Extract common code from runtime implementations. #152

mathetake · 2021-04-12T08:38:12Z

fixes #148

Signed-off-by: Takeshi Yoneda [email protected]

~~still wip since I haven't tested on Envoy yet~~

Signed-off-by: Takeshi Yoneda <[email protected]>

src/common/wasm_util.cc

src/common/wasm_util.h

src/v8/v8.cc

Signed-off-by: Takeshi Yoneda <[email protected]>

mathetake · 2021-04-12T13:55:08Z

src/v8/v8.cc

+    wasm::vec<byte_t> code_vec = wasm::vec<byte_t>::invalid();
+    if (stripped.empty()) {
+      // Use the original bytecode.
+      code_vec = wasm::vec<byte_t>::make(code.size(), (char *)(code.data()));


(char *)(code.data()) is a temporary workaround for now to pass build and we should fix. This is necessary since wasm::vec<byte_t>::make requires us to pass char vector instead of const char vector. If we want to do zero copy, we need to do cast like this or change the arg type of code to std::string&.. but I'm not sure why ::make require non-const bytes code since V8 seems to copy the code internally anyway..

@PiotrSikora I need your inputs here.. Should we change the signature of load to std::string &code from const std::string &code so that we can have char * instead of const char* from code.data(), or we allow (char *)(code.data())..?

The following is the error we get without modifying signature and just pass code.data()

external/com_googlesource_chromium_v8/wee8/third_party/wasm-api/wasm.hh:149:18: error: cannot initialize an array element of type 'char' with an rvalue of type 'typename std::remove_reference<const char *&>::type' (aka 'const char *') T data[] = { std::move(args)... }; ^~~~~~~~~~~~~~~ external/proxy_wasm_cpp_host/src/v8/v8.cc:293:37: note: in instantiation of function template specialization 'wasm::vec<char>::make<unsigned long, const char *>' requested here code_vec = wasm::vec<byte_t>::make(code.size(), code.data());

I think this might be the reason why we originally copied the bytecode. Perhaps make getStrippedSource always return std::string (either stripped version or a copy of the original bytecode) and then use it as input here?

so we cannot avoid copying the original bytecode anyway.. thanks for suggestion.

done ed0f3a5

The bytecode read from the configuration is passed to both: canary and base WasmVMs at startup, so I'm afraid that we have to copy it, unfortunately.

Perhaps we can rewrite this logic somehow, but that should probably be done in a separate PR.

mathetake · 2021-04-12T13:57:14Z

made ready for review at least now this passes Envoy tests

Signed-off-by: Takeshi Yoneda <[email protected]>

PiotrSikora

Thanks for working on this!

Wavm is an odd duck now. We're using its internal parsing to extract precompiled module in Wavm::load, but replaced Wavm::getCustomSection with WasmUtil. I think we should either replace the iterator in Wavm::load with WasmUtil::getCustomSection or retain getCustomSection as part of the interface, and use WasmUtil-based default, so that Wavm could override it.
Could you also extact code responsible for mapping ABI versions? Right now, we do the same string_view to AbiVersion conversion in each runtime.

src/common/wasm_util.cc

src/v8/v8.cc

src/common/wasm_util.h

Signed-off-by: Takeshi Yoneda <[email protected]>

mathetake · 2021-04-13T01:01:45Z

changed to use BytecodeUtil::getCustomSection in WAVM: d807f06

mathetake · 2021-04-13T01:07:53Z

OK verified that WAVM change also passes Envoy tests (locally😄).

Signed-off-by: Takeshi Yoneda <[email protected]>

mathetake · 2021-04-13T02:44:02Z

added ABI version extraction in BytecodeUtil and use it in getAbiVersion of runtimes except NullVm 305d48d 41a9ac6 ,

Signed-off-by: Takeshi Yoneda <[email protected]>

src/common/bytecode_util.h

src/common/bytecode_util.cc

PiotrSikora · 2021-04-13T05:29:44Z

src/v8/v8.cc

+    wasm::vec<byte_t> code_vec = wasm::vec<byte_t>::invalid();
+    if (stripped.empty()) {
+      // Use the original bytecode.
+      code_vec = wasm::vec<byte_t>::make(code.size(), (char *)(code.data()));


I think this might be the reason why we originally copied the bytecode. Perhaps make getStrippedSource always return std::string (either stripped version or a copy of the original bytecode) and then use it as input here?

src/v8/v8.cc

src/common/bytecode_util.cc

Signed-off-by: Takeshi Yoneda <[email protected]>

PiotrSikora

Thanks! Could you update Envoy PR to make sure it passes with latest changes?

mathetake · 2021-04-13T08:05:53Z

Thanks! Once the test passes in Envoy, I will merge this and update Envoy to point to the main branch here.

PiotrSikora · 2021-04-13T08:13:04Z

Thanks! Once the test passes in Envoy, I will merge this and update Envoy to point to the main branch here.

We probably shouldn't update every single commit in Enovy, especially since sometimes they get merged out-of-order.

This is mostly refactoring, so I'd skip it.

mathetake · 2021-04-13T10:35:58Z

Yeah agreed👍

…

On Tue, Apr 13, 2021 at 17:13 Piotr Sikora ***@***.***> wrote: Thanks! Once the test passes in Envoy, I will merge this and update Envoy to point to the main branch here. We probably shouldn't update every single commit in Enovy, especially since sometimes they get merged out-of-order. This is mostly refactoring, so I'd skip it. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#152 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADHDJ6JJDMTOEBOOQDDJ4ZLTIP4J7ANCNFSM42YZ22BQ> .

PiotrSikora · 2021-04-26T06:16:05Z

Thanks! Once the test passes in Envoy, I will merge this and update Envoy to point to the main branch here.

