runtime-c-api tests fail with singlepass-backend

[ethanfrey]

Describe the bug

echo "`wasmer -V` | `rustc -V` | `uname -m`"

wasmer: command not found
 | rustc 1.38.0-nightly (c43d03a19 2019-08-14) | x86_64

git log -1 provides:

commit 6d351b514ccc15fa2f344de4ed68506a6d3baa23 (HEAD -> master)
Merge: 569d0386 91b199fe
Author: bors[bot] <26634292+bors[bot]@users.noreply.github.com>
Date:   Thu Aug 29 20:48:53 2019 +0000

---

I realized this was an old commit.

I then updated to current master, fixed a typo, and while cargo test passes,
cargo test --features singlepass-backend --no-default-features returns a linking error:

   Compiling wasmer-runtime-c-api v0.7.0 (/home/ethan/IdeaProjects/wasmer/lib/runtime-c-api)
error: linking with `cc` failed: exit code: 1
  |
...
note: /usr/bin/ld: /home/ethan/IdeaProjects/wasmer/target/debug/deps/libwasmer_runtime_core-dcfea6f3cbfc5444.rlib(image-loading-linux-x86-64.o): relocation R_X86_64_PC32 against symbol `get_boundary_register_preservation' can not be used when making a shared object; recompile with -fPIC
          /usr/bin/ld: final link failed: Bad value
          collect2: error: ld returned 1 exit status

from:

commit 0ea5b1b2bd87a704e2963bb92b9c0d96e2bf66ba (HEAD -> master, origin/staging, origin/master)
Merge: 174cc9f0 71b865aa
Author: bors[bot] <26634292+bors[bot]@users.noreply.github.com>
Date:   Wed Sep 18 16:59:37 2019 +0000

Steps to reproduce

cd lib/c-runtime-api
cargo test --features singlepass-backend --no-default-features

fails on current master

Note that I had to use --no-default-features flags so this compiles at all.

Expected behavior

Tests pass, as with cranelift or llvm

Actual behavior

Result:

test 11
      Start 11: test-module-serialize

11: Test command: /home/ethan/IdeaProjects/wasmer/lib/runtime-c-api/tests/test-module-serialize
11: Test timeout computed to be: 9.99988e+06
11: test-module-serialize: /home/ethan/IdeaProjects/wasmer/lib/runtime-c-api/tests/test-module-serialize.c:25: main: Assertion `serialize_result == WASMER_OK' failed.
11/15 Test #11: test-module-serialize ............***Exception: Child aborted  0.17 sec

Additional context

[Hywan]

Thanks for the bug report!

@losfair Can you take a look at this bug please?

[ethanfrey]

I had a long discussion on this with @nlewycky outside of github. And resolved some parts of this.

The linkage error for single-pass has been fixed with #814 I don't really understand the meaning of the change, but Nick debugged it and I verified it worked on my machine. (He had a different, unsupported, version of ld and it failed with a different error earlier).

The other error was tracked down to the fact that singlepass compiler never had support for serialization. This just was never tested before. In the end, I managed to get out this message (why test-module-serialize fails with singlepass backend): the singlepass backend doesn't support caching yet.

And found the smoking gun:

wasmer/lib/singlepass-backend/src/codegen_x64.rs

Line 464 in 98d3e04

"the singlepass backend doesn't support caching yet".to_string(),

Handling this - either implementing it or documenting it as the wasmer team decides - can be another issue when you figure out your approach. I would be fine with closing this when #814 is merged, and linking to the new focused issue on how to handle the now known serialization issue.

[syrusakbary]

We have now caching in singlepass. I'll do a quick check to see if the runtime-c-api tests pass with it :)

[ethanfrey]

Amazing. Can you link the pr that added caching to singlepass? I'm definitely excited about the 0.12 release now.

[syrusakbary]

Here's the PR adding caching to singlepass: #1022

We also just merged this PR which tests the capi with all the backends (singlepass, cranelift and llvm): #1060

So we should be good closing this issue now since caching is implemented for singlepass and the capi works with it!