src: fixes for V8 6.9 and 7.0

[mmarchini]

Several fixes for V8 6.9 (#202, #226) and 7.0 (#232). Still a work in progress though, I'm sending a PR to run CI on these changes.

• PositionInfo removed from SFI and PreParsedData (v8/v8@39e2d97)
  • PositionInfo is now available on ScopeInfo or UncompiledData
• DebugInfo was removed from SFI (v8/v8@c51bcd1)
  • InferredName is available on ScopeInfo or UncompiledData (v8/v8@c941f11)
• StackLocal removed from ScopeInfo (v8/v8@467eb14)
• ParamCount (from ScopeInfo) size decreased (v8/v8@53d4dfc)

TODO

• Improve commit message
• Check if there's no stalled code in the PR
• Update Travis CI to test Node.js v11.x

[codecov-io]

Codecov Report

Merging #247 into master will increase coverage by 11.58%.
The diff coverage is 100%.

@@             Coverage Diff             @@
##           master     #247       +/-   ##
===========================================
+ Coverage   80.84%   92.43%   +11.58%     
===========================================
  Files          33        8       -25     
  Lines        4120      476     -3644     
===========================================
- Hits         3331      440     -2891     
+ Misses        789       36      -753

Impacted Files	Coverage Δ
test/addon/jsapi-test.js	97.05% <ø> (ø)	⬆️
test/plugin/workqueue-test.js	88.88% <100%> (-11.12%)	⬇️
test/common.js	87% <0%> (-0.57%)	⬇️
test/plugin/scan-test.js	98.91% <0%> (-0.07%)	⬇️
src/llnode_module.cc
src/printer.cc
src/constants.cc
src/llnode_api.h
src/addon.cc
... and 15 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 74414fc...e708c92. Read the comment docs.

[mmarchini]

Timer check was removed from workqueue test because of nodejs/node#25806. The test was a duplicate of the previous test anyway, so this shouldn't decrease coverage.

[mmarchini]

While working on this I noticed we use Error::Failures a lot when trying to decode an object from the heap. The problem is those Error::Failures propagate and usually abort our command.

For example: llnode was unable to output the content of process objects in Node.js 11 because it couldn't get the script and source position for some methods in the object, but it could get everything else. llnode didn't give any output to the user because it couldn't retrieve a small piece of information.

IMO we should change the code to be more tolerant when it can't decode only a piece of information. Most users of llnode use it to analyze core dumps, which may have corrupted memory addresses, and being fault-tolerant is essential to analyze such core dumps.

We should move most occurrences of Error::Failure to Error::PrintInDebugMode and save Error::Failures for unrecoverable errors.

This would also allow llnode to work better for future Node.js versions, with fewer features affected by V8 changes.

[mmarchini]

This is ready for review :)

cc @nodejs/llnode

[mmarchini]

CI failures are coming from recent changes in core (bisected and found nodejs/node#26382 to be the reason). Since apparently we were relying on a type indirectly defined internally on core ((Object), which is an object with a non-js-function constructor according to llnode code), I dropped this type from our tests. Remaining basic types should still be enough for testing.

[mmarchini]

CI is green. I'll start working on fixes for Node.js v12 (V8 7.1 to 7.6) in the next few days in an attempt to have llnode working on 12 soon after it goes LTS. But it would be nice to merge this first. cc @nodejs/llnode @nodejs/diagnostics

[mmarchini]

Landed in 99d06e7...13f7034 😄