Test Plan: compiled file_util_test

Reviewed By: soren@fb.com

FB internal diff: D809250

Summary:
Calling CHECK() in folly will force the program to abort in case of a failure.
On the other hand, for range checking, the standard library throws std::out_of_range for many functions.

Thus it could be a good idea to throw the same exception in folly so that errors can be handled using try {} catch (...) {} blocks.

Test Plan:
from fbcode, type:
fbconfig -r folly; fbmake opt -j32; fbmake runtests_opt -j 32

What other tests should I run?

Reviewed By: tudorb@fb.com

FB internal diff: D808204

Summary:
Calling CHECK() in folly will force the program to abort in case of a failure.
On the other hand, for range checking, the standard library throws std::out_of_range for many functions.

Thus it could be a good idea to throw the same exception in folly so that errors can be handled using try {} catch (...) {} blocks.

Test Plan:
from fbcode, type:
fbconfig -r folly; fbmake opt -j32; fbmake runtests_opt -j 32

What other tests should I run?

Reviewed By: tudorb@fb.com

FB internal diff: D808204

Summary:
Also move manually-generated stuff away from folly-config.h and into
Portability.h.  This should be only for things that differs between the
various (compiler, library) pairs used internally at FB; everything else
should be autoconf-ed and therefore go into folly-config.h.

Test Plan: built and ran file_util_test on various platforms

Reviewed By: delong.j@fb.com

FB internal diff: D807067

Blame Revision: D806781

Summary: Strings can be moved around while the vector grows.

Test Plan: file_util_test

Reviewed By: lucian@fb.com

FB internal diff: D806966

Blame Revision: D806781

Summary: zero-copy

Test Plan: record_io_test

Reviewed By: lucian@fb.com

FB internal diff: D806813

Summary:
Testing incomplete reads / writes is hard, so I'm only testing the wrappers
(template functions that take the underlying operation and retry it in
case of incomplete operations).  Note the cute hack of using variadic
templates to use the same wrapper for both file pointer- and explicit-position
flavors of the functions (the offset argument becomes optional).

Test Plan: test added

Reviewed By: lucian@fb.com

FB internal diff: D806781

Summary:
You can now submit to the same AsyncIO object from different threads, but you
must still reap from only one thread at a time.

Test Plan: async_io_test, added MT test

Reviewed By: philipp@fb.com

FB internal diff: D804914

Summary:
gcc and clang both give you these three macros pre-defined. I've found a
reference on the interweb to this being broken
(http://gcc.gnu.org/ml/gcc/2011-08/msg00152.html) but that was with gcc
4.4. I've tested with 4.6, 4.7 and clang 3.2 and they all do the right
thing on little-endian (internal build and OS X). Alas, I don't have a
big-endian system handy to test with.

Test Plan: fbconfig/fbmake runtests.

Reviewed By: simpkins@fb.com

FB internal diff: D799416

Summary:
Unfortunately, this doesn't exist on all platforms. Fortunately, both
gcc and clang support builtins that do the same thing as the bswap_*
functions we were including byteswap.h for.

I changed the test file to hardcode expected results for the swapping
functions. Is that OK? It seemed a little silly to be testing the
Endian::whatever functions by comparing them against the functions that
they're implemented in terms of.

Test Plan:
fbconfig/fbmake runtests with gcc 4.6 and 4.7. Verified
separately that clang supports these builtins.

Reviewed By: simpkins@fb.com

FB internal diff: D799244

Summary:
It doesn't exist on some systems (at least Mac OS X). We're only using
it for __GNUC_PREREQ, which is easy to provide our own definition for.

I moved the definitions of FOLLY_FINAL and FOLLY_OVERRIDE into
folly-config.h so we can autoconf them in the open-source build. The
hardcoded stuff for the internal build is a little ugly, unfortunately.

folly can't be built with gcc versions earlier than 4.6, so that check
in ThreadLocal.h is pointless by now. (Plus we use noexcept without a
macro wrapper all over the place.) That stuff was also not
clang-friendly. clang has supported static_assert since 2.9 and noexcept
since... I'm not sure but at least 3.0.

Test Plan:
fbconfig/fbmake runtests, with gcc 4.6 and 4.7. clang can't
build folly right now, but I verified separately that it supports
noexcept and static_assert.

