Insert third_party in to the beginning of PATH

Insert third_party in to the beginning of PATH

This allows to avoid conflicts with libraries in users' PATH

R=pgervais@chromium.org, tandrii@chromium.org

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=295497

[tandrii(chromium), 2015-06-02 18:16:48]

lgtm

[Vadim Sh., 2015-06-02 18:26:08]

Somewhat related FYI: protobuf lives in "google" package. If there's other stuff
needed from "google" package more elaborate path hack is needed, e.g. on GAE we
do this:
https://github.com/luci/luci-py/blob/master/appengine/components/components/u...

[Sergiy Byelozyorov, 2015-06-02 18:29:17]

On 2015/06/02 18:26:08, Vadim Sh. wrote:
> Somewhat related FYI: protobuf lives in "google" package. If there's other
stuff
> needed from "google" package more elaborate path hack is needed, e.g. on GAE
we
> do this:
>
https://github.com/luci/luci-py/blob/master/appengine/components/components/u...

cq_pb2.py imports only from google.protobuf and the google.protobuf library that
I have in depot_tools also imports only from itself.

[nodir, 2015-06-02 18:36:21]

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

[nodir, 2015-06-02 18:41:10]

This CL LGTM, but I am not sure inserting a path to the beginning of sys.path
will help if a package installed system-wide. AFAIU, system-wide packages are
higher priority. That's what I saw, but I may be wrong.

[tandrii(chromium), 2015-06-02 18:50:33]

On 2015/06/02 18:41:10, nodir wrote:
> This CL LGTM, but I am not sure inserting a path to the beginning of sys.path
> will help if a package installed system-wide. AFAIU, system-wide packages are
> higher priority. That's what I saw, but I may be wrong.

AFAIR, Unless the system-wide package is already loaded prior to this path
modification, 
the imports will iterate through various paths in order of sys.path . 
That's why I suggested this change to sergiyb@ . 
I verified this locally, but I couldn't immediately find the C code doing this.

[pgervais, 2015-06-02 20:31:39]

On 2015/06/02 18:50:33, tandrii(chromium) wrote:
> On 2015/06/02 18:41:10, nodir wrote:
> > This CL LGTM, but I am not sure inserting a path to the beginning of
sys.path
> > will help if a package installed system-wide. AFAIU, system-wide packages
are
> > higher priority. That's what I saw, but I may be wrong.
> 
> AFAIR, Unless the system-wide package is already loaded prior to this path
> modification, 
> the imports will iterate through various paths in order of sys.path . 
> That's why I suggested this change to sergiyb@ . 
> I verified this locally, but I couldn't immediately find the C code doing
this.

Well, there is no good solution to this problem outside of having a virtualenv.

Changing sys.path inside the script takes precedence over everything else,
including the system-wide path and user-defined PYTHONPATH. This is good for
this use case, but not true in general.

Another example of this is the competition between packages installed through
apt-get, through easy_install, and those pointed to by PYTHONPATH. 
Packages reached through PYTHONPATH shadow packages installed by apt-get. This
is the reverse with easy_install: they take precedence over packages reached
through PYTHONPATH. This is because easy_install modifies sys.path somewhere
inside sitecustomize.py and forcefully put their packages at the beginning
(which is real bad imho).

virtualenv gives the solution because the global sitecustomize.py is not
executed. You can achieve a similar effect by running 'python -s -S -E' but it's
kind of impractical in the present case.

So lgtm, waiting for a better solution (something like virtualenv).

[Sergiy Byelozyorov, 2015-06-02 21:00:28]

The CQ bit was checked by sergiyb@chromium.org

[Sergiy Byelozyorov, 2015-06-02 21:00:29]

The patchset sent to the CQ was uploaded after l-g-t-m from tandrii@chromium.org
Link to the patchset: https://codereview.chromium.org/1161803004/#ps20001
(title: "Added README.chromium to protobuf module to document its version")

[commit-bot: I haz the power, 2015-06-02 21:00:58]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1161803004/20001

[Sergiy Byelozyorov, 2015-06-02 21:01:08]

On 2015/06/02 20:31:39, pgervais wrote:
> On 2015/06/02 18:50:33, tandrii(chromium) wrote:
> > On 2015/06/02 18:41:10, nodir wrote:
> > > This CL LGTM, but I am not sure inserting a path to the beginning of
> sys.path
> > > will help if a package installed system-wide. AFAIU, system-wide packages
> are
> > > higher priority. That's what I saw, but I may be wrong.
> > 
> > AFAIR, Unless the system-wide package is already loaded prior to this path
> > modification, 
> > the imports will iterate through various paths in order of sys.path . 
> > That's why I suggested this change to sergiyb@ . 
> > I verified this locally, but I couldn't immediately find the C code doing
> this.
> 
> Well, there is no good solution to this problem outside of having a
virtualenv.
> 
> Changing sys.path inside the script takes precedence over everything else,
> including the system-wide path and user-defined PYTHONPATH. This is good for
> this use case, but not true in general.
> 
> Another example of this is the competition between packages installed through
> apt-get, through easy_install, and those pointed to by PYTHONPATH. 
> Packages reached through PYTHONPATH shadow packages installed by apt-get. This
> is the reverse with easy_install: they take precedence over packages reached
> through PYTHONPATH. This is because easy_install modifies sys.path somewhere
> inside sitecustomize.py and forcefully put their packages at the beginning
> (which is real bad imho).
> 
> virtualenv gives the solution because the global sitecustomize.py is not
> executed. You can achieve a similar effect by running 'python -s -S -E' but
it's
> kind of impractical in the present case.
> 
> So lgtm, waiting for a better solution (something like virtualenv).

We can also add 'python -s -S -E' into the commit_queue and commit_queue.bat
binaries.

[commit-bot: I haz the power, 2015-06-02 21:07:15]

Committed patchset #2 (id:20001) as
http://src.chromium.org/viewvc/chrome?view=rev&revision=295497