future / fiber integration

Summary:
make future::wait() use fiber's baton, so wait works in threads or in fibers.

Much cleaner than making a new FiberRequest type in thrift

Test Plan: tests

Reviewed By: andrii@fb.com

Subscribers: doug, alandau, bmatheny, mshneer, andrii, folly-diffs@, yitingli, yfeldblum, jsedgwick, chalfant

FB internal diff: D1996283

Signature: t1:1996283:1429144165:da5dc6b1f2a053a45efd39877e79169e3fba810c

when* -> collect*

Summary: title

Test Plan: tests

Reviewed By: hans@fb.com

Subscribers: laser-diffs@, trunkagent, mhl, rhe, fbcode-common-diffs@, chaoyc, search-fbcode-diffs@, hero-diffs@, zeus-diffs@, vikas, cold-storage-diffs@, rtgw-diffs@, unicorn-diffs@, targeting-diff-backend@, netego-diffs@, apollo-diffs@, everstore-dev@, zhuohuang, zhguo, jying, darshan, apodsiadlo, alikhtarov, folly-diffs@, wch, lins, tingy, jsedgwick, thom, yfeldblum, petchean, iaroslav, qhuang, gus, san, tomasz, pknowles, lyang, chalfant, paggarw, stevenkim

FB internal diff: D2003300

Tasks: 6025255

Signature: t1:2003300:1429659170:e18999cba45e8aa9019aa94f1f29732076a274ad

fix collect() for move-only types

Summary:
as above. it never ends.

Test Plan: added unit

Reviewed By: hans@fb.com

Subscribers: folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D2011569

Signature: t1:2011569:1429660210:930cb17682d5c86a11881a23efe0a91f4c6a36b1

then-with-Executor

Summary:
Pass an Executor to `then`, which applies only for that callback. This is on
the one hand just a convenience method, but it's a major convenience when it's
needed, because grabbing, storing, and restoring the old Executor in the
middle of a chain is very inconvenient indeed.

Test Plan:
new unit
docblock comments

Reviewed By: jsedgwick@fb.com

Subscribers: folly-diffs@, davejwatson, chalfant, yfeldblum, nkgupta, jsedgwick, exa, robbert

FB internal diff: D2011542

Tasks: 6771589, 6838553

Signature: t1:2011542:1429660204:f5959b1e0b3b36dfb8c3c7091302d19101dde93b

fix collect for non-default-constructible types, for real this time

Summary:
this was a fun one. Add a specialized implementation that builds up the results in a map with their indices and aggregates them into a vector at the end

Test Plan: unit tests

Reviewed By: hans@fb.com

Subscribers: folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D2002444

Signature: t1:2002444:1429642589:ee5aa5e8c461db97a28642b9887b3158df317813

map()

Summary:
simple sugary syrup, best used in conjunction with collect/whenAll

Test Plan: unit

Reviewed By: hans@fb.com

Subscribers: trunkagent, folly-diffs@, jsedgwick, yfeldblum, chalfant, cgthayer

FB internal diff: D1999047

Tasks: 6045623

Signature: t1:1999047:1429570631:62361bb43dc5489fe3d4eb31af404faf8a765402

Assume exception when Executor::add throws

Summary:
Rather than crashing spectacularly, if `Executor::add` throws (e.g. because the queue is full), then discard the result we got and assume the exception the executor threw instead.

Alternatively, we could pass this exceptional Try to the callback (without an executor, as it is here), but not perturb `result_`. This would mean two different world views in these two code snippets:

auto f1 = makeFuture(42).via(&crappyExecutor);
f1.value(); // 42 (no callback happened)
f1.then(...); // would see the executor's exception. Would also be ill-advised to do this after value()

auto f2 = makeFuture(42).via(&crappyExecutor)
.then([](int x) { return x * 2; }); // skipped
f2.value(); // throws executor's exception

It feels rude to throw away the result, but it feels too potentially dangerous to allow this split view of the world.

Test Plan: modified unit

Reviewed By: jsedgwick@fb.com

Subscribers: exa, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D2007729

Tasks: 5306911

Signature: t1:2007729:1429627114:b627ce758ce9231298f1b28e203ccc1ee415ed9a

revert D1985475, clang still borked

Summary:
This reverts commit bb08e2405dc68e6dab1f0066b15faa2821ad3dc7.

