Without that flag, LLVM generates unaligned memory access instructions, which are not allowed on ARMv5.
For example, the 'hello world' example from `cargo --new` failed with:
```
$ ./hello
Hello, world!
thread 'main' panicked at 'assertion failed: end <= len', src/libcollections/vec.rs:1113
note: Run with `RUST_BACKTRACE=1` for a backtrace.
```
I traced this error back to the following assembler code in `BufWriter::flush_buf`:
```
6f44: e28d0018 add r0, sp, #24
[...]
6f54: e280b005 add fp, r0, #5
[...]
7018: e5cd001c strb r0, [sp, #28]
701c: e1a0082a lsr r0, sl, #16
7020: 03a01001 moveq r1, #1
7024: e5cb0002 strb r0, [fp, #2]
7028: e1cba0b0 strh sl, [fp]
```
Note that `fp` points to `sp + 29`, so the three `str*`-instructions should fill up a 32bit - value at `sp + 28`, which is later used as the value `n` in `Ok(n) => written += n`. This doesn't work on ARMv5 as the `strh` can't write to the unaligned contents of `fp`, so the upper bits of `n` won't get cleared, leading to the assertion failure in Vec::drain.
With `+strict-align`, the code works as expected.
After compiling a project (e.g. libstd, libtest, or librustc) rustbuild needs to
copy over all artifacts into the sysroot of the compiler it's assembling.
Unfortunately rustbuild doesn't know precisely what files to copy! Today it has
a heuristic where it just looks at the most recent version of all files that
look like rlibs/dylibs and copies those over. This unfortunately leads to bugs
with different versions of the same craet as seen in #42261.
This commit updates rustbuild's strategy of copying artifacts to work off the
list of artifacts produced by `cargo build --message-format=json`. The build
system will now parse json messages coming out of Cargo to watch for files being
generated, and then it'll only copy over those precise files.
Note that there's still a bit of weird logic where Cargo prints that it's
creating `libstd.rlib` where we actually want `libstd-xxxxx.rlib`, so we still
do a bit of "most recent file" probing for those. This commit should take care
of the crates.io dependency issues, however, as they're all copied over
precisely.
Closes#42261
rust-src: include everything needed to compile libstd with jemalloc
I am not very happy about all this `Path::new`, but did not find a nice way to avoid it. Also, this shouldn't be very performance-critical.
With this patch, rust-src-1.19.0-dev.tar.gz grows from 1.4 to 3.1 MiB (new uncompressed size: 15.5 MiB). Not great, but shipping incomplete sources is also not great, and this is still much smaller than pre-#41546. Excluding the entire `src/jemalloc/test` does not work, unfortunately; there is a file in there that is needed to build libstd. (And anyway there's just 190 KiB uncompressed left in that folder.)
In principle, we could try excluding the Rust test suite directories (that would be `libcore/tests` and `libcollection/tests`). I don't know enough about how this component is used to judge whether that would cause any problems. Anyway this is just 600 KiB uncompressed.
Fixes#41952
incr.comp.: Remove DepGraph::write() and its callers
After months of yak shaving, we are finally there `:)`
The existence of `DepGraph::write()` was one of the two main ways for introducing cycles into the dep-graph -- something we need to avoid in the future. The other way, re-opening nodes to add more edges, is next on the list.
r? @nikomatsakis