Reviewed By: simpkins@fb.com

FB internal diff: D799143

Summary:
I am trying to move more code into a HPHP extension, and it's complaining about
this.

Test Plan: `fbconfig -r folly && fbmake runtests`

Reviewed By: delong.j@fb.com

FB internal diff: D799096

Summary:
Discovered this while trying to build folly on OS X. It would print a
warning about Boost being missing that got drowned out in the rest of
the configure output, and then the build would fail. This makes the
failure more friendly.

Test Plan:
`autoreconf; ./configure` on a system without boost
installed; make sure configure fails and the last thing it prints is the
boost error message.

Reviewed By: delong.j@fb.com

FB internal diff: D797389

Summary:
- Add some convenience methods for reading std::string objects via
folly::Cursor

Test Plan: - Unit tests

Reviewed By: simpkins@fb.com

FB internal diff: D795428

Summary:
There are quite a few places where we split strings into a fixed number
of fields. This enables this to be done a bit faster by not using any
variable-length structures at runtime.

Test Plan: Unit tests, Benchmarks

Reviewed By: philipp@fb.com

FB internal diff: D794523

Summary:
RecordIO provides an interface to write and read streams of variable-length
records that is resilient in the face of failure and data corruption.  If
the stream is corrupted in any way, you will lose records, but the stream
will resynchronize.

We have one implementation of RecordIO reader/writer that reads from / writes
to regular files (using mmap(), even for writes, so we can preserve state
even in case of process death -- the buffer cache will flush things to disk
eventually) and we expose enough of the guts (in a reasonably clean way)
so you can build your own on top of other backends.

Test Plan: test added

Reviewed By: mmcurtiss@fb.com

FB internal diff: D790275

Summary:
Test required a separate process, as fcntl locks are always re-granted to
a process that holds the lock.

Test Plan: new tests, all tests under folly

Reviewed By: lucian@fb.com

FB internal diff: D791370

Summary:
This wasn't building with an error from a warning about
narrowing conversions---probably we didn't notice because it was
previously broken due to folly docs being broken forever.  Folly tests
should always pass.  (I want to add it to the fbcode-tests-must-pass
commit hook.)

Test Plan:
fbmake runtests_opt in this platform, except with stl_tests
turned off because it takes years to compile.

Reviewed By: andrewjcg@fb.com

FB internal diff: D792059

Summary: It was broken because the `dynamic`'s iterator wasn't complete.

Test Plan: Unit test

Reviewed By: jbrewer@fb.com

FB internal diff: D793618

Test Plan: Unit tests

Reviewed By: jbrewer@fb.com

FB internal diff: D791149

Summary: Comparisons with values can lead to confusion, especially when the value itself is truthy. To avoid this confusion, I'm disabling comparison with contained value. Note that Optionals can still be constructed and assigned these values, but comparsion must be done separately for the container and the contained.

Test Plan: Unit tests, contbuild

Reviewed By: tudorb@fb.com

FB internal diff: D783621

Test Plan: ran it

Reviewed By: lucian@fb.com

FB internal diff: D787733

Summary:
The BucketedTimeSeries::rateHelper() function is generic, and not
specific to the BucketedTimeSeries class.  This diff moves it to
folly/detail/Stats.h, so other parts of the stats code can access it as
well.

Test Plan: Ran the folly unit tests.

Reviewed By: delong.j@fb.com

FB internal diff: D778115

Summary:
Make rate() and countRate() more accurate when queried for a specific
time range.

Previously these functions divided the estimated sum/count by the entire
time range specified.  This underestimated the rate if we don't actually
have data for the entire time period.  (Since the sum computed only
takes into account the time range for which we have data.)

For example, if the timeseries duration was 60 seconds, but only 30
seconds of data had been entered so far, rate(now - 60, now) would
underestimate the rate by half, since there was only 30 seconds worth of
data available.  The no-argument version of rate() did work correctly in
that case.

Test Plan:
Added a new unit test for this behavior.

Also fixed the existing rate test code, which had the same bug and
expected the underestimated rate.

Reviewed By: delong.j@fb.com

FB internal diff: D778114

Summary:
21.4.2 [string.cons] / 23 says

> If *this and str are the same object, the member has no effect.