Test Plan: revert-hammer

Reviewed By: hans@fb.com

Subscribers: mathieubaudet, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D2006283

Signature: t1:2006283:1429556243:41e6b3189ce2da5a0f1a32a663ff8761628ca574

test failure conditions in LengthFieldBasedFrameDecoder

Summary: Moar unittests.

Test Plan:
unittests

Reviewed By: hans@fb.com

Subscribers: doug, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1959161

Signature: t1:1959161:1429292487:b7d10be35c2cf1d0bc1b399f4f523392a138a217

Revert "Revert "Deprecating folly::is_complete""

Summary:
This reverts commit 95a80fd4cb5a4b31b41b34fd24d591e0fc1b0650.

Test Plan: n/a

Reviewed By: andrewjcg@fb.com

Subscribers: trunkagent, folly-diffs@, yfeldblum, chalfant

FB internal diff: D2004342

Tasks: 6804947

Signature: t1:2004342:1429332777:4dd1c45e22ee4da69e5e84d3c5f67bada263f307

Blame Revision: D2002345

Revert "Deprecating folly::is_complete"

Summary: This reverts commit 42add531d4bdb1f95c40e41981f908a1b70865fb.

Test Plan: n/a

Reviewed By: andrewjcg@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D2002345

Tasks: 6804947

Signature: t1:2002345:1429296810:7db71fe4748d5c71f0f0486751ee426c2cdff653

then-with-Executor

Summary:
Pass an Executor to `then`, which applies only for that callback. This is on
the one hand just a convenience method, but it's a major convenience when it's
needed, because grabbing, storing, and restoring the old Executor in the
middle of a chain is very inconvenient indeed.

Test Plan:
new unit

Reviewed By: jsedgwick@fb.com

Subscribers: robbert, exa, folly-diffs@, jsedgwick, nkgupta, yfeldblum, chalfant, davejwatson

FB internal diff: D1985475

Signature: t1:1985475:1429148056:0450120263ba8110e8825420cbefe3b1887f7306

fix collect for non-default-constructible types

Summary: as above

Test Plan: compile

Reviewed By: agoder@fb.com

Subscribers: folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1999218

Signature: t1:1999218:1429217794:0278955bb471fe73a14ccedb59eae80b98acdbf3

README: Fixed typo, formatted code, and linked to docs

Summary:
(from @njormrod) This changes the READMEs to be .md,
appropriate for github. I viewed the docs on github, and they looked
WAY better as .md files.

Signed-off-by: Nicholas Ormrod <njormrod@fb.com>
Test Plan: view on github

Reviewed By: markisaa@fb.com

Subscribers: fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1999531

Tasks: 6783581

Signature: t1:1999531:1429220432:68c6d8dc70806d993c83cacce6369cb7e6e964c8

Remove unused variables

Summary:
This broke the HHVM (on ubuntu at least)

Test Plan:
compiled it

Signed-off-by: Nicholas Ormrod <njormrod@fb.com>
Reviewed By: andrei.alexandrescu@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1999341

Tasks: 6783581

Signature: t1:1999341:1429219016:235beef7522c4ada020eedf4f0a2ae200e94b7c8

Adds x64 arch-specific test

Summary:
Adds x64 arch-specific test inside the
if HAVE_X86_64

Signed-off-by: Nicholas Ormrod <njormrod@fb.com>
Test Plan: tests

Reviewed By: markisaa@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1999428

Tasks: 6783581

Signature: t1:1999428:1429220350:4287342a5a99823c9e4ffe20d0e47039f9fdbc0b

Update Copyright 2014 -> 2015

Summary:
Signed-off-by: Nicholas Ormrod <njormrod@fb.com>
Test Plan: Comments only

Reviewed By: markisaa@fb.com, yfeldblum@fb.com

Subscribers: fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1999373

Tasks: 6783581

Signature: t1:1999373:1429219027:225446bb418f3f8d65fb195682e4e7fe315ed41a

Deprecating folly::is_complete

Summary: Deprecating folly::is_complete. Replacement is fatal::is_complete.

Test Plan: unit tests

Reviewed By: njormrod@fb.com

Subscribers: njormrod, folly-diffs@, yfeldblum, chalfant

FB internal diff: D1979124

Tasks: 6697454, 6697600

Signature: t1:1979124:1429153237:dc5736ed0f546dd4dd57bda9176e3c3c13b1235d

We might avoid some temporaries in putMessagesImpl

Summary:
This is folly github pull request https://github.com/facebook/folly/pull/184

It seems we might avoid some temporaries in putMessagesImpl
function. To do so we pass arguments directly to constructors
with the help of emplace_back member.

