By doing so, rustdoc will also emit workspace-relative filenames for the doctests.
This was first landed in #8954 but later backed out in #8996 because it changed the CWD of rustdoc test invocations.
The second try relies on the new `--test-run-directory` rustdoc option which was added in https://github.com/rust-lang/rust/pull/81264 to explicitly control the rustdoc test cwd.
fixes#8993
This commit fixes an issue with LTO calculation for various units when
`lto=off` is specified in the profile. This ensures now that `lto=off`
is passed to all transitive dependencies as well to disable thin-local
LTO. As an added bonus this also passed `embed-bitcode=no` whenever
`lto=off` is specified since we know we won't be using bitcode anyway.
Closes#9171
This commit updates Cargo's build of host dependencies to build them
with optimization level 0 by default instead of matching the profile of
the final binary.
Since Cargo's inception build dependencies have, by default, been built
in a profile that largely matches the profile of the final target
artifact. Build dependencies, however, rarely actually need to be
optimized and are often executing very small tasks, which means that
optimizing them often wastes a lot of build time. A great example of
this is procedural macros where `syn` and friends are pretty heavyweight
to optimize, and the amount of Rust code they're parsing is typically
quite small, so the time spent optimizing rarely comes as a benefit.
The goal of this PR is to improve build times on average in the
community by not spending time optimizing build dependencies (build
scripts, procedural macros, and their transitive dependencies). The PR
will not be a universal win for everyone, however. There's some
situations where your build time may actually increase:
* In some cases build scripts and procedural macros can take quite a
long time to run!
* Cargo may not build dependencies more than once if they're shared with
the main build. This only applies to builds without `--target` where
the same crate is used in the final binary as in a build script.
In these cases, however, the `build-override` profile has existed for
some time know and allows giving a knob to tweak this behavior. For
example to get back the previous build behavior of Cargo you would
specify, in `Cargo.toml`:
[profile.release.build-override]
opt-level = 3
or you can configure this via the environment:
export CARGO_PROFILE_RELEASE_BUILD_OVERRIDE_OPT_LEVEL=3
There are two notable features we would like to add in the future which
would make the impact of a change like this smaller, but they're not
implemented at this time (nor do we have concrete plans to implement
them). First we would like crates to have a way of specifying they
should be optimized by default, despite default profile options. Often
crates, like lalrpop historically, have abysmal performance in debug
mode and almost always (even in debug builds) want to be built in
release mode. The second feature is that ideally crate authors would be
able to tell Cargo to minimize the number of crates built, unifying
profiles where possible to avoid double-compiling crates.
At this time though the Cargo team feels that the benefit of changing
the defaults is well worth this change. Neither today nor directly after
this change will be a perfect world, but it's hoped that this change
makes things less bad!
In the case that LTO is happening but we're also generating a
cdylib/rlib simultatneously that means that the final artifact will use
the rlib but the cdylib still needs to be produced. To get this to work
the cdylib's dependency tree needs to be compiled with embedded bitcode.
The cdylib itself will be linked with the linker because we can't LTO an
rlib+cdylib compilation, but the final executable will need to load
bitcode from all its deps.
This is meant to address rust-lang/rust#72268
Turns out a case I thought didn't happen can indeed happen. Units may
depend on other units which are LTO-able because integration tests can
depend on binaries. Handle that case internally and remove a few panics.
Closes#8223
This commit updates Cargo's implementation of LTO builds to do less work
and hopefully be faster when doing a cold build. Additionaly this should
save space on disk! The general idea is that the compiler does not need
object files if it's only going to perform LTO with some artifacts. In
this case all rustc needs to do is load bitcode from dependencies. This
means that if you're doing an LTO build generating object code for
intermediate dependencies is just wasted time!
Here Cargo is updated with more intrusive knowledge about LTO. Cargo
will now analyze the dependency graph to figure out which crates are
being compiled with LTO, and then it will figure out which dependencies
only need to have bitcode in them. Pure-bitcode artifacts are emitted
with the `-Clinker-plugin-lto` flag. Some artifacts are still used in
multiple scenarios (such as those shared between build scripts and final
artifacts), so those are not compiled with `-Clinker-plugin-lto` since
the linker is not guaranteed to know how to perform LTO.
This functionality was recently implemented in rust-lang/rust#71528
where rustc is now capable of reading bitcode from `-Clinker-plugin-lto`
rlibs. Previously rustc would only read its own format of bitcode, but
this has now been extended! This support is now on nightly, hence this
PR.
This commit is the Cargo half of support necessary for
rust-lang/rust#70458. Today the compiler emits embedded bytecode in
rlibs by default, but compresses it. This is both extraneous disk space
and wasted build time for almost all builds, so the PR in question there
is changing rustc to have a `-Cembed-bitcode` flag which, when enabled,
places the bitcode in the object file rather than an auxiliary file (no
extra compression), but also enables `-Cembed-bitcode=no` to disable
bitcode emission entirely.
This Cargo support changes Cargo to pass `-Cembed-bitcode=no` for almost
all compilations. Cargo will keep `lto = true` and such working by not
passing this flag (and thus allowing bitcode to get embedded), but by
default `cargo build` and `cargo build --release` will no longer have
any bitcode in rlibs which should result in speedier builds!
Most of the changes here were around the test suite and various
assertions about the `rustc` command lines we spit out. One test was
hard-disabled until we can get `-Cembed-bitcode=no` into nightly, and
then we can make it a nightly-only test. The test will then be stable
again once `-Cembed-bitcode=no` hits stable.
Note that this is intended to land before the upstream `-Cembed-bitcode`
change. The thinking is that we'll land everything in rust-lang/rust all
at once so there's no build time regressions for anyone. If we were to
land the `-Cembed-bitcode` PR first then there would be a build time
regression until we land Cargo changes because rustc would be emitting
uncompressed bitcode by default and Cargo wouldn't be turning it off.
