Add new compound Signature, Struct and Array types #3012

dcodeIO · 2020-08-01T14:19:32Z

Introduces TypeInfo as a common descriptor to represent compound Signature, Struct and Array types in addition to the existing Tuple types and refactors internal maps to use TypeInfo as their key.

src/support/hash.h

dcodeIO

Some of my thoughts about the changes made to provide some context.

src/wasm-type.h

src/wasm/wasm-type.cpp

src/wasm-type.h

tlively

Looking good so far!

src/wasm-type.h

src/wasm/wasm-type.cpp

src/wasm-type.h

…c-type-refactor

test/example/typedef.txt

This leads to simpler code and is a prerequisite for #3012, which makes it so that not all `Type`s are backed by vectors that `expand` could return.

Co-authored-by: Thomas Lively <7121787+tlively@users.noreply.github.com>

src/wasm/wasm-type.cpp

tlively

I think this should be the last round of comments :)

src/wasm-type.h

test/example/typedef.cpp

src/wasm-type.h

…ement hashing of all the public things

src/wasm-type.h

dcodeIO · 2020-08-21T12:36:35Z

src/wasm-type.h

@@ -165,12 +165,12 @@ class Type {
  static Type get(unsigned byteSize, bool float_);

  // Returns true if left is a subtype of right. Subtype includes itself.
-  static bool isSubType(Type left, Type right);
+  static bool isSubType(const Type& left, const Type& right);


Since Types are small anyway we are either pushing its uintptr_t id or an equally sized reference, so that's mostly style, right?

Side note: binaryen master is still in a weird state where we've changed anyref to externref, but the other reference types changes haven't yet been applied; in this case, there is no longer any subtyping relationship, so I assume this method needs to be refactored out as part of removing nullref, adding an immediate type tag to ref.null, etc.

Apologies if this causes issues for folks, I only had spare cycles to do the initial work.

Thanks for the reminder. Plan forward might be:

Complete this PR so we have a foundation for compound types

Remove nullref, add eqref, i31ref and the new anyref alongside initial handling of the new compound types

Think about RTTs

Update ref.null and ref.is_null alongside adding the new GC instructions

That seems good to me. Apologies if it was already top of mind 🤡

cc @aheejin, who I think was planning to make some changes to reflect the current state of reference types. It would be good to coordinate with her on who is going to make what changes.

On topic here, I think it would be best to keep passing the types directly here. The extra indirection of passing a reference (which means the values must be in memory rather than in registers) is unnecessary. The only reason we use const Type& for operator== and friends is that those are the standard kinds of types used for those operator overloads.

I have a half-baked (or, more precisely, quarter-baked) patch for bringing reference types proposal up to date in my local machine, which I started some weeks ago and didn't get around to finish. I'm now trying to pick up the patch.. Do you need that changes for GC? I don't want to block you from proceeding to wait for my patch, but at the same time, I think it is good to add that reference type patch separately from GC and make sure all existing (and to-be-updated) tests the fuzzer pass with reference types proposal, because currently the fuzzer is testing reference type instructions already.

I was away for most of the last week. I was gonna make this change this week, but please let me know if your plan is blocked by this or interfere with this schedule, in which case we can change the plan. I wasn't able to take a look at this patch yet, but I think I'll soon.

The GC proposal restores anyref and nullref, so all we really need to do right now is add an externref type. But some current uses of anyref might need to use externref instead. @aheejin, you'll have a better idea than me whether it's better to rename the current anyref to externref then add a new anyref as part of the GC work or to just add a new externref to what we already have.

I'm confused.

Does the GC proposal really restore nullref? After the recent reference type changes, every reference type can be nullified, or in other words, have a null value. Why would we need to restore a separate nullref type? I think what you quoted is just not up-to-date, no?

I think we should add anyref; anyref and externref seem to be two separate types. anyref is a supertype of all reftypes while externref is not.

Iiuc there is no nullref anymore.

