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This change was originally made to allow for logging to the log_sink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>
These are unused, and I'm not sure they ever actually worked.
Change-Id: I6bb6b4257b7815932581ded7929437dba80961c9
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1885
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: tazjin <mail@tazj.in>
The various bits (including the unix socket that we listen on) append
/nix to this - so it isn't expected to be in there.
Change-Id: I0d95f2cc5641f610c1bdb03b036fac61dac9842d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1890
Reviewed-by: tazjin <mail@tazj.in>
Tested-by: BuildkiteCI
The nix package is expected to install /etc/profile.d/nix.sh and
/etc/profile.d/nix-daemon.sh into its output - these set the environment
variables expected by the client-side nix user commands, eg NIX_DAEMON.
Fixes: b/52
Change-Id: I5fb964c076f092d0ef7e1d49aca37875c866fb90
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1889
Reviewed-by: tazjin <mail@tazj.in>
Tested-by: BuildkiteCI
Rather than sending the entire AddTextToStore request along in a single
message, send it in a stream of chunks using the same metadata-first
approach we've been using for the other store gRPC requests. This fixes
a bug where certain builds could send more data than the maximum gRPC
request size (4194304 bytes, it would appear), resulting in a
RESOURCE_EXHAUSTED error.
The initial chunk size, which is currently constant but should be made
dynamic at some point in the future, has been chosen based on the IPC
bandwidth delay product for tazjin's desktop, rounded up.
Change-Id: I6f0232cdbc98653484816b39855126873fc59a03
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1835
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
Reviewed-by: kanepyork <rikingcoding@gmail.com>
These are the scripts I use to test the nix daemon interaction with a
non-/nix store directory during development, copied almost verbatim from
my cmake build directory. As such, there's likely a *lot* of cleanup and
deduplication to be done here, but I'm committing these as is in the
hope that others can benefit from them somehow.
Change-Id: I42a10a85e6731fa2014c7ea9738224d678a8376b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1881
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Tested-by: BuildkiteCI
The command line options --arg and --argstr that are used by a bunch of
CLI commands to pass arguments to top-level functions in files go
through the same code-path as auto-calling top-level functions with
their default arguments - this, however, was only passing the arguments
that were *explicitly* mentioned in the formals of the function - in the
case of an as-pattern with an ellipsis (eg args @ { ... }) extra passed
arguments would get omitted. This fixes that to instead pass *all*
specified auto args in the case that our function has an ellipsis.
Submitted upstream at https://github.com/NixOS/nix/pull/3965Fixes: #46
Change-Id: I32b7ee0e5bacf75b2bc43a3f0796f533f4bd5959
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1863
Reviewed-by: tazjin <mail@tazj.in>
Tested-by: BuildkiteCI
Previously, MixEvalArgs (a generic data type used to handle --arg,
--argstr, and -I arguments to `nix-build`, `nix eval`, etc.) was storing
the difference between --arg and --argstr by prepending a single
character (either 'E' or 'S') to the value of the arg. This is messy and
un-type-safe, so this commit refactors that to use a proper enum and a
std::pair, which allows us to add a switch and get totality checking.
yay, types!
Change-Id: I883f8db3136d05dda190fac0a1b494386c2ff87b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1862
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
This is the clang-tidy lint 'google-explicit-constructor'.
There's a whole bunch of breakage that was introduced by this, and we
had to opt out a few types of this (esp. the string formatting crap).
In some cases minor other changes have been done to keep the code
working, instead of converting between types (e.g. an explicit
comparison operator implementation for nix::Pid).
Change-Id: I12e1ca51a6bc2c882dba81a2526b9729d26988e7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1832
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: glittershark <grfn@gws.fyi>
The previous clang-tidy invocation missed some header files, which has
now been rectified.
Change-Id: I31547754fbf52f439dc7aeefb08ab90bd50c4156
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1831
Reviewed-by: glittershark <grfn@gws.fyi>
Tested-by: BuildkiteCI
.. turns out producing an entire build log on a single line is not
useful.
