Merge missing changes from 3.14-ppc into libv8-3.14

[andrewlow]

There are several commits @mtbrandy made to clean up the 3.14-ppc stream that should be incorporated into libv8-3.14

There is also at least one change I missed when making libv8-3.14 from src/objects-inl.h

This item exists so I don't forget to take care of this.

[kapouer]

Interested, since there's a good chance these 3.14 ppc branch patches make their way to the libv8-3.14 debian package.

[andrewlow]

@kapouer That's the plan. I helped Adam create the original patch into the ubuntu libv8-314 for PPC (BE/LE).

[andrewlow]

Big bunch of commits merged from 3.14-ppc branch and some cleanup

commit 7d1b8b5
commit a891130
commit e65316c
commit bd647a1 << upgrade patch to 3.14.5.9
commit 6a4dae6
commit 3104c7a
commit 1fce584
commit dd1c665
commit 8564b62
commit b9c7b1a

[andrewlow]

commit b74ef76
commit 1079d0e
commit eef4397
commit 16022ca
commit 2f80bb1

One note - this upgrades to 3.14.5.9 which is a a single commit: bd647a1 but appears to be Node related and probably a good idea.

[andrewlow]

Submitted a bug on launchpad https://bugs.launchpad.net/ubuntu/+source/libv8-3.14/+bug/1289490 to track inclusion of this in the Ubuntu builds.

[kapouer]

Had a second look.
Debian ppc port supports low-end POWER instructions sets.
It's not that obvious what exactly (but i'll ask for more info).
At least the build servers set a minimum requirement, see for example poulenc or praetorius.
That means libv8-3.14 package cannot support "powerpc" architecture unless you manage to
merge the main ppc port with the g4compat branch.

[andrewlow]

@kapouer - I looked at poulenc (Xserve Dual Core G5) and praetorius (IBM p5 52A, 4 x POWER5+ 1.6GHz) - I would have thought that the current libv8-3.14 did work fine here.

It won't work on a G4 processor - the currently unmerged g4compat branch is for those. I can understand that debian wants to be as widely supported as possible, so this is another vote to merge those changes in. (I may try again to do that.. it was just a pretty big diversion for quite an old processor).

I would like to understand if you have evidence that the two build servers listed above are actually incompatible with the code we have in libv8-3.14. If so, that's a different bug than G4 compatibility and I'd like to address that first.

[kapouer]

No i don't have any evidence the build servers are actually incompatible.
(I thought the G5 was < POWER5).
Here's some information about which instruction sets are supported in debian:
https://lists.debian.org/debian-powerpc/2014/04/msg00005.html

[andrewlow]

Len Soremsem's comments outlines PowerPC levels we do support.

He also correctly points out that there are issues with some of the Freescale processors due to our assumption of FPU support

Standard powerpc FPU is expected to exist, which makes the SPE powerpc
versions not able to run it (which is mainly e500 core powerquicks as
far as I know)."

So, based on this data - the code should be compatible. If you hit problems with those servers / that CPU specification we will happily accept issues and resolve them.

[kapouer]

Fine by me, will try the patch and report, then.

[kapouer]

Which branch is the one i can review ? libv8-3.14 or 3.14-ppc ?

[andrewlow]

The libv8-3.14 branch was prepared specifically to make the merge into the Ubuntu builds as simple as possible. The parent of the libv8-3.14 branch is 3.14-ppc.

We use 3.14-ppc as our branch to support the PowerPC version of Node.js.

For debian, I'd be tempted to start with libv8-3.14 and see if it's a close match to what you have now on other platforms. If there are big issues - then I'd be happy to take a look at the deltas and figure out how we can best provide code to you that requires the minimum effort on your part.

If it's a big effort, and you want to do the work yourself, then starting from 3.14-ppc makes the most sense.