[kripken]

General idea is wasm-as is a parallel to llvm-as, and so forth. But no strong feelings about names, and perhaps anyhow all tools should be able to load both binary and text eventually.

This just adds the binary writer, and checks that it compiles.

[jfbastien]

2016 now.

[kripken]

fixed

[jfbastien]

Could you add the following to check.py:

import test.waterfall.src.link_assembly_files as link_assembly_files
wasm_as_torture_out = os.path.abspath(os.path.join('test', 'wasm-as-torture-out'))
if os.path.isdir(wasm_as_torture_out):
  shutil.rmtree(wasm_as_torture_out)
os.mkdir(wasm_as_torture_out)
unexpected_result_count = link_assembly_files.run(
    linker=os.path.abspath(os.path.join('bin', 'wasm-as')),
    files=os.path.abspath(os.path.join('test', 'torture-s', '*.s')),
    fails=os.path.abspath(os.path.join('test', 'wasm_as_known_gcc_test_failures.txt')),
    out=wasm_as_torture_out)
assert os.path.isdir(wasm_as_torture_out), 'Expected output directory %s' % wasm_as_torture_out
shutil.rmtree(wasm_as_torture_out)
if unexpected_result_count:
  fail(unexpected_result_count, 0)

It'll also require a new test/wasm_as_known_gcc_test_failures.txt file, and an update of the test/waterfall submodule.

[kripken]

Wait, what does that code do? Run wasm-as on stuff? It's not ready yet! :)

[jfbastien]

Not even mildly ready to fail on most-but-not-quite-all tests? :p

[kripken]

No, it aborts on "hello world" :)

I need to do some other refactoring. This pull is mainly to finish getting it building, as I don't want to have a long-term side branch.

[jfbastien]

Ah OK, lgtm then :)

[ghost]

Just out of curiosity, the llvm backend currently emits a .s file in a format that is not vaguely like wast. This tools is describe as 'Assemble a .wast (WebAssembly text format) into a .wasm (WebAssembly binary format)' and if so then the wasm-as naming looks confusing. Is it planned that the current .s format becomes the text format and superseded wast or might there be a separate .s format?

[jfbastien]

"Link and assemble a .s into a .wasm"

[jfbastien]

Just out of curiosity, the llvm backend currently emits a .s file in a format that is not vaguely like wast. This tools is describe as 'Assemble a .wast (WebAssembly text format) into a .wasm (WebAssembly binary format)' and if so then the wasm-as naming looks confusing.

It consumes a .s file, the help is wrong but I hadn't noticed. Added a comment.

Is it planned that the current .s format becomes the text format and superseded wast or might there be a separate .s format?

The .s format is just convenient for LLVM as it is now, because that got things off the ground faster without introducing alien concepts into LLVM (which the LLVM community may frown upon). It's meant to look similar to .s files for other architectures, LLVM makes that easy. @sunfishcode is experimenting with outputting binary directly from LLVM, which is more involved but then you get assembler and disassembler for free if you Do It The Right Way™.

The plans for text format are still very up in the air, and the current s-expression format isn't agreed on either. .s seems less readable than .wast so I doubt we'll standardize it as The One True Text Format™, but it may be useful to standardize between compilers, the same way standardizing .o / .a / .so files between compilers could be useful: they're not meant to get shipped to a browser, but tools can get along and that's nice.

[kripken]

Actually no, the text as it stands is correct. This tool assembles a .wast (s-expression wasm text format, and eventually the official wasm text format whatever that will be) into a .wasm (binary wasm format, in current v8 proposal).

In other words:

• wasm-as: .wast => .wasm
• wasm-dis: .wasm => .wast - not yet implemented

These are parallel to llvm-as, llvm-dis. They translate between the text and binary formats of the same thing, either LLVM IR or wasm AST.

Other tools convert to/from external or unofficial formats, like

• asm2wasm: .asm.js = .wasm (well, right now it emits text, but that should change)
• s2wasm: LLVM .s => .wasm (again, right now it emits text)
• etc.

A sad point of confusion is that .s is LLVM's .s format, while wasm-s-parser stands for wasm-s-expression-parser, which might trip people up. Suggestions for a better name?

[jfbastien]

@kripken ah right, I double-misread :-)

@binji was using sexpr, but in general wasm-as seems intuitive. So would wast2wasm and similar.

[binji]

@jfbastien I named it that before .wast existed :)