Fix g++-multilib conflict between install-build-deps scripts

Fix g++-multilib conflict between install-build-deps scripts

Currently install-build-deps-android.sh installs g++-multilib which
conflicts with packages installed by install-build-deps.sh (since
gcc-multilib contains a /usr/include/asm symlink and the crosscompiler
packagers don't want you to accidentally include the wrong
architecture's asm headers).

The g++-multilib package was wrongly commented as being required for
Android SDK; it turns out these requirements have changed, so I
updated the libraries required for the Android SDK based on their docs.

g++-4.8-multilib is however still (likely) needed, alongside
linux-libc-dev:i386, for compiling the V8 snapshot. So I've
consolidated those deps, and the logic for selecting them, under the
--lib32 flag of install-build-deps (to avoid duplicating the code in
both). I updated the `if (trusty) install g++-4.8-multilib` logic so
it chooses the right version of the multilib automatically, rather
than hardcoding the version needed for trusty (which will break in
utopic or beyond).

Finally I removed the `sudo apt-get update` in
install-build-deps-android which is redundant now that we call
install-build-deps which already does that. And I added `sudo dpkg
--add-architecture i386` which seems to have been missing from these
scripts.

BUG=435056
Committed: https://crrev.com/49bb458a54a0e959f91ef612c79fad89e7d65c91
Cr-Commit-Position: refs/heads/master@{#305991}

[johnme, 2014-11-20 11:40:33]

johnme@chromium.org changed reviewers:
+ sbc@chromium.org, sivachandra@chromium.org, thestig@chromium.org

[Sam Clegg, 2014-11-25 03:20:48]

I don't know enough about how the android build works on what it needs to know
if this change is good.  If lei nor siva do then you might want to try torne@.

[johnme, 2014-11-25 14:22:18]

johnme@chromium.org changed reviewers:
+ torne@chromium.org

[johnme, 2014-11-25 14:22:19]

+Torne, any thoughts?

[Torne, 2014-11-25 16:56:20]

Not quite the right fix as discussed offline with John; he'll prepare another
version. The packaging issues here are *really* dumb :)

[Sam Clegg, 2014-11-25 17:10:47]

On 2014/11/25 16:56:20, Torne wrote:
> Not quite the right fix as discussed offline with John; he'll prepare another
> version. The packaging issues here are *really* dumb :)

I never did get to the bottom of why the arm compiler conflicted with
g++-multilib.  Is that some rational, or a packaging mistake?

[Torne, 2014-11-25 17:13:59]

On 2014/11/25 17:10:47, Sam Clegg wrote:
> On 2014/11/25 16:56:20, Torne wrote:
> > Not quite the right fix as discussed offline with John; he'll prepare
another
> > version. The packaging issues here are *really* dumb :)
> 
> I never did get to the bottom of why the arm compiler conflicted with
> g++-multilib.  Is that some rational, or a packaging mistake?

Short answer: gcc-multilib contains a /usr/include/asm symlink and the
crosscompiler packagers don't want you to accidentally include the wrong
architecture's asm headers.

[Primiano Tucci (use gerrit), 2014-11-26 16:41:42]

primiano@chromium.org changed reviewers:
+ primiano@chromium.org

[Primiano Tucci (use gerrit), 2014-11-26 16:41:43]

The rationale and logic look good but need a fix to actually work :) (see
comments below)

https://codereview.chromium.org/740963002/diff/60001/build/install-build-deps.sh
File build/install-build-deps.sh (right):

https://codereview.chromium.org/740963002/diff/60001/build/install-build-deps...
build/install-build-deps.sh:144: lib32_list="linux-libc-dev:i386"
Can you move these lines below (near the next +++ chunk), so all the lib32_list
logic is together?

https://codereview.chromium.org/740963002/diff/60001/build/install-build-deps...
build/install-build-deps.sh:290: lib32_list=
Uhm there seems to be no use of this lib32_list.
I guess you meant to add an apt-get install step below after the
add-architecture step?

