management of positional parameters and the
environment are not available. This is fine for the intended purpose of
Execline (glueing together shell commands efficiently without messing
around with the FFI in BQN for pipe(2) etc.).
Change-Id: Ief69b1bab919c16b6e39c3f5dc3db628766c5a8c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13180
Reviewed-by: sterni <sternenseemann@systemli.org>
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
These days highlight.js in mdbook includes syntax highlighting for Nix
out of the box, there's no more need to ship our own variant of it.
Change-Id: Idd453c2efe97707ae861de9fa08c9dab71ea7876
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13178
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: adisbladis <adisbladis@gmail.com>
Tested-by: BuildkiteCI
Currently ingest_entries panics. This makes it hard to use
ingest_entries in parallel with other processes. When another process
runs into an error and wants to return so, while ingest_entries has
already started and has no entries yet, we're forced to panic.
With this change ingest_entries will return an error when there are no
entries, thus allowing the user to handle the errors as they please.
Change-Id: I78b85bf18f52af8c157d6bedad6019fd4398250a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13146
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: Bob van der Linden <bobvanderlinden@gmail.com>
Reviewed-by: Bob van der Linden <bobvanderlinden@gmail.com>
Tested-by: BuildkiteCI
Something recently caused us to replace Docker with Podman (I guess a default
changed in nixpkgs? I don't remember making the change explicitly), which broke
the reindexing unit.
Change-Id: I1d3453ed970e536abb540c6ef79765cfda271810
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13173
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
Autosubmit: tazjin <tazjin@tvl.su>
I've managed to (several times!) lock myself out of the office network by
rebuilding while the configuration was unavailable. Might as well assert that.
Change-Id: Ib2685c0e8dab977a3cb576606e70f8960968b574
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13168
Autosubmit: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
We're not saying 2.3 here anymore in all cases, but use the version
specified in the `nix_version` function argument.
Change-Id: I86c34f97eea4141b8431de7c6a784750d6b7c2cb
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13166
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Brian Olsen <me@griff.name>
This provides a shell containing the necessary packages to run the
integration tests too.
//tvix:shell stays the same.
Change-Id: Iac7977cbb7dc085da698993be9baca662d09ff5f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13164
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Instead of collecting all rows into a Vec, use the stream_rows function,
and massage Ok values into the right shape.
This avoids collecting everything into memory.
Change-Id: I2f03434c7e1e25ee8395c69685b16100aec36f89
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13163
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Since cl/12661 this doesn't need to be Arc'd.
Change-Id: I3c14c45df7815867ddea6f461e81649772648962
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13155
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
make the Display impl of B3Digest use the `blake3-` prefix rather than
`b3:`, which is more close to the SRI hashes we use in NixHash.
Also use Span::current().record() to record the `blob.digest` field (and
only when the digest has the right lenght). It doesn't make sense to
brand it as blake3 before anyways, and allows us to get rid of the
manual implementation.
Change-Id: Iee349557ef4761807af1fb3942387de1942ab12b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13162
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Ilan Joselevich <personal@ilanjoselevich.com>
Tested-by: BuildkiteCI
This is probably a nicer default, and also aligns more with Serialize.
Change-Id: I3a8f18607c305dbf9fb93a2e2dd02ad07cd99f24
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13161
Reviewed-by: Ilan Joselevich <personal@ilanjoselevich.com>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Included changes:
* targeted primary NixOS channel back at `nixos-unstable`
* //3p/chicago95: disable new broken symlink check
* //tvix: omit unnecessary lifetime clippy recommends
* //users/sterni/blipqn: wait a bit for data to arrive in test so it
succeeds under load.
* //fun/paroxysm: force pkg-config flag of pq-sys by adding a bogus
dependency on it. Otherwise, pq-sys will try to use pg_config
which does not work correctly in pkgs.libpq at the moment.
* //users/flokli/keyboards/dilemma: disable temporarily
Change-Id: I6d53bd7bca6886f3457e1f41505e97314f4cd191
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13119
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: aspen <root@gws.fyi>
Reviewed-by: sterni <sternenseemann@systemli.org>
This got disabled in 5f17df8548 without
much investigation.
After some more digging, I found the root cause -
173de320197ba97f2daf46b7d229055cd3732df4 in nixpkgs changed the files
produced by `mkBinaryCache` to use FileHash, not NarHash to name NAR
files.