That means we have to support self-move-assignment.

Test Plan: added test which triggered assertion, ran it

Reviewed By: andrei.alexandrescu@fb.com

FB internal diff: D785057

Test Plan: test added

Reviewed By: delong.j@fb.com

FB internal diff: D783100

Summary: Seems like a good idea.

Test Plan: compiled all of folly

Reviewed By: delong.j@fb.com

FB internal diff: D783104

Test Plan: fbconfig $(find folly -name test) && fbmake runtests_opt

Reviewed By: tjackson@fb.com

FB internal diff: D777186

Summary:
Now that the third_party link was updated in
https://phabricator.fb.com/D778617, we're good.

Test Plan: fbconfig -r thrift && fbmake runtests_opt

Reviewed By: chip@fb.com

FB internal diff: D778707

Summary: Unbreak build.  third-party link not yet updated after https://phabricator.fb.com/D776457

Test Plan: no

Reviewed By: chip@fb.com

FB internal diff: D778669

Summary: Cleaned up from common/strings/URL.h, and it should be URI, not URL.

Test Plan: tests added

Reviewed By: chip@fb.com

FB internal diff: D768880

Summary:
impl_.e_ += newB - impl_.b_; fails when the difference
between newB and impl_.b_ isn't a multiple of sizeof(T).

Test Plan: .

Reviewed By: oyamauchi@fb.com

FB internal diff: D774754

Summary:
We can't throw after the process is created, because we don't know what to do
with it (and the Subprocess object goes up in smoke, so we can't rely on the
caller to clean up, either).  So don't throw.

If we throw before the process is created, make sure we clean up.

Test Plan: subprocess_test

Reviewed By: delong.j@fb.com

FB internal diff: D774722

Summary:
folly/Range.h is extensively used in fbcode, try to avoid including
<iostream> (which is pretty heavy), include forward declarations
(<iosfwd>) instead.

Also transitioned it from 'std type_traits' + 'boost type_traits' to
'std type_traits'.

Test Plan: compiled, ran tests

Reviewed By: soren@fb.com

FB internal diff: D774834

Summary:
Move Histogram.h into the stats/ subdirectory, along side the
BucketedTimeSeries code.  Eventually I plan to land more of our stats
code in this subdirectory too.

This also renames Histogram-inl.h to Histogram-defs.h, and no longer
includes it by default from Histogram.h.  Instead, this adds a new
stats/Instantiations.cpp file which explicitly instantiates
Histogram<int64_t> and BucketedTimeSeries<int64_t>.  Most callers use
these instantiations, and they now no longer need to include
Histogram-defs.h.  Only callers that need other instantiations need to
include Histogram-defs.h.  This will speed up build times.

Test Plan:
Ran "arc unit" to build all projects depending on folly, and verified
they all still built and passed tests.

Reviewed By: ldbrandy@fb.com

FB internal diff: D761377

Summary: rebinding allocators is too cumbersome, this diff implements a helper to make this job easier.

Test Plan: unit test added

Reviewed By: tudorb@fb.com

FB internal diff: D766451

Summary:
currently, StlAllocator<Alloc, void> can't be used when you
want an untyped allocator that can be rebound later since it doesn't
carry the SimpleAllocator pointer with it. This diff fixes that.

Test Plan: unit test added

Reviewed By: jon.coens@fb.com

FB internal diff: D766559

Summary:
Clang throws error because the callback refers to itself inside its lambda
function definition.  I prevent this by declaring the variable first then the
compiler does not complain when it is used within its lambda definition.

folly/test/TimeoutQueueTest.cpp:99:37: error: variable 'cb' is uninitialized
when used within its own initialization [-Werror,-Wuninitialized]

Test Plan:
fbconfig --clang --platform=gcc-4.7.1-glibc-2.14.1 --with-project-version
boost:1.51.0 folly/test/
fbmake opt
fbmake runtests_opt

Reviewed By: ldbrandy@fb.com

FB internal diff: D753061

Summary: Create an artificial limit on an arena to start throwing bad_alloc before running out of system memory

Test Plan: adjust unit test

Reviewed By: marcelo.juchem@fb.com

FB internal diff: D762695

Test Plan: all folly unit tests

Reviewed By: andrewjcg@fb.com

FB internal diff: D757374