Land Ben's 32-bit symlink script.

Land Ben's 32-bit symlink script.

BUG=skia:

Committed: https://skia.googlesource.com/skia/+/7e80c889deb5ef3df2afa80520d380ae225b4e72

[mtklein, 2015-02-02 23:49:02]

mtklein@google.com changed reviewers:
+ bungeman@google.com, mtklein@google.com

[mtklein, 2015-02-02 23:49:02]

Did we land this somewhere already?

[mtklein, 2015-02-02 23:50:53]

+CC Ravi

[rmistry, 2015-02-02 23:53:44]

On 2015/02/02 23:50:53, mtklein wrote:
> +CC Ravi

I had independently come up with the symlinks and package installations detailed
here:
https://skia.googlesource.com/buildbot/+/master/compute_engine_scripts/buildb...
(some might be overkill).

[bungeman-skia, 2015-02-03 15:19:18]

On 2015/02/02 23:53:44, rmistry wrote:
> On 2015/02/02 23:50:53, mtklein wrote:
> > +CC Ravi
> 
> I had independently come up with the symlinks and package installations
detailed
> here:
>
https://skia.googlesource.com/buildbot/+/master/compute_engine_scripts/buildb...
> (some might be overkill).

Note that this script doesn't try to install any packages, it just makes up for
a lameness in current Ubuntu multi-arch packages. In Debian multi-arch you can
have libraries for multiple platform architectures installed simultaneously
(this works well), and in theory you can also install the XXX-dev packages for
multiple platform architectures simultaneously. Unfortunately, multiple arch
XXX-dev packages is not an actually tested configuration, so a bunch of dev
packages claim to require that the host platform gcc build tools (for example)
be of it's particular arch. Or the various XXX-dev packages for different archs
claim (usually falsely) to conflict with other archs. This is just broken
dependency issues, but it's a lot to clean up and test, so they haven't. I
believe there are several launchpad bugs about this. The end result, however, is
it's a royal pain to install both x86-64 and x86 XXX-dev packages at the same
time.

Most of the time, all the XXX-dev packages are is a set of symlinks, setting up
the generic names of libraries. This script essentially looks through all of the
x86-64 XXX-dev packages, figures out which symlinks they created, and the
mirrors that on the x86 side if the :i386 package is installed. It then also
warns you about :i386 packages which aren't installed, and when the x86-64
package actually delivered a 'real' .so (which happens quite infrequently). I
wrote this script some time ago so that I could test 32 bit only issues locally
(mostly floating point), but I haven't had to in a while. This was on the docs
site for a while, looks like it's going missing in the new one?

[mtklein, 2015-02-03 15:22:27]

> This was on the docs site for a while, looks like it's going missing in the
new one?

Yep.  That's why I want to check it in.  It was useful exactly as described:
install libfoo, libfoo:i386, and libfoo-dev, then run the script to simulate the
effect of installing libfoo-dev:i386 without the stupid side effect of
uninstalling libfoo-dev.

[bungeman-skia, 2015-02-03 15:39:57]

On 2015/02/03 15:22:27, mtklein wrote:
> > This was on the docs site for a while, looks like it's going missing in the
> new one?
> 
> Yep.  That's why I want to check it in.  It was useful exactly as described:
> install libfoo, libfoo:i386, and libfoo-dev, then run the script to simulate
the
> effect of installing libfoo-dev:i386 without the stupid side effect of
> uninstalling libfoo-dev.

Since this is being divorced from the prose it was originally attached to, we
should include that prose as a comment here, or at least more or less comment #5
here. I think the old prose had a link to a general launchpad bug... at the
moment the ones I can find are ones like
https://bugs.launchpad.net/ubuntu/+source/libpng/+bug/1360285 and
https://bugs.launchpad.net/ubuntu/+source/mesa/+bug/1317113 . As this file
stands (and how it's named) it's not entirely obvious what it's used for by
itself.

[mtklein_C, 2015-02-03 15:43:33]

Committed patchset #1 (id:1) manually as
7e80c889deb5ef3df2afa80520d380ae225b4e72 (presubmit successful).

[mtklein, 2015-02-03 15:43:51]

On 2015/02/03 15:39:57, bungeman1 wrote:
> On 2015/02/03 15:22:27, mtklein wrote:
> > > This was on the docs site for a while, looks like it's going missing in
the
> > new one?
> > 
> > Yep.  That's why I want to check it in.  It was useful exactly as described:
> > install libfoo, libfoo:i386, and libfoo-dev, then run the script to simulate
> the
> > effect of installing libfoo-dev:i386 without the stupid side effect of
> > uninstalling libfoo-dev.
> 
> Since this is being divorced from the prose it was originally attached to, we
> should include that prose as a comment here, or at least more or less comment
#5
> here. I think the old prose had a link to a general launchpad bug... at the
> moment the ones I can find are ones like
> https://bugs.launchpad.net/ubuntu/+source/libpng/+bug/1360285 and
> https://bugs.launchpad.net/ubuntu/+source/mesa/+bug/1317113 . As this file
> stands (and how it's named) it's not entirely obvious what it's used for by
> itself.

Sounds good.  Please attach appropriate prose.