Test Plan:
All folly/tests, make check for 37 tests, passed.

Signed-off-by: Nicholas Ormrod <njormrod@fb.com>
Reviewed By: davejwatson@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1998515

Tasks: 6783581

Signature: t1:1998515:1429207683:056db129405bed212f50a50821f095c8d3694ec8

fix build

Summary:
^

Test Plan: compiles

Reviewed By: hans@fb.com

Subscribers: folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1998822

Signature: t1:1998822:1429211074:3ff9b0827a8ce029efa06a4a78ecea2b07eae3a8

runImmediatly fix

Summary: Should also work if eventbase isn't looping

Test Plan: added new unittest

Reviewed By: hans@fb.com

Subscribers: doug, folly-diffs@, yfeldblum, chalfant

FB internal diff: D1984259

Signature: t1:1984259:1428691788:edb89424a48b8071a776d828300b7c1094f1c48f

collect()

Summary: title

Test Plan: unit

Reviewed By: hans@fb.com

Subscribers: fbcode-common-diffs@, targeting-diff-backend@, zhuohuang, thom, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1992144

Tasks: 6025255

Signature: t1:1992144:1429120337:7678f790dd0f383295c036e6627bdf417ae43fc7

fulfil -> setWith, fulfilTry -> setTry

Summary: title

Test Plan: tests

Reviewed By: hans@fb.com

Subscribers: cgist, cold-storage-diffs@, fugalh, atlas2-eng@, zhuohuang, folly-diffs@, jsedgwick, yfeldblum, chalfant, andrii

FB internal diff: D1994472

Tasks: 6768508

Signature: t1:1994472:1429117362:218c4fac3c88fcc8d37dc22ff8fe4135a73ec5d5

Instrument fibers execution

Summary: Add option to setup fibers' instrumentation.

Test Plan: Unit tests

Reviewed By: andrii@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1991638

Tasks: 6347990

Signature: t1:1991638:1429144487:3c3586b9943b4cc6bbef0297e48b8bffaf586f0d

Make sure we can't access LocalData when destroying it

Summary: We can access LocalData while currentFiber is set. We should make sure it's set to null when LocalData::reset is called.

Test Plan: unit test

Reviewed By: alikhtarov@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant, bwatling

FB internal diff: D1996071

Tasks: 6725667

Signature: t1:1996071:1429135408:d549d577e140ce2867aff4130e73be3884dbd2ed

onError(exception_wrapper)

Summary: title

Test Plan: unit

Reviewed By: hans@fb.com

Subscribers: folly-diffs@, jsedgwick, yfeldblum, chalfant, hannesr, vloh

FB internal diff: D1984864

Tasks: 6045789

Signature: t1:1984864:1429116418:b4a9cdbb88f605a09b5753eea41dd970c96b9d4e

remove unnecessary & deprecated activate() call

Summary: let's not do this to ourselves :)

Test Plan: unit

Reviewed By: yfeldblum@fb.com

Subscribers: folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1993491

Signature: t1:1993491:1429063112:b8664b365ad12fa4b51e9bd7fde9e4d197dbc648

Ensure the loop callback is scheduled when the ready queue is not empty

Summary: Previously we'd call ensureLoopScheduled() but it'd be a no-op since the loop was already scheduled. Delaying the call to ensureLoopScheduled() fixes the issue

Test Plan: unit tests (FiberManager.yieldTest fails without the changes to FiberManager-inl.h)

Reviewed By: andrii@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1993686

Signature: t1:1993686:1429070253:af933abbbbb33868a402f1d643e4e6f5fef1be83

fix race in ~ThreadWheelTimekeeper

Summary:
During destruction, we enqueue a callback to `eventBase_` that references `wheelTimer_`.

So long as that callback exists, we have a circular dependency as both reference the other and we -cannot- safely destroy this object (and its members). The fix is to break this dependency someway. In this case, I've chosen to simply `AndWait` until the eventBase has consumed the message referencing the timer. From here normal destruction of members can proceed safely (destorying HHWheelTimer first, EventBase second).

The problem in the current code is that the Eventbase will attempt to consume all messages during it's own destruction. If `wheelTimer_->cancelAll()` is still enqueued, it will attempt to use the now destroyed `wheelTimer_`.