While this is more correct and consistent with how it's produced by Nix,
it unfortunately means our uploads will now get rejected.
We didn't notice this, as curl (and rush) were not configured to bail
out - which now is the case again.
Deal with the change, by decompressing twice, once to get the NarHash
again, the second time to feed the data to curl.
Change-Id: Ic1deb470801250e7967083ac78cd46be39f3caa3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13122
Reviewed-by: Ilan Joselevich <personal@ilanjoselevich.com>
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
git-rev-list(1) orders the printed revisions before printing. This is
quite complicated and I noticed that it behaves in unexpected ways
w.r.t. commits given individually: sometimes they are reordered,
sometimes not. This is problematic when interdependent commits are given
via the command line individually: If they aren't re-ordered by
git-rev-list(1), we'd be cherry picking them in reverse order.
To solve this, just call git-rev-list(1) on all arguments individually,
allowing ranges to be resolved, but always picking individual arguments
in the correct order.
Notably, this means that excluding commits via individual REV^ arguments
won't ever work, but I believe this wasn't possible previously
anyways (maybe due to --no-walk?).
Change-Id: If8e83e0b47a74baf37d77b6c4f68a55af944b366
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13149
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
Autosubmit: sterni <sternenseemann@systemli.org>
The TVL overlay can lead to diverging store paths, e.g. starting with
CL/13119, firefox depends on some FODs that in turn depend on Nix. Thus
the TVL overlay (with Nix 2.3) will come up with a different drvPath
than just using (import pkgs.path {}).
To fix this, just automatically use the same arguments to nixpkgs as
Tvix for expectedPath.
Change-Id: I77d39df9092442a49c8ee94b012dec046ff6a07e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13147
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
These bots are doing unthrottled requests to cgit 24/7, and it's starting to
annoy me.
Change-Id: I6b7d7a68e9becb8ed4b5c52b376c2a60febc6ec6
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13145
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
- Reduce visual noise, e.g. parens where it hurts readability.
- Separate parts of trains visually where it helps readability.
- Reduce amount of unnecessarily pointfree code.
Change-Id: Ie0ef03af7d344863a81bf0557b90a4dfb7974e84
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13126
Autosubmit: sterni <sternenseemann@systemli.org>
Reviewed-by: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Normal agents can easily go from 16 -> 24 (proportionally to whitby, this makes
more sense).
I've kind of randomly decided to label 6 agents as large ones. We will filter
things like eval, or building tvix tests (until b/431 is resolved).
Change-Id: Ib38d2c56410c2ad9d86a01546c00192f87274bb3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13121
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
Adds a new tagging system to Buildkite agents, where agents are tagged with
large/small slots. All agents have small slots, only some agents have large
slots. The small slots are purely informative - nothing targets them, whereas
large slots will be used for filtering agents.
This allows us to target large slots in some builds and minimise the concurrent
execution of extremely large builds, while keeping a large number of small slots
around for all the light targets.
This will need some tuning over time (also because tagging is a manual process).
Change-Id: I15aa657773ed874d84d98e55238fb31c75d4efa7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13120
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
Passing a namespace as 𝕩 to •CurrentError is not allowed since it is
reserved for future use (not sure how it'll change the behavior
exactly). Since WithFlipdot always passes a namespace as 𝕩, this is a
relevant problem for blipqn. To work around it, we wrap the passed
argument in a list (and unwrap it for 𝔾).
Change-Id: I3382f754cbb64ad799e47b7d174d641c43a0f2e0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13125
Reviewed-by: sterni <sternenseemann@systemli.org>
Autosubmit: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Included changes/fixes:
* bumped all `wasm-bindgen` usages again
* regenerated protobuf files
* keycloak terraform provider has been migrated to new name
This also included a state migration in the bucket, which I've already
performed.
* tvix/boot: disable tests that are broken in CI
* users/aspen/yeren: avoid upgrading kernel to 6.12
digimend depends on a fix: https://github.com/NixOS/nixpkgs/pull/378830/
Change-Id: I657dcf5c4d0d08f231bfe30e37c8062bfcfaaa32
Reviewed-on: https://cl.tvl.fyi/c/depot/+/13098
Reviewed-by: aspen <root@gws.fyi>
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
Autosubmit: tazjin <tazjin@tvl.su>
