Modify base_unittests.exe to use tcmalloc

Modify base_unittests.exe to use the allocator shims
required to use tcmalloc on Windows.

This fixes failing SecurityTest.MemoryAllocationRestriction
tests which are testing the tcmalloc 2Gb memory limits
imposed for security

BUG=242304
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=203907

[Will Harris, 2013-05-24 09:15:56]

Hi Jar - debate to be had here is whether to enable tcmalloc
on base_unittests or not.  I vote 'yes' because
base_unittests does enough low level stuff that it's
probably a good thing to use the same allocator
as we use in the chrome binaries - but another
option here would be to shift the 2Gb limit tests into
the alloactor tests, although this requires getting
the allocator_unittests target building on platforms
other than Windows.

[jar (doing other things), 2013-05-24 17:56:15]

We ship with TCMalloc... so I'd sure like to see as much testing with TCMalloc
as possible.

One hassle is that perhaps the DLL flavor build uses the default allocator...
and I *think* that maybe we <sigh> use the DLL build for testing :-(.

If you can make this fly... I'm in favor.

Added Ricardo to review list in case he has comments, corrections, or cautions
to add.

[rvargas (doing something else), 2013-05-24 18:30:36]

This should work, but I'm not so sure about the change.

On one hand, from the point of view of base, tcmalloc is a configurable feature
that embedders can enable as they wish, and given that the feature is made
visible through base that would mean things should work with any allocator. So
maybe base should be tested with all allocators :(

On the other hand, the product that we care the most is published with tcmalloc,
so maybe we should test more with tcmalloc.

But there is the whole issue of unit tests being really unit tests and not
integration tests. And that would call for independent features to be fully unit
tested so that other components (and unit tests!) can concentrate on what they
should do instead of worry about every dependency that they have. And that would
call for things to be unit tested at the allocator layer instead of depending on
base_unittests.

So my vote is to do any specific test that we *really* need at the allocator
layer (like the 2GB limit) so that base_unittests  can pretty much ignore which
allocator is used to build the executable.

[Will Harris, 2013-05-26 09:13:50]

I agree with the goals that if we're shipping multiple allocators then we really
should be testing them all, and also any allocator specific tests should
probably be in unittests specific to that allocator.

so - perhaps we should move the 2Gb security tests into the tcmalloc unit tests
(since they are specific to tcmalloc) - and also change the test harness to test
all four allocators (via CHROME_ALLOCATOR=) in the allocator_unittests?

If you guys agree, I can prepare a CL for this change.  jln@ do you have a view
on this?

[jln (very slow on Chromium), 2013-05-28 22:30:56]

On 2013/05/26 09:13:50, Will Harris wrote:
> I agree with the goals that if we're shipping multiple allocators then we
really
> should be testing them all, and also any allocator specific tests should
> probably be in unittests specific to that allocator.
> 
> so - perhaps we should move the 2Gb security tests into the tcmalloc unit
tests
> (since they are specific to tcmalloc) - and also change the test harness to
test
> all four allocators (via CHROME_ALLOCATOR=) in the allocator_unittests?
> 
> If you guys agree, I can prepare a CL for this change.  jln@ do you have a
view
> on this?

Having Linux use the allocator shim would be a fine solution, but I fear it's
going to be too long for fixing this.

Right now base_unittests is using tcmalloc on Linux, and I would favor landing a
change to make base_unitests use tcmalloc on Windows as well as soon as
possible.

The alternative: to disable important security tests on Windows until the work
to unify the allocator shim is done seems dangerous. I didn't really look at it,
but I suspect that if the people who wrote the allocator shim didn't make it
work on Linux in the first place, there might be a reason.

[Will Harris, 2013-05-29 11:46:25]

shall I just go ahead and push this CL to enable tcmalloc in windows
base_unittests and we can defer the debate on exactly where the 2Gb
tests should be until we can scope out the cost of the Linux work?

I think this gives a meaningful improvement to our test coverage -
as base_unittests will then execute with the same allocator that we
use in a default chrome configuration.

[Will Harris, 2013-06-03 09:18:45]

gentle poke - can I get an LGTM on this please?

[rvargas (doing something else), 2013-06-03 18:06:07]

lgtm

[jar (doing other things), 2013-06-03 18:51:53]

lgtm

[commit-bot: I haz the power, 2013-06-03 18:57:58]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/wfh@chromium.org/15674003/20001

[commit-bot: I haz the power, 2013-06-04 08:38:22]

Change committed as 203907