This does not use `<< std::endl` because we have run into buffering
issues with the implementation of the logs->gRPC sink, but intend to
replace this in the future using a structured sink for BuildEvent
protos rather than a raw stream.
Change-Id: Ia9b05fa804391d389e2ef53ab4436c0ec5cc452e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1828
Reviewed-by: glittershark <grfn@gws.fyi>
Tested-by: BuildkiteCI
Introduces the actual log sink in several places where we actually
want the build logs to thread through correctly.
This should cover *most* build paths.
Change-Id: I735dff8a79f7e35a5874eb89b4abb980f9703dc2
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1827
Reviewed-by: glittershark <grfn@gws.fyi>
Tested-by: BuildkiteCI
This *should* wire up the builder's logs all the way back through the
gRPC client, where they are then conveniently discarded.
Change-Id: I65f22526d0b5a8b8d90f28665bc1b4bc7f7c802a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1825
Reviewed-by: glittershark <grfn@gws.fyi>
Tested-by: BuildkiteCI
This passes an output stream for build logs to almost all relevant
functions inside of build.cc by threading it through the
`Goal`-abstraction.
Store calls that create goals but don't have a sink available use the
DiscardLogsSink().
Change-Id: I2c0cb1aec1f9150f33113f4752055cea518ede8b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1824
Reviewed-by: glittershark <grfn@gws.fyi>
Tested-by: BuildkiteCI
This part of the store API needs to carry a handle to the log sink
from now on, so that it can be passed in as appropriate from the gRPC
handlers.
In all places where there is no such handler available at the moment,
the discarding log sink has been inserted. This can be used as a
convenient grep target in the future.
Change-Id: I26628e30b4c6437dccdf8f722ca2e8ed827dfc19
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1797
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: glittershark <grfn@gws.fyi>
this shuts up a few lints ...
Change-Id: I51ee9418767e97247f98b170cdad137e5b12b44d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1796
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: glittershark <grfn@gws.fyi>
This is not actually legal code, but it kind of ... works. There are
more of these around, these were just the ones Griffin stumbled upon
while working on the build logs.
Change-Id: Iff9821d8fe145dd426648a8ff4510a73f67c9b7d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1795
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: glittershark <grfn@gws.fyi>
Introduces a class which implements std::streambuf by sending build
log lines to the provided gRPC stream writer as individual messages.
This can be used in the implementations of calls which trigger builds
to forward logs back to the clients.
Change-Id: I3cecba2219cc24d56692056079c7d7e4e0fc1e2c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1794
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: glittershark <grfn@gws.fyi>
Introduces a new `BuildEvent` proto type which is streamed in response
to calls that trigger builds of derivations.
This type can currently supply build statuses, log lines and
information about builds starting.
This is in preparation for threading build logs through the processes.
Since we have nowhere to send the logs (yet), a null sink is used
instead.
Co-authored-by: Griffin Smith <grfn@gws.fyi>
Change-Id: If7332337b89506c7e404cd20174acdaa1a3be4e8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1793
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
Reviewed-by: kanepyork <rikingcoding@gmail.com>
In some cases we don't have anywhere for the build logs to go. Until
we understand those cases fully and can get rid of them, this null
sink implementation can be used.
Change-Id: Ib93c43caf268e2c01c43d59737a829e8c43d223e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1792
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Reviewed-by: glittershark <grfn@gws.fyi>
If the global nix config, or any available user nix config location, does not exist, then the loadConfFile() function will throw and not finish initalizing the Nix configuration.
Change-Id: I6db83bca54d9b66699356d107720603476e32c23
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1657
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
Now that we have access to a store in tests, we can enable the tests that needed a store.
Additionally, move the expected output files for disabled tests into the disabled folder.
Change-Id: I2492d49d43b93c7c9b0463e4d3d2855a5a51365d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1758
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
Value now carries a shared_ptr<Bindings>, and all Bindings constructors return a unique_ptr<Bindings>.