We probably shouldn't update every single commit in Enovy, especially since sometimes they get merged out-of-order.

This is mostly refactoring, so I'd skip it.

Well, that was terrible timing on my part... For anything that requires changes to code and/or tests, even if it's only because of refactoring, we need to update Envoy as well, otherwise we're might be blocking other PRs.

Do you mind reopening that Envoy PR?

mathetake · 2021-04-26T06:18:52Z

yeah will do

mathetake added 2 commits April 12, 2021 17:36

Extract common code from runtime implementations.

4d46e16

Signed-off-by: Takeshi Yoneda <[email protected]>

Add license

57bb777

Signed-off-by: Takeshi Yoneda <[email protected]>

PiotrSikora reviewed Apr 12, 2021

View reviewed changes

review: stop copying source

700d5b8

Signed-off-by: Takeshi Yoneda <[email protected]>

mathetake force-pushed the runtime-common branch from ad1909c to 700d5b8 Compare April 12, 2021 13:45

Fix build

7172764

Signed-off-by: Takeshi Yoneda <[email protected]>

mathetake marked this pull request as ready for review April 12, 2021 13:49

mathetake commented Apr 12, 2021

View reviewed changes

Fix doc comment

d951111

Signed-off-by: Takeshi Yoneda <[email protected]>

mathetake mentioned this pull request Apr 12, 2021

wasm: update proxy-wasm-cpp-host to use refactored code around runtimes. envoyproxy/envoy#15923

Closed

PiotrSikora reviewed Apr 12, 2021

View reviewed changes

src/common/wasm_util.cc Show resolved Hide resolved

src/v8/v8.cc Show resolved Hide resolved

src/v8/v8.cc Show resolved Hide resolved

src/v8/v8.cc Show resolved Hide resolved

PiotrSikora reviewed Apr 12, 2021

View reviewed changes

src/common/wasm_util.h Outdated Show resolved Hide resolved

mathetake added 6 commits April 13, 2021 08:40

rename files for better alignment

c5b123c

Signed-off-by: Takeshi Yoneda <[email protected]>

Check Wasm header

cefe656

Signed-off-by: Takeshi Yoneda <[email protected]>

Simplify Wasmtime

d749021

Signed-off-by: Takeshi Yoneda <[email protected]>

Add precompiled section test on runtime

265de54

Signed-off-by: Takeshi Yoneda <[email protected]>

fix build

a103f8f

Signed-off-by: Takeshi Yoneda <[email protected]>

Use BytecodeUtil in WAVM for precompiled section

d807f06

Signed-off-by: Takeshi Yoneda <[email protected]>

mathetake added 3 commits April 13, 2021 10:28

Rename wasm_util -> bytecode_util

0f7eac3

Signed-off-by: Takeshi Yoneda <[email protected]>

Add ABI parse in BytecodeUtil

305d48d

Signed-off-by: Takeshi Yoneda <[email protected]>

Use BytecodeUtil in getAbiVersion on WasmVm except NullVm

41a9ac6

Signed-off-by: Takeshi Yoneda <[email protected]>

mathetake force-pushed the runtime-common branch from 6534a25 to 41a9ac6 Compare April 13, 2021 02:40

Construct string only when it is function export.

808a869

Signed-off-by: Takeshi Yoneda <[email protected]>

PiotrSikora reviewed Apr 13, 2021

View reviewed changes

mathetake added 2 commits April 13, 2021 15:04

review: fix style and remove redundant WasmHeader check

1dd2bc1

Signed-off-by: Takeshi Yoneda <[email protected]>

review: copy original bytecode.

ed0f3a5

Signed-off-by: Takeshi Yoneda <[email protected]>

style: align var names across v8 and wasmtime

c639e2c

Signed-off-by: Takeshi Yoneda <[email protected]>

PiotrSikora approved these changes Apr 13, 2021

View reviewed changes

PiotrSikora merged commit 8f8bcc8 into proxy-wasm:master Apr 13, 2021

PiotrSikora mentioned this pull request Apr 13, 2021

Add support for WebAssembly Micro Runtime (WAMR). #142

Merged

mathetake deleted the runtime-common branch April 13, 2021 10:36

Extract common code from runtime implementations. #152

Extract common code from runtime implementations. #152

Uh oh!

Conversation

mathetake commented Apr 12, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mathetake Apr 12, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mathetake Apr 13, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mathetake Apr 13, 2021

Choose a reason for hiding this comment

Uh oh!

PiotrSikora Apr 13, 2021

Choose a reason for hiding this comment

Uh oh!

mathetake Apr 13, 2021

Choose a reason for hiding this comment

Uh oh!

mathetake Apr 13, 2021

Choose a reason for hiding this comment

Uh oh!

PiotrSikora Apr 13, 2021

Choose a reason for hiding this comment

Uh oh!

mathetake commented Apr 12, 2021

Uh oh!

PiotrSikora left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mathetake commented Apr 13, 2021

Uh oh!

mathetake commented Apr 13, 2021

Uh oh!

mathetake commented Apr 13, 2021

Uh oh!

Uh oh!

Uh oh!

Uh oh!

PiotrSikora Apr 13, 2021

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

PiotrSikora left a comment

Choose a reason for hiding this comment

Uh oh!

mathetake commented Apr 13, 2021

Uh oh!

PiotrSikora commented Apr 13, 2021

Uh oh!

mathetake commented Apr 13, 2021 via email

Uh oh!

PiotrSikora commented Apr 26, 2021

Uh oh!

mathetake commented Apr 26, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

mathetake commented Apr 12, 2021 •

edited

Loading

mathetake Apr 12, 2021 •

edited

Loading

mathetake Apr 13, 2021 •

edited

Loading