Please read b/108 to make sense of this.
This gets rid of the explicit list of exposed packages from nixpkgs,
and instead makes the entire package set available at
`third_party.nixpkgs`.
To accommodate this, a LOT of things have to be very slightly shuffled
around. Some of this was done in already submitted CLs, but this
change is unfortunately still quite noisy.
Pay extra attention to:
* overlay-like functionality that was partially moved to actual
overlays (partially as in, the minimum required to get a green
build)
* modified uses of the package set path, esp. in NixOS systems
Special notes:
* xanthous has been disabled in CI because of issues with the Haskell
overlay
* //third_party/nix has been disabled because of other unclear
dependency issues
Both of these will be tackled in a followup CL.
Change-Id: I2f9c60a4d275fdb5209264be0addfd7e06c53118
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2910
Reviewed-by: glittershark <grfn@gws.fyi>
Reviewed-by: sterni <sternenseemann@systemli.org>
Tested-by: BuildkiteCI
Main motivation for this is to get the openldap update that fixes
10 CVEs: CVE-2020-36221 to including CVE-2020-36230. See also this
issue which lists them all: https://github.com/NixOS/nixpkgs/issues/113490
Someone should also redeploy whitby as soon as this lands in canon and
all build failures have been fixed.
Things done to resolve upstream breakages:
* grpc no longer takes abseil-cpp as an input, it has also been removed
in the override.
* Upgrade glittershark's kernel to 5.11 since the linuxPackages_5_9
attribute has been removed by upstream and the patch used by them is
available for 5.11 as well.
* The fixed output hash for third_patry.apereo-cas changed for some reason.
* Remove the pin of haskellPackages.vector from the haskell overlay. It
broke as the most recent version of vector in nixos-unstable no longer
depends on semigroups. This effectively updates vector from 0.12.1.2
to 0.12.2.0.
* Align two comments in tvix/libstore/worker-protocol.hh because the
updated clang-format now demands that.
Change-Id: I2ecf10a98de935e9222acf1feaea447d4c11ed2d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2538
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
Reviewed-by: glittershark <grfn@gws.fyi>
Reviewed-by: sterni <sternenseemann@systemli.org>
Nix internally differentiates between lambdas and primops, but their
type in the nix expression language is the same (lambda). The
implementation of builtins.functionArgs only checks if the given
expression is of type tLambda and fails if the type is tPrimop or
tPrimopApp which are also functions. This most notably breaks
lib.generators.toPretty when called on a builtin making for example
yants fail if a primop is typechecked and an error message is
generated.
This fix generates an empty set for primops like for plain lambdas
and is based upstream commit b2748c6e99239ff6803ba0da76c362790c8be192.
Additionally we add to two tests:
* eval-okay-functionargs now includes a few test cases checking that
builtins.functionArgs always returns an empty set for builtins and
also works as expected for normal functions.
* eval-okay-types now also checks if builtins are functions.
Future work would be to make builtins.functionArgs work as users would
expect for builtins like builtins.fetchurl, builtins.fetchGit etc. which
take a set as an argument. These currently don't register as formal
arguments, but it would be an usability improvement at least if they
did.
See also https://github.com/NixOS/nix/pull/3626#issuecomment-698546704
Change-Id: I2bf4cb80d44a4b72ade13d3e0dbd7dfb1d049f32
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2477
Tested-by: BuildkiteCI
Reviewed-by: Profpatsch <mail@profpatsch.de>
Reviewed-by: glittershark <grfn@gws.fyi>
Since we don't have a Bindings implementation with unstable order this
function is not required, as its callers can just iterate over the
attributes instead.
Change-Id: I01b35277b5a2dde69d684bc881dbd7c0701bcbb3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2291
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
To make it easier for future tests to use the arbitrary specialisations
we've defined for some nix types, centralize them all in a single
arbitrary.hh header file.
