Hi, Here are two things that might help you get faster builds.
TL;DR: 1. On Linux, make sure you have lld installed, because it's a *much* faster linker than gold, and it's now used by default if installed. 2. Upgrade your sccache to version 0.2.7 to get faster rebuilds of changed Rust files. DETAILS: 1. lld was made the default linker on Linux (when installed) in https://bugzilla.mozilla.org/show_bug.cgi?id=1473436. Thank you, glandium! If you don't already have it installed, run `mach bootstrap` to get a copy in ~/.mozbuild/clang/bin/, which you can then add to your PATH. Alternatively `mach artifact toolchain --from-build linux64-clang` will download lld (and other clang tools) under the current directory. To confirm that lld is being used, look for this output from configure: checking for linker... lld On other platforms, lld is not yet the default but you *might* be able to use it. On Windows, again use `mach bootstrap` and add this to your mozconfig: export LINKER=lld-link This used to cause problems with debugging (bug 1458109) but that has since been fixed. On Mac, if you have lld installed, add this to your mozconfig: export LDFLAGS=-fuse-ld=lld but it might cause build errors, such as "No rule to make target `libmozavutil_dylib.list', needed by `libmozavutil.dylib`". https://bugzilla.mozilla.org/show_bug.cgi?id=1384434 is the tracking bug for making lld the default on all platforms. Any bugs filed about problems should block that bug. Also, if you are building a Rust-only project, something like this might work (on Linux; I'm not sure about other builds): RUSTFLAGS="-Clinker=clang -Clink-arg=-fuse-ld=lld" cargo build See https://github.com/rust-lang/rust/issues/50584#issuecomment-398988026 for an example improvement. https://github.com/rust-lang/rust/issues/39915 is the issue for making lld the default linker for Rust. On non-Linux platforms, I recommend testing with the resulting builds before you place full confidence in them. 2. When the Rust compiler is invoked with incremental compilation (which happens in any Firefox build that doesn't have --enable-release) sccache 0.2.7 will skip the file, which is good because incremental compilation has much the same effect, and sccache used to cause significant slowdowns in the cases where a cache miss occurred. Thank you to ted for this improvement! For example, on my fast Linux box, if I `touch` servo/components/style/ stylist.rs and rebuild (resulting in an sccache cache hit), I get these times: - old sccache disabled: 28s - old sccache enabled: 25s - new sccache enabled: 28s I.e. sccache's new behaviour causes a tiny slowdown. But if I insert a comment at the top of that file (resulting in an sccache cache miss), I get these times: - old sccache disabled: 37s - old sccache enabled: 1m53s(!) - new sccache enabled: 37s I.e. sccache's new behaviour is a big win. And this "make a small change and recompile" case is extremely common. See https://github.com/mozilla/sccache/issues/257 for more details. Nick _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