Test Plan:
Running this test repeatedly was able to repro:

./folly/futures/futures-test --gtest_filter=Timekeeper.futureWithinVoidSpecialization

Reviewed By: yfeldblum@fb.com

Subscribers: enis, folly-diffs@, jsedgwick, yfeldblum, darshan, chalfant

FB internal diff: D1985967

Tasks: 6332729, 6741095

Signature: t1:1985967:1428726270:997ec277c6a73554e54b8cf673acd36ff62976e6

parse flags in unit tests

Summary: This code has useful logging when turned up high enough (e.g. --v=11). But we need to parse the flags to see it.

Test Plan: trial & error

Reviewed By: yfeldblum@fb.com

Subscribers: enis, folly-diffs@, jsedgwick, yfeldblum, darshan, chalfant

FB internal diff: D1985965

Signature: t1:1985965:1428713065:4607610ec438416fffb5fced1867efa13544eef0

(rx) Include types.h first

Summary: We need to do forward declarations first, and types.h already does declarations, so just make sure it gets included first in each file.

Test Plan:
tests

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fugalh, exa, folly-diffs@, jsedgwick, yfeldblum

FB internal diff: D1861040

Tasks: 6258585

Signature: t1:1861040:1428604059:de6393e3c87e0694a6b04ad6d440cc9635a1008d

Bump version to 35:0

SharedMutex potential lost wakeup with exactly 3 or 4 contending writers

Summary:
SharedMutex used a saturating counter that records the number of
waiting lock() calls, but an ABA problem on futexWait could lead to a lost
wakeup when there was exactly 3 or 4 threads contending on the RW lock
in W mode.  This diff changes the kWaitingE count to be heuristic (it is
possible that the count says 1 but there are two waiters), saturates at
2 instead of 3 (because there is no benefit from differentiating those
two), and doesn't decrement the count on a successful wakeup.

Also, I noticed while debugging this that boost::noncopyable was causing
SharedMutex to be 8 bytes when it should only be 4.

One way the wakeup could be lost in the old code:

1. A calls lock()
2. A updates state <- kHasE
3. A returns
4. B calls lock()
5. B spins
6. B updates state <- kHasE + 1 * kIncrWaitingE
7. A calls unlock()
8. A updates state <- 0
9. A calls futexWake(), which returns 0
10. A calls lock()
11. A updates state <- kHasE
12. A returns
13. C calls lock()
14. C spins
15. C updates state <- kHasE + 1 * kIncrWaitingE
16. C calls futexWait, expecting kHasE + 1 * kIncrWaitingE
17. B calls futexWait, expecting kHasE + 1 * kIncrWaitingE
18. A calls unlock()
19. A updates state <- 0
20. A calls futexWake(), which returns 1
21. C receives the wakeup
22. C updates state <- kHasE
23. C returns
24. C calls unlock()
25. C updates state <- 0

B missed the wakeup that was intended for it (sent at step 9, wait
started at step 17), but went to sleep anyway because it saw the write
state at step 17. Now there are two waiters but only 1 recorded in the
SharedMutex, at which point failure is inevitable.

Test Plan:
1. DeterministicSchedule test using uniformSubset that can repro the problem
2. Test in production scenario that produced occasional deadlocks under high stress

Reviewed By: yfeldblum@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1980210

Tasks: 6720328

Signature: t1:1980210:1428623932:ef1c00c3f88154578b2b253ac0cfdbadf9f31d8c

Move StringHandler to its own file

Summary: Move this from a test to its own file in codec folder

Test Plan:
already has a unittest
fbconfig -r folly/wangle; fbmake runtests

Reviewed By: hans@fb.com

Subscribers: doug, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1959152

Signature: t1:1959152:1427935371:9d1338986df97fd94f98ad2bf08c731d229bc2b1

LineBasedFrameDecoder

Summary: Copy of netty's line based decoder.

Test Plan:
unittests
fbconfig folly/wangle/codec; fbmake runtests

Reviewed By: hans@fb.com

Subscribers: doug, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1959155

Signature: t1:1959155:1427935150:e11280c5567df9ad9964dbb656aa090267856f57

Pipeline OutputBufferingHandler

Summary: Part 2 of pipeline.  Use OutputBufferingHandler, which is exactly the same output buffering thrift does currently.