Change-Id: I767f27949cfe7ec55c79901f7d7aa538d2f98c6b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2182
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Add missing break statements to the big switch block in
BuildResult::FromProto, and cover the whole thing with a rapidcheck
round-trip test.
Change-Id: I7a07ca398cc5f02ca4fd8e6256fd563c6f3aea9d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2178
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Similarly to how we did for buildPaths, add a std::ostream& log_sink
parameter to the build_derivation method on Store, and pass it std::cerr
when called at the top level by nix commands - most notably, the
build-remote hook binary, so that we get build logs when using tvix as a
remote builder.
Change-Id: I0f8f729ba8429d4838a0a135a5c2ac1e1a95d575
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2176
Tested-by: BuildkiteCI
Reviewed-by: andi <andi@notmuch.email>
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Upstream doesn't, and for good reason - this might not be a store path,
for example if it's /bin/sh
Fixes: b/73
Change-Id: If9aa96de2cd8ab941c098a9f122b8b302a92ec38
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2175
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Tested-by: BuildkiteCI
The --quiet and -v/--verbose flags were removed during the glog
conversion in d0c44425e1, which has
previously broken programs like e.g. home-manager, which passes --quiet
to nix-build.
A nix-build-specific workaround was added in
24f9354d5b, which manipulates the
FLAGS_stderrthreshold global variable from glog. This commit moves the
--quiet logic back into the argument handling code in libmain, and adds
corresponding handling for -v/--verbose.
Change-Id: I13d860ebbb78541d9f1236691a1efe8bd2163c67
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2170
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
This was referencing a nonexistent note in buildPaths, for one, but for
another let's get log outputs when calling this RPC.
Change-Id: Ic9d17834b356ea84d69692ccc0249d09777e833b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2173
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Subclasses of std::streambuf are expected to override overflow in
addition to xsputn, as it's called in certain cases by the non-virtual
methods. In our case, this was preventing endlines from getting sent
over the log stream.
Change-Id: I70d00f0c7cb8f8cf2f744f58974c21e7a70a715b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2172
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
- A couple of log messages had double-newlines still
- Remove the [ drvPath ] bit from the "building path" output, since the
derivation path is already present in the message itself.
Change-Id: Ib4b514a18fcb63c7191f4d7cf0a9b857aaa3f3fa
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2171
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
This allows users to use C-A C to access the QEMU console while within
the VM terminal. From there you can control the process (e.g. kill it).
Change-Id: If2eec2e385fd27fc4e20921dcad78be7390f9d2b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2167
Reviewed-by: glittershark <grfn@gws.fyi>
Tested-by: BuildkiteCI
For the build of tvix we can safely ignore all the files in src/protos
as well as all the nix files in the root of the repository. By passing
the input sources through builtins.filterSource we can make a per-file
decision whether or not we want to copy it into the store (in other
words: if we consider it relevant for the build).
This enables much faster development on the actual build expression.
Change-Id: Ib4c01dbe9cbfa9770922f6257af2a7259814ce0d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2159
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
Reviewed-by: lukegb <lukegb@tvl.fyi>
The different signedness of level and withLevel was causing implicit conversions. Use a nullopt instead of a -1 sentinel value.
third_party/nix/src/libexpr/nixexpr.cc:242:21: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
Change-Id: I7c2cadb6fd6bbff6c5b84028651ad4ebba423297
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2157
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
Reviewed-by: glittershark <grfn@gws.fyi>
Implicit signedness conversion was causing warnings. Perform a static_cast; this value should never be out of range.
third_party/nix/src/libmain/shared.cc:310:12: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
Change-Id: I34fe53d186ae33f970147135bbe3894fb39a0ec7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2156
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
Reviewed-by: glittershark <grfn@gws.fyi>
third_party/nix/src/libstore/sqlite.cc:89:33: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
Change-Id: I1dbc01acaecc4f382f63e7108429cd814ff05f4f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2155
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
Reviewed-by: glittershark <grfn@gws.fyi>
This is passed by some third-party tools, notably home-manager. At some
point we should fully revamp argument parsing and log verbosity setting,
but for now this gets us to more of a working state
Fixes: b/78
Change-Id: I4134b17315e30145fd43d3659f407d7121ce8876
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2136
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
It seems like the amount of logging is causing things to move a little
slower - even if that's not really the case, it gets in the way of
debugging things.