Aha, I've been looking at the wrong GC document. Thanks for the clarification, @dcodeIO!

src/wasm-type.h

src/wasm/wasm-type.cpp

test/example/typeinfo.cpp

tlively

Looks good to me with those small changes. Let me know if you agree that it's ready to land :)

src/wasm-type.h

Co-authored-by: Thomas Lively <7121787+tlively@users.noreply.github.com>

dcodeIO · 2020-08-24T10:34:08Z

This should be consistent in itself even though it doesn't yet address recursive types (probably either mutable compound types one can populate before canonicalization or type pointers) or RTTs (probably a new kind of compound type), but yeah, should be good to merge as a basis for further work :)

Invocations so far:
   FuzzExec: 1125
   CompareVMs: 301
   CheckDeterminism: 78
   Wasm2JS: 294
   Asyncify: 299

Extends compound types introduced in #3012 with a representation of `Rtt`s as described in the GC proposal, by also introducing the concept of a `HeapType` shared between `TypeInfo` and `Rtt`. Again, this should be a non-functional change since `Rtt`s are not used anywhere yet. Subtyping rules and updating the `xref` aliases is left for future work.

dcodeIO added 3 commits August 1, 2020 16:01

Refactor types to represent arbitrary complex types

927bd6d

fix

e84664f

fix

185f887

dcodeIO force-pushed the gc-type-refactor branch from ac5eedd to 185f887 Compare August 1, 2020 14:39

MaxGraey reviewed Aug 1, 2020

View reviewed changes

src/support/hash.h Outdated Show resolved Hide resolved

dcodeIO added 4 commits August 1, 2020 19:16

I know nothing

415353f

layers

941bde4

unify printing

37eb81f

tuple terminology, Tuple::toString via wrapping

f4067bd

dcodeIO commented Aug 2, 2020

View reviewed changes

src/wasm-type.h Outdated Show resolved Hide resolved

src/wasm-type.h Outdated Show resolved Hide resolved

src/wasm-type.h Outdated Show resolved Hide resolved

src/wasm/wasm-type.cpp Outdated Show resolved Hide resolved

dcodeIO added 4 commits August 2, 2020 18:11

make nullable a conditional property of specific TypeDefs

5146da2

remove now unused function parameters

7299028

move 'null' to TypeDef when printing

a8dce05

ValueType -> ID, efficient lookups

e006f21

dcodeIO changed the title ~~Refactor types to represent arbitrarily complex types~~ Refactor types to represent various constructed types Aug 2, 2020

dcodeIO commented Aug 2, 2020

View reviewed changes

src/wasm-type.h Outdated Show resolved Hide resolved

dcodeIO commented Aug 2, 2020

View reviewed changes

src/wasm-type.h Outdated Show resolved Hide resolved

dcodeIO commented Aug 2, 2020

View reviewed changes

src/wasm-type.h Show resolved Hide resolved

dcodeIO mentioned this pull request Aug 3, 2020

Prototype GC instructions #2935

Closed

tlively reviewed Aug 3, 2020

View reviewed changes

binji reviewed Aug 3, 2020

View reviewed changes

src/wasm-type.h Outdated Show resolved Hide resolved

kripken mentioned this pull request Aug 3, 2020

Refactor getMaxBits() out of OptimizeInstructions and add beginnings of unit testing for it #3019

Merged

tlively and others added 8 commits August 3, 2020 14:38

Extract kind

a41c3d5

rename Type::ID to Type::BasicID

2764af7

rename ID -> BasicID in other comments

8e159d1

avoid using default cases

f1a25a4

Merge remote-tracking branch 'tlively/dcodeIO-gc-extract-kind' into g…

1e901fc

…c-type-refactor

fix, remove getKind

34c2b55

pass Signature by value

b4ffb98

packed field types

12183ec

parens it is

d92e416

dcodeIO commented Aug 18, 2020

View reviewed changes

test/example/typedef.txt Outdated Show resolved Hide resolved

dcodeIO mentioned this pull request Aug 19, 2020

Replace Type::expand() with an iterator-based approach #3061

Merged

tlively mentioned this pull request Aug 19, 2020

Recursive Types #3063

Closed

tlively pushed a commit that referenced this pull request Aug 19, 2020

Replace Type::expand() with an iterator-based approach (#3061)

a18d30f

This leads to simpler code and is a prerequisite for #3012, which makes it so that not all `Type`s are backed by vectors that `expand` could return.

