R=@domenic, @bnoordhuis, @indutny

Stable is on .14; we should not be on .15

4.2.77.15 was the HEAD of branch-heads/4.2 at the time the original PR was merged into next. We're two commits behind, by the way; it's at .17 now.

I'm sorry; I totally forgot my own instructions for picking which tag to use :(. We should probably grab 17 then...

This is probably a silly set of questions, but:

• Where do we fetch V8 from?
• Are there any patches (or other, manual actions) we have to apply to add a new V8?

@chrisdickinson when you find answers perhaps you could document somewhere? either the wiki or in docs/

@chrisdickinson

$ git clone https://chromium.googlesource.com/v8/v8.git
$ git clone https://chromium.googlesource.com/chromium/tools/depot_tools.git
$ cd v8
# if you want to check out a release branch
$ git checkout branch-heads/4.2
$ git` pull --rebase && env PATH=$HOME/src/depot_tools:$PATH ../depot_tools/gclient sync
# to compile a release build
$ make -j8 native  # or x64.release
# to compile debug build without i18n
$ make -j8 x64.debug extrachecks=on i18nsupport=off

Are there any patches (or other, manual actions) we have to apply to add a new V8?

Only if there are floating patches. In most cases you can just git log --oneline deps/v8 and cherry-pick them after upgrading.

Hm, I'm running into issues doing this:

• Git doesn't see any remote (or local) branch-heads/* refs.
• gclient can't find a .gclient file. I created one using gclient config $(git remote -v | grep fetch | awk '{print $2}') (using the git url as the origin url), but gclient sync seems to just re-clone v8.

@chrisdickinson I don't think you have to use gclient unless you want to push stuff back. Cloning the git repo, retrieving commits between known states (current tag vs intended next tag) and go through floating patches to see if they've been merged or interferes with applying the new commits could be enough.

@jbergstroem Cool – the other problem is that I don't see any refs (tags or branches) corresponding to branch-heads in my (fresh) git clone.

@chrisdickinson to see the branch-heads you need to edit the .git/config file as per instructions from @domenic https://gist.github.com/domenic/aca7774a5d94156bfcc1#checking-out-the-head-of-the-corresponding-branch

@ofrobots Thanks for the link. I'll pull this into a document in /doc/ for folks to reference in the future!

To update: I'm seeing errors on tests now that I've merged in V8 4.2.77.18, specifically:

FATAL ERROR: v8::Context::Exit() Cannot exit non-entered context

Ok, tracked down the error. It turns out I hadn't pulled in @indutny's SealHandleScope patch from upstream (I had partially applied enough to get node working, barely.) Updated. PTAL @indutny, @bnoordhuis.

@chrisdickinson wut? was it fixed by re-introducing SealHandleScope?

@indutny When grabbing the new V8 tree, I checked out branch-heads/4.2, which did not have v8's 1f8555 on them. When I tried to compile, it was missing SealHandleScope. I erroneously reintroduced that class def + implementation manually and recompiled. It seemed to work, until I ran tests at which point the above error started spamming out. Running a debug build did not have the same errors. I looked at a5244d3 and noticed the link to 1f8555. I cherry-picked that commit onto my v8 working tree, then vendored that into iojs and recompiled, and the tests passed. Specifically, I had missed these changes.

let's get this show on the road folks!

Branch head is up to 18, would be good to update this PR to just be a clean upgrade to 18 + any floating patches.

Yep, this is updated to 18.

On Apr 26, 2015, at 9:01 PM, Domenic Denicola notifications@github.com wrote:

Branch head is up to 18, would be good to update this PR to just be a clean upgrade to 18 + any floating patches.

—
Reply to this email directly or view it on GitHub.

@domenic Ah – did you mean that the upgrade commits should be squashed down to a single commit? I'll do that on merge. Until then I'm leaving the commits I introduced (3d57c7b and a1b16ca) separate so @indutny and @bnoordhuis can review them.

LGTM but it would be easier if you split off the out-of-tree patches into separate commits. You seem to be doing that for the postmortem fixes but not SealHandleScope.

Ah the commit log typo is what confused me. Also the additional tmpdir() commit.

Ah, updated the commit log, sorry about that.

@bnoordhuis: Ah, my goal in partitioning the history they way I did was to make sure that each commit compiled & passed the tests – for the future, I'll leave them separate.

Also, I'm working on documenting the process over here and will PR that soon.

Merged in 509b59e. Ended up leaving the commits separate.