Refs: b/76
Change-Id: I9ea99a3b16e3307a0b0371bad22d03b0e2175af6
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2134
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
To aid in both debugging and (eventually) printing stacktraces, add a
Pos member to a few more Expr variants.
Change-Id: Ic1d2a056fc7e6c07bc3e79fa38845cb4a5da5ca5
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2133
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
Included fixes for random breakage:
* 3p/awscli: pick from the stable channel; it is broken on unstable
* 3p/googletest: bumped version & removed patches that nixpkgs applies
* 3p/lisp/cffi: bumped library version for SBCL compat
* 3p/nix: fix libsystemd attribute
* 3p/nix: reformatted (clang-format handling of ternaries changed)
* glittershark/home: Use home-manager from nixkpgs
* glittershark/kernel: bumped linux-ck patch hash
* glittershark/kernel: removed "patch patch"
* multi/whitby: Use home-manager from nixpkgs
* tazjin/frog: drop Sourcetrail (it doesn't build currently)
Note that in addition to these changes, some previous CLs updated the
versions of git and cgit which was necessary for this channel bump,
but which could not be done in the same commit due to the nature of
the subtree merges.
Change-Id: If2563e8a68e2750c4b913a976ff7b93b42e8b7f3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2110
Tested-by: BuildkiteCI
Reviewed-by: multi <depot@in-addr.xyz>
Reviewed-by: glittershark <grfn@gws.fyi>
The Abseil version of `StatusOr` does not come with the status macros
or the `Consume*` family of functions.
This change modifies the existing code to use the common denominator
of the API that is available between Abseil's own implementation of
`StatusOr` and the one from Tensorflow that we are currently using.
Change-Id: I5c37f68636a1fd54d153f95d7303ab8644abb774
Passing a string directly to add_paths like this causes the proto class
to take ownership over the string, meaning when it is destructed it
will *explicitly* free the string. When the string's actual owner (the
derivation struct) then goes out of scope it'll get freed again, causing
a double-free. This fixes that to instead use the copy constructor to
assign to a pointer to a new path, and covers the whole to_proto method
with a rapidcheck test.
Fixes: b/64
Change-Id: I84235bed9104ff430a0acf686d4a96f1e2e9a897
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2106
Reviewed-by: tazjin <mail@tazj.in>
Tested-by: BuildkiteCI
This was accidentally using the proto arena API to assign the derivation
field of a BuildDerivationRequest. We *thought* this was causing a
double free, but even with this change that's still happening. That
said, this change is probably still a good idea since it's using the
proto API as intended.
References: b/64
Change-Id: I950a4eafb214e9113639ea54d2dfd4659b7be931
Reviewed-on: https://cl.tvl.fyi/c/depot/+/2104
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
The code that calls queryPathInfoUncached explicitly catches the
InvalidPath exception and translates it into a null result - but the RPC
code was throwing a regular old Error for invalid paths. At some point
we should get rid of all the exception-driven control flow in this whole
thing, but in the meantime this gets us back to functional.
Change-Id: I2a38790ee0c691ab0c8394c7738d7693fa42aa10
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1980
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
The static_assert is present to force us to change the constructor when a member grows.
Change-Id: Ifa4f4a03eb7ce13cff109cf26ecf0032045905e9
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1972
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
Make all static std::strings constexpr std::string_views, and replace
concatenation with absl::StrCat where necessary.
Technically all of these are constant, so they really don't need to be
top-level statics - and since I'm trying to get rid of as much global
state as possible in preparation for making the nix daemon properly
multithreaded I figured I'd knock these out while I was at it.
