[PATCH] libc++: Add option to disable access to the global filesystem namespace

[Jonathan Roelofs]

REPOSITORY
  rL LLVM

================
Comment at: libcxx/trunk/test/libcxx/test/format.py:63
@@ -62,1 +62,3 @@
 
+        if test.config.unsupported:
+            return (lit.Test.UNSUPPORTED,
----------------
EricWF wrote:
> jroelofs wrote:
> > EricWF wrote:
> > > jroelofs wrote:
> > > > EricWF wrote:
> > > > > I don't think we should be using this. Why not put the metadata into the tests?
> > > > This is how the tests in llvm/test/CodeGen work. I think it cleans things up **significantly**. I view it as a mistake that I did the libcpp-has-no-threads UNSUPPORTEDs this way.... it's a bit painful to maintain them.
> > > My concern is that we accidentally don't run tests because they were mistakenly put in a directory with a lit.site.cfg that marks all the tests as unsupported.
> > > 
> > > I also think that lit.TestRunner.parseIntegratedTestScript may already handle this but I'll have to double check.
> > > My concern is that we accidentally don't run tests because they were mistakenly put in a directory with a lit.site.cfg that marks all the tests as unsupported.
> > 
> > `--show-unsupported` takes care of that, no?
> > 
> > > I also think that lit.TestRunner.parseIntegratedTestScript may already handle this but I'll have to double check.
> > 
> > Apparently not, or there's some flag that I missed that makes it happen... I debugged it a bit this morning before suggesting these lines to @ed
> `--show-unsupported` should *help* but you still need to be watching it carefully and understand what tests are actually unsupported.
> 
> At least with requiring the metadata in the test I can check the commit that put that line in there. If a test is added to a directory that has a `lit.site.cfg` it's a little harder to figure out.
> 
> I don't really mind the extra maintenance cost because it make things more explicit and clear to readers of the test.
> 
> Anyway I don't strongly object to this change so I'll drop the subject.
What if they got marked in the `lit.local.cfg` as xfail instead of unsupported? I could see that being quite a bit less risky as the tests would still get run regardless of the feature.

http://reviews.llvm.org/D8194

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/