Add initial support for ppc64

[luke-jr]

Note, I couldn't get a complete build to work... Got as far as (95/96) BUILDING cargo v0.30.0 though.

[RUSTC] -o output/prefix-s/lib/rustlib/powerpc64le-unknown-linux-gnu/lib/libcore.rlib
make: *** [Makefile:120: output/prefix-s/lib/rustlib/powerpc64le-unknown-linux-gnu/lib/libcore.rlib] Segmentation fault

[r-burns]

It looks like the issue you ran into was not architecture-specific and has been fixed since you opened this. I was able to apply this patch on top of db097f7 and bootstrap rustc 1.29 on powerpc64le-linux (using a workaround in #187).

[luke-jr]

All of -C run_rustc completes now! But the resulting toolchain appears to be unable to build rustc 1.29.1, so something is still wrong :/

The build is looking for _ZN4core3num19ASCII_LOWERCASE_MAP17hf5d18b4e3429b9aeE, but the libstd built has _ZN4core3num19ASCII_LOWERCASE_MAP17hf935b1a9ae318099E instead. Not sure what the random value at the end is supposed to be, so kinda dead in the water again.

[luke-jr]

   Compiling bootstrap v0.0.0 (file:///var/tmp/portage/dev-lang/rust-1.29.1_p20210830/work/rustc-1.29.1-src/src/bootstrap)
error: /var/tmp/portage/dev-lang/rust-1.29.1_p20210830/work/rustc-1.29.1-src/build/bootstrap/debug/deps/libserde_derive-b505111b61dd22f0.so: undefined symbol: _ZN4core3num19ASCII_LOWERCASE_MAP17hf5d18b4e3429b9aeE
   --> bootstrap/lib.rs:124:1
    |
124 | extern crate serde_derive;
    | ^^^^^^^^^^^^^^^^^^^^^^^^^^

error: aborting due to previous error

error: Could not compile `bootstrap`.

[thepowersgang]

That random number is the symbol hash, it's derived from a whole host of things. I've seen it be different before (I think as part of 1.39), and traced it to a (now fixed) mrustc bug. Is your branch based on a recent master?

[luke-jr]

I was trying to reproduce @r-burns's success, so based my work on db097f7. I think I was intimidated by master, since it seems to have a lot of incomplete work toward newer versions. I can try it, though - what target is known to work currently (the newer the better)?

[thepowersgang]

1.39 is still the highest fully supported version.

[r-burns]

FWIW, I built mrustc as a nix package, so it was using all of the deterministic build mechanisms enforced by nixpkgs stdenv such as fixed -frandom-seed and SOURCE_DATE_EPOCH. I don't know anything about how that random value is derived, so not sure if this is relevant.

[thepowersgang]

It's not random, it's a stable hash of the crate ID and the (incremental-safe) definition ID.
I think the original issue I saw was due to not using enum niche optimisations where they were expected.

[luke-jr]

Went to try master w/ 1.39 and realised I did already (ffb0961 to be specific).

It failed pretty early with:

Process was terminated with signal 7
FAILING COMMAND:  /var/tmp/portage/dev-lang/rust-1.40.0_p20211027/work/rustc-1.40.0-src/mrustc/output/build_compiler_builtins-0_1_18_H22_run > /var/tmp/portage/dev-lang/rust-1.40.0_p20211027/work/rustc-1.40.0-src/mrustc/output/build_libc-0_2_62_Hd_run

(note that the inevitable target of this package would be 1.40.0, built using the mrustc-built 1.39.0 - it never got that far; but I am not trying to build 1.40.0 with mrustc here)

[thepowersgang]

SIGBUS - that's odd, does make test (which runs a basic "hello, world") work?

[luke-jr]

This was during RUSTC_VERSION=1.39.0 MRUSTC_TARGET_VER=1.39 OUTDIR_SUF=-1.39.0 make -f minicargo.mk LIBS, so long before hello world would be applicable?

[luke-jr]

FWIW, playing around, I tried 953b8a9 w/ 1.54.0 and got that same error

[thepowersgang]

I guess you'll have to do some debugging (e.g. run the relevant program in gdb) to find the codegen error.

[luke-jr]

Moved that topic to #194