[johnme, 2014-11-26 17:14:37]

https://codereview.chromium.org/740963002/diff/60001/build/install-build-deps.sh
File build/install-build-deps.sh (right):

https://codereview.chromium.org/740963002/diff/60001/build/install-build-deps...
build/install-build-deps.sh:144: lib32_list="linux-libc-dev:i386"
On 2014/11/26 16:41:43, Primiano Tucci wrote:
> Can you move these lines below (near the next +++ chunk), so all the
lib32_list
> logic is together?

I don't want to move this too far from the definition of the other _lists, so
instead I've moved the g++-X.Y-multilib section up closer below.

https://codereview.chromium.org/740963002/diff/60001/build/install-build-deps...
build/install-build-deps.sh:290: lib32_list=
On 2014/11/26 16:41:43, Primiano Tucci wrote:
> Uhm there seems to be no use of this lib32_list.
> I guess you meant to add an apt-get install step below after the
> add-architecture step?

Done.

[Sam Clegg, 2014-11-26 17:24:22]

https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
File build/install-build-deps.sh (right):

https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
build/install-build-deps.sh:214: # When cross building for arm/Android on 64-bit
systems the host binaries
How about for building 32-bit chrome?  Or is that all taken care of by the
chroot environment?

https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
build/install-build-deps.sh:349: sudo dpkg --add-architecture i386
Seems strange to me that we never needed this before.  Any idea why?  (Perhaps
its the default for some distros?)

[Primiano Tucci (use gerrit), 2014-11-26 17:27:42]

LGTM. If you want before committing I can kick off two VMs (precise and trusty)
and check that it actually works :)

https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
File build/install-build-deps.sh (right):

https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
build/install-build-deps.sh:214: # When cross building for arm/Android on 64-bit
systems the host binaries
On 2014/11/26 17:24:22, Sam Clegg wrote:
> How about for building 32-bit chrome?  Or is that all taken care of by the
> chroot environment?

When building for 32-bit chrome this is not sufficient. You need a full chroot.

https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
build/install-build-deps.sh:349: sudo dpkg --add-architecture i386
On 2014/11/26 17:24:22, Sam Clegg wrote:
> Seems strange to me that we never needed this before.  Any idea why?  (Perhaps
> its the default for some distros?)

This is needed for the :i386 packages. Before (e.g. precise) there was a
catch-call package (ia32-libs) for dealing (Badly) with this

[Sam Clegg, 2014-11-26 17:33:33]

On 2014/11/26 17:27:42, Primiano Tucci wrote:
> LGTM. If you want before committing I can kick off two VMs (precise and
trusty)
> and check that it actually works :)
> 
>
https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
> File build/install-build-deps.sh (right):
> 
>
https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
> build/install-build-deps.sh:214: # When cross building for arm/Android on
64-bit
> systems the host binaries
> On 2014/11/26 17:24:22, Sam Clegg wrote:
> > How about for building 32-bit chrome?  Or is that all taken care of by the
> > chroot environment?
> 
> When building for 32-bit chrome this is not sufficient. You need a full
chroot.
> 
>
https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
> build/install-build-deps.sh:349: sudo dpkg --add-architecture i386
> On 2014/11/26 17:24:22, Sam Clegg wrote:
> > Seems strange to me that we never needed this before.  Any idea why? 
(Perhaps
> > its the default for some distros?)
> 
> This is needed for the :i386 packages. Before (e.g. precise) there was a
> catch-call package (ia32-libs) for dealing (Badly) with this

I'm just a little confused because the arm_list previously contains
linux-libc-dev:i386 and this is enabled by default (both nacl and arm are
enabled by default, at the request of infra who wanted the default to do the
right thing for creating vm images for build machines that might need to build
everything).  So it seem that anyone who ran this script previously would
already be exposed to :i386 installs.. but nobody complained about lack of
--add-architecture...