Change-Id: Ibd3ad9ef68f0a0eacb135541b39fdb13dae042e1
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1939
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
This is constant, so let's make it a constexpr.
Change-Id: I6a9eb9f740979740f730ebe142d2cbf23dbcfd70
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1961
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Since the daemon is running in threads now rather than forking a process
per connection (thanks to grpc) this static flag to prevent accidentally
initializing a new worker during a build is getting stepped on by
multiple threads. This converts it to thread-local, and also adds an
actual message to the check so that if it hits in the future we know
what's going on.
Fixes: b/58
Change-Id: I07a2f1582e56709c104f79935e5405fa24888f59
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1940
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: V <v@anomalous.eu>
This is mostly to help in tracing through the core dumps for b/58, but
is also generally a good idea while we're still in hardcore
debugging-mode.
Fixes: b/63
Change-Id: Ib816a2a98bf3b8afa6e8b255a74ccb93af6508c5
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1937
Reviewed-by: tazjin <mail@tazj.in>
Tested-by: BuildkiteCI
This reverts commit a41c3dedb1.
Reason for revert: I believe its time has come.
Change-Id: I4133dc8a3b57b40b280646d1622ed7a0409c5246
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1941
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Write *all* build logs (essentially everything that wasn't a DLOG), not
just the stdout of the builder, to the client-bound log_sink_ that's
threaded through build.cc. These logs are significantly more useful on
the client, both when debugging tvix itself and (more importantly) when
building things with it.
To make this all work, this commit also refactors the passing-around of
the log sink in build.cc slightly - the main reference is now owned by
the Worker class, which already stores pointers to itself on the Goals
it creates, so those Goals can then load the log sink from the Worker.
This change was originally made to allow for logging to the log_sink
from instance methods of the worker, but actually feels significantly
more natural than storing the sink on the goal anyway.
Change-Id: I461685bc7f5c0a326054940ac5971b568dd43f8e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1925
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
These were not used anyway, and *may* have been causing a weird series
of problems leading to things sometimes aborting. Or maybe it has
nothing to do with it - the only correlation we've found is from
disassembling the output of a core dump. Regardless, this feels
stylistically like a good idea.
Change-Id: I288b50945e74ea90c118a21767fa4387c9f47f7d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1921
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Rather than printing "Nix" in the version string, print "Tvix". This is
cosmetically nice, but also is very useful when making sure I actually
have installed tvix on a system successfully.
Change-Id: Idd1a9954bd66509327f52b7457cfba2f090ab30c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1924
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Tested-by: BuildkiteCI
This allows distinguishing between an empty value and no value.
Patch ported from upstream at
ba87b08f85
Change-Id: I061cc8e16b1a7a0341adfc3b0edca1c0c51d5c97
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1884
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
When the nix daemon starts up, first check (using sd_listen_fds) whether
we have been systemd socket-activated. If so, instead of passing the nix
daemon socket path to grpc, start a manual accept(2) loop, passing the
client file descriptors to grpc via AddInsecureChannelFromFd. There's an
open grpc issue at https://github.com/grpc/grpc/issues/19133 for
building support into grpc to do this automatically, but as of right now
this appears to be the only way to make this happen.
Making this happen, by the way, was a bit of a journey - at one point I
attempted to ServerBuilder's experimental AddExternalConnectionAcceptor
API, and that didn't work either - it appears that the final missing
piece to getting this working was explicitly fcntl(2)ing the client file
descriptors to set O_NONBLOCK before passing them into gRPC. With that
set, this all works inside of the test vm.
Fixes: b/56
Change-Id: I5d2ab2b5b02eb570249b30a9674e115c61b0ab0e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1882
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: tazjin <mail@tazj.in>
Tested-by: BuildkiteCI
Expose depot to the test vm via NIX_PATH, so we can test the build of
depot packages from inside the test vm.
Change-Id: I26314634895ab1339f5715c0b0b261028eefdaa3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1923
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