Test Plan:
A couple canaries, unittests.
Specific issues of previous pipeline diffs I think are fixed already and should be fine with this one:
* Previously there were missing DestructorGuards.  I don't think anything in OutputBUfferHandler needs a DG
* previously broke http GET handling.  There is a unittest for it now.

Canary

Reviewed By: alandau@fb.com

Subscribers: doug, fugalh, alandau, bmatheny, mshneer, folly-diffs@, jsedgwick, yfeldblum, chalfant, vloh, andreib

FB internal diff: D1970949

Signature: t1:1970949:1428360621:a0db142032894525f69c2b144cf946d63f790fe1

folly: MemoryMapping: add offset/size params to advise

Test Plan: ran it

Reviewed By: philipp@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1974936

Signature: t1:1974936:1428453023:1a83ed8336f75b745bffc633b8471231423b0fb0

folly: MemoryMapping: respect writable when opening the file

Summary: the file needs to be opened with write-permissions before it can be mmaped with PROT_WRITE.

Test Plan: n/a

Reviewed By: philipp@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1974895

Signature: t1:1974895:1428451664:1106019abf02b9de659ce0d305e3d788ae7d60bc

AsyncSocketHandler cacheChainLength

Summary:
Is there any reason we shouldn't just cache the chain length everywhere?

Currently the pipeline does a lot of passing of buffers between IOBufQueues, I wonder if we can get even smarter about passing the length around

Test Plan: used in telnet server diff

Reviewed By: hans@fb.com

Subscribers: doug, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1959167

Signature: t1:1959167:1427934568:eb9a49514b2bb7d157879bb8efa99f12092d5e95

Remove FiberManager move-constructor

Summary:
LoopController and Fibers keep references to FiberManager, so implementing move-constructor for it is non-trivial.
The only purpose of move constructor was to have a create() static method, replaced it with a constructor accepting a type tag.

Test Plan: unit test + tao build

Reviewed By: stepan@fb.com, pavlo@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1971881

Signature: t1:1971881:1428374073:945d913e69eaa6f957dace981c23835105d91935

AsyncUDPServerSocket passes socket in callback

Summary: AsyncUDPServerSocket doesn't make it easy to write to the same socket you read from.  Add the socket as a callback param, similar to AsyncServerSocket

Test Plan:
fbconfig -r folly; fbmake dbg

Will fixup any other spots contbuild finds

Reviewed By: hans@fb.com

Subscribers: bmatheny, doug, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1948936

Signature: t1:1948936:1427841651:20d13d73c06d31c75056624f051a6fd35b9701fb

Add ability to yield execution of the currently running fiber

Summary: This diff allows fibers to explicitly yield execution to other fibers and the event loop

Test Plan: unit tests

Reviewed By: andrii@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant, tao-eng@

FB internal diff: D1965297

Signature: t1:1965297:1428082686:e524e9dd21b3fb951e1d3556e4cb3eedc3e6511a

Fix open-source Makefiles after fibers were moved to folly

Test Plan: contbuild

Reviewed By: pavlo@fb.com

Subscribers: alikhtarov, folly-diffs@, yfeldblum, chalfant

FB internal diff: D1966698

Signature: t1:1966698:1428367351:62f2218c23d4911c15055917c0fbae31a2420903

GlobalExecutor.h include Executor.h and IOExecutor.h

Summary: We were forward-declaring `Executor` and `IOExecutor` for great compilation speed. However, this leaves users of this header to include at least one of those headers in their cpp file to actually use it, and that's kind of annoying. `Executor` and `IOExecutor` are not templated and are quite small, so it seems like the compilation speed benefit is minimal compared to the convenience factor, so this diff just includes the two headers. (Probably including `Executor.h` is redundant since `IOExecutor.h` probably includes it, but for completeness and explicitness, and in case `IOExecutor` also just forward-declares `Executor` now or in the future.)

Test Plan: stuff still builds

Reviewed By: davejwatson@fb.com

Subscribers: fugalh, exa, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1970179

Signature: t1:1970179:1428351076:e9c3700a868688035e7ff50486b8cead751d3c02

Add FiberManager local type

Summary: This adds local type for each FiberManager. Only local of given type can be created on fibers of this FiberManager. Locals of other types will be just treated as regualar thread-locals.