The test that wanted to compare two Bindings by putting them into Values has been modified to use the new Equal() method on Bindings (extracted from EvalState).
Change-Id: I8dfb60e65fdabb717e3b3e5d56d5b3fc82f70883
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1744
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
Reviewed-by: tazjin <mail@tazj.in>
This is in preparation for making some of Value's members into refcounted ('smart') pointers.
Change-Id: Ibc54e23ac35766a2fd4e14871c9a7c936a603778
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1743
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
Previously the outlist would not be used, and it would sort god knows
what in the out value.
This was probably introduced by the std::vector refactoring, and the
language test for builtins.sort was disabled. Whatever reason there
was for disabling it seems to be gone, so we're re-enabling it.
Change-Id: I98941c2cad78df58ff7bea1ece3aaa4133e94bf8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1757
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Tested-by: BuildkiteCI
This allows you to write EXPECT_OK(statusor), as well as
EXPECT_THAT(status, IsStatusCode(StatusCode::kInvalidArgument).
Change-Id: I53bed694d812c501eb305ed4ddb358e1f9a68277
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1704
Tested-by: BuildkiteCI
Reviewed-by: glittershark <grfn@gws.fyi>
Make Store::buildPaths return a Status with [[nodiscard]] rather than
throwing exceptions to signal failure. This is the beginning of a long
road to refactor the entire store API to be status/statusor based
instead of using exceptions.
Change-Id: I2e32371c95a25b87ad129987c217d49c6d6e0c85
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1745
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
We have decided that leaking memory is a better fate than random,
non-debuggable memory corruption. Future CLs will begin changing
various fields to std::unique_ptr and std::shared_ptr.
It turns out that disabling the GC does not have disasterous impact.
The Nix evaluator only runs on the client CLI, never in any long-
running process. Even the REPL does not leak too badly under this
change, because it uses one EvalState for the duration of the REPL.
Building an explicitly tracing garbage collector is likely in the
future of this project, but that giant amount of work cannot be
done under a nix evaluator that is constantly crashing. We need to
restore development velocity here, and this is the best way we've
figured out to do it.
Change-Id: I2fcda8fcee853c15a9a5e22eca7c5a784bc2bf76
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1720
Reviewed-by: glittershark <grfn@gws.fyi>
Reviewed-by: tazjin <mail@tazj.in>
Tested-by: BuildkiteCI
The parser contained vectors, and the primary parser state, that
were not participating in GC tracing.
Change-Id: Ie198592cd7acffd390e3e2ae9595138b56416838
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1706
Tested-by: BuildkiteCI
Reviewed-by: tazjin <mail@tazj.in>
Reviewed-by: glittershark <grfn@gws.fyi>
Invocations of the MakeError macro that were not followed by a
semicolon messed up indentation in the next lines.
Change-Id: I03d7d1443f062a38af2c7da3da8928e0ed05e274
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1708
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Aded a few test cases covering the scanForReferences function, which had
been accidentally broken in 976a36c (which is now partially-reverted).
As part of this, since the test needed to generate hashes for store
paths, the logic in MakeStorePath to compress a sha256 hash down to 20
bytes and convert it to base32 has been extracted to a member function
on the Hash class.
Fixes: #34
Change-Id: Ie2d914688a80f42d0234d351a7cc0714fd15709e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1698
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
All of the miscellanious structs that make up a nix Value should inherit
from gc, as they contain pointers to GC'ed things as members
Fixes: #42
Change-Id: I057d8c9f7dafbee7de7644ff152c9cba1aa6bc03
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1696
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
Now that we've fully implemented the RPC-based store client, we can get
rid of the UDSRemoteStore, whose only use was connecting to the locally
running nix daemon. The RemoteStore still needs to be around to connect
to remote upstream nix stores over SSH.
Change-Id: I0699819803cbfe966b9a46786f2c927d8e4bf1a2
Reviewed-on: https://cl.tvl.fyi/c/depot/+/1693
Tested-by: BuildkiteCI
Reviewed-by: kanepyork <rikingcoding@gmail.com>