[johnme, 2014-11-26 18:28:44]

On 2014/11/26 17:33:33, Sam Clegg wrote:
>
https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
> > build/install-build-deps.sh:349: sudo dpkg --add-architecture i386
> > On 2014/11/26 17:24:22, Sam Clegg wrote:
> > > Seems strange to me that we never needed this before.  Any idea why? 
> (Perhaps
> > > its the default for some distros?)
> > 
> > This is needed for the :i386 packages. Before (e.g. precise) there was a
> > catch-call package (ia32-libs) for dealing (Badly) with this
> 
> I'm just a little confused because the arm_list previously contains
> linux-libc-dev:i386 and this is enabled by default (both nacl and arm are
> enabled by default, at the request of infra who wanted the default to do the
> right thing for creating vm images for build machines that might need to build
> everything).  So it seem that anyone who ran this script previously would
> already be exposed to :i386 installs.. but nobody complained about lack of
> --add-architecture...

It's possible that all Goobuntu machines have that architecture by default. But
https://wiki.debian.org/Multiarch/HOWTO says you need to run this before running
`sudo apt-get update` or it won't fetch package lists for i386, and there's no
harm in running it more than once, so seems better to explicitly include it than
to rely on default configurations of certain distros.

[Sam Clegg, 2014-11-26 18:30:06]

On 2014/11/26 18:28:44, johnme wrote:
> On 2014/11/26 17:33:33, Sam Clegg wrote:
> >
>
https://codereview.chromium.org/740963002/diff/100001/build/install-build-dep...
> > > build/install-build-deps.sh:349: sudo dpkg --add-architecture i386
> > > On 2014/11/26 17:24:22, Sam Clegg wrote:
> > > > Seems strange to me that we never needed this before.  Any idea why? 
> > (Perhaps
> > > > its the default for some distros?)
> > > 
> > > This is needed for the :i386 packages. Before (e.g. precise) there was a
> > > catch-call package (ia32-libs) for dealing (Badly) with this
> > 
> > I'm just a little confused because the arm_list previously contains
> > linux-libc-dev:i386 and this is enabled by default (both nacl and arm are
> > enabled by default, at the request of infra who wanted the default to do the
> > right thing for creating vm images for build machines that might need to
build
> > everything).  So it seem that anyone who ran this script previously would
> > already be exposed to :i386 installs.. but nobody complained about lack of
> > --add-architecture...
> 
> It's possible that all Goobuntu machines have that architecture by default.
But
> https://wiki.debian.org/Multiarch/HOWTO says you need to run this before
running
> `sudo apt-get update` or it won't fetch package lists for i386, and there's no
> harm in running it more than once, so seems better to explicitly include it
than
> to rely on default configurations of certain distros.

ok lgtm

[Torne, 2014-11-27 14:54:14]

lgtm

[johnme, 2014-11-27 14:59:20]

The CQ bit was checked by johnme@chromium.org

[commit-bot: I haz the power, 2014-11-27 15:00:41]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/740963002/120001

[johnme, 2014-11-27 15:03:35]

On 2014/11/26 17:27:42, Primiano Tucci wrote:
> LGTM. If you want before committing I can kick off two VMs (precise and
trusty)
> and check that it actually works :)

Primiano ran these checks, and it seems that running either or both scripts
works in both gPrecise and gTrusty, though there's a pre-existing issue
(g++-arm-linux-gnueabihf conflicts with g++-4.8-multilib) when running
install-build-deps.sh with vanilla Trusty. Since the pre-existing issue isn't
aggravated by these changes (which still install the same g++-4.8-multilib on
Trusty), landing this now, and the pre-existing issue can be dealt with
orthogonally.

[commit-bot: I haz the power, 2014-11-27 15:45:38]

Committed patchset #7 (id:120001)

[commit-bot: I haz the power, 2014-11-27 15:46:29]

Patchset 7 (id:??) landed as
https://crrev.com/49bb458a54a0e959f91ef612c79fad89e7d65c91
Cr-Commit-Position: refs/heads/master@{#305991}