dcodeIO and others added 4 commits August 20, 2020 09:29

Merge branch 'master' into gc-type-refactor

e8e111d

Update src/wasm-type.h

9517da1

Co-authored-by: Thomas Lively <7121787+tlively@users.noreply.github.com>

add an assert

c20b61f

revert Tuple move constructor until expert feedback

f6b9f83

dcodeIO commented Aug 20, 2020

View reviewed changes

src/wasm/wasm-type.cpp Outdated Show resolved Hide resolved

tlively reviewed Aug 20, 2020

View reviewed changes

src/wasm/wasm-type.cpp Outdated Show resolved Hide resolved

src/wasm/wasm-type.cpp Outdated Show resolved Hide resolved

remove unnecessary this reference

0de9d01

dcodeIO mentioned this pull request Aug 20, 2020

Use const modifier when dealing with types #3064

Merged

Merge branch 'master' into gc-type-refactor

0cdc596

tlively reviewed Aug 20, 2020

View reviewed changes

src/wasm-type.h Outdated Show resolved Hide resolved

test/example/typedef.cpp Outdated Show resolved Hide resolved

src/wasm-type.h Outdated Show resolved Hide resolved

dcodeIO added 2 commits August 21, 2020 10:18

move remains of TypeDef to wasm-type.cpp, rename to TypeInfo and impl…

85fb78f

…ement hashing of all the public things

apply dangerous half-knowledge

cecf38a

dcodeIO commented Aug 21, 2020

View reviewed changes

interesting tests

fa620d9

dcodeIO commented Aug 21, 2020

View reviewed changes

test/example/typeinfo.cpp Show resolved Hide resolved

revert dangerous half-knowledge

9d206af

tlively approved these changes Aug 24, 2020

View reviewed changes

src/wasm-type.h Outdated Show resolved Hide resolved

src/wasm-type.h Outdated Show resolved Hide resolved

dcodeIO and others added 3 commits August 24, 2020 11:22

Update src/wasm-type.h

b8d6faf

Co-authored-by: Thomas Lively <7121787+tlively@users.noreply.github.com>

Update src/wasm-type.h

fd81477

Co-authored-by: Thomas Lively <7121787+tlively@users.noreply.github.com>

Merge branch 'master' into gc-type-refactor

5d617c9

tlively merged commit fa43b50 into WebAssembly:master Aug 24, 2020

dcodeIO mentioned this pull request Aug 26, 2020

Add new compound Rtt type #3076

Merged

aheejin mentioned this pull request Aug 31, 2020

Update reference types #3084

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add new compound Signature, Struct and Array types #3012

Add new compound Signature, Struct and Array types #3012

dcodeIO commented Aug 1, 2020 •

edited

Loading

dcodeIO left a comment

tlively left a comment

tlively left a comment

dcodeIO Aug 21, 2020

jayphelps Aug 22, 2020

dcodeIO Aug 22, 2020

jayphelps Aug 22, 2020

tlively Aug 23, 2020

aheejin Aug 24, 2020

tlively Aug 24, 2020 •

edited

Loading

aheejin Aug 24, 2020 •

edited

Loading

dcodeIO Aug 24, 2020

tlively Aug 24, 2020

tlively left a comment

dcodeIO commented Aug 24, 2020

Add new compound Signature, Struct and Array types #3012

Add new compound Signature, Struct and Array types #3012

Conversation

dcodeIO commented Aug 1, 2020 • edited Loading

dcodeIO left a comment

Choose a reason for hiding this comment

tlively left a comment

Choose a reason for hiding this comment

tlively left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tlively Aug 24, 2020 • edited Loading

Choose a reason for hiding this comment

aheejin Aug 24, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tlively left a comment

Choose a reason for hiding this comment

dcodeIO commented Aug 24, 2020

dcodeIO commented Aug 1, 2020 •

edited

Loading

tlively Aug 24, 2020 •

edited

Loading

aheejin Aug 24, 2020 •

edited

Loading