Test Plan: unit test

Reviewed By: bwatling@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1969739

Signature: t1:1969739:1428345931:aff5deb526c179158e5881b29330ff3f6698149a

Option to record precise stack on every N fiber

Summary: title + some logging around

Test Plan: mcrouter unit tests

Reviewed By: andrii@fb.com

Subscribers: alikhtarov, folly-diffs@, yfeldblum, chalfant

FB internal diff: D1965097

Signature: t1:1965097:1428103637:341ff705c9d11c730197828256ec0b790ee7afaa

shared ptr vector sockets

Summary: promote the sockets vector to a shared_ptr, since both ServerWorkerPool and ServerBootstrap use it.  Otherwise there are destruction order issues between ServerBootstrap and any IOThreadPoolExecutor you use

Test Plan: Saw use after free in D1942242, gone after this.

Reviewed By: yfeldblum@fb.com

Subscribers: chalfant, doug, fugalh, folly-diffs@, jsedgwick, yfeldblum

FB internal diff: D1947553

Signature: t1:1947553:1427484417:5b78f5c9c70d244d3f52a6f71b6d1fab7b29d106

Use Fiber locals for TAO Fiber-locals

Summary:
This simplifies TAO fibers locals, using FiberManager based fiber-local storage.
This depends on D1958135.

Test Plan: integration tests

Reviewed By: bwatling@fb.com

Subscribers: alikhtarov

FB internal diff: D1962214

Signature: t1:1962214:1427997755:b546b9039382a7600f234b0a2a60cc96da34e662

Fiber-local context

Summary: This adds fiber-local context, which behaves more like static objects+fork rather than thread-locals.

Test Plan: unit test

Reviewed By: pavlo@fb.com, bwatling@fb.com

Subscribers: rushix, alikhtarov, bwatling

FB internal diff: D1958135

Signature: t1:1958135:1427999426:8e4b89f4af53a1a119b2e5a765fb549dd8442c50

Add static method to skip SSL init

Summary:
Some things like gSOAP, initialize SSL only to have SSLContext stomp over their
work. Add a method to allow flagging SSLContext as "initialized" to use the
existing global state.

Test Plan:
Tested running mcrouter and paymentnegined to verify existing
functionality appears normal.

Reviewed By: subodh@fb.com

Subscribers: ssl-diffs@, folly-diffs@, yfeldblum, chalfant, dihde

FB internal diff: D1959936

Tasks: 6655251

Signature: t1:1959936:1427994292:8b2be1742ef17a5ec2c7b3ad37787d2cf4d142a6

Bump version to 34:0

Update Makefile.am for AsyncSignalHandler

Summary: facepalm

Test Plan: Built proxygen OSS that depends on this symbol

Reviewed By: davejwatson@fb.com

Subscribers: doug, folly-diffs@, yfeldblum, chalfant

FB internal diff: D1963679

Signature: t1:1963679:1428006035:2c4ceffc046b995431cde2fbcef757fa7ea1ea0b

Move mcrouter/lib/fibers to folly/experimental/fibers

Summary:
The mcrouter fibers library is quite general purpose and reusable by other projects. Hence I'm moving it into folly.

Test Plan:
fbconfig -r folly/experimental/fibers && fbmake runtests
fbconfig -r mcrouter && fbmake runtests
fbconfig -r tao && fbmake
fbconfig -r ti && fbmake

Reviewed By: andrii@fb.com

Subscribers: vikas, zhuohuang, jmkaldor, jhunt, pavlo, int, aap, trunkagent, fredemmott, alikhtarov, folly-diffs@, jsedgwick, yfeldblum, chalfant, chip

FB internal diff: D1958061

Signature: t1:1958061:1428005194:b57bfecfe9678e81c48526f57e6197270e2b5a27

Bump version to 33:0

UDP bind via port or address

Summary: Bind via port or address, similar to tcp

Test Plan: I wrote a unittest that tested v4 vs v6 binding - can't check it in because we have v6 only clusters though.  Not sure how eles to write a unittest without picking specific ports

Reviewed By: hans@fb.com

Subscribers: doug, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1948938

Signature: t1:1948938:1427841721:621031320981bf366a8252a0ea1d4491b694ea24

Move TAsyncSignalHandler into folly

Summary:
TODO item, trying to remove unecessary dependencies on thrift

Test Plan: Unit tests

Reviewed By: davejwatson@fb.com

Subscribers: doug, fbcode-common-diffs@, davejwatson, andrewcox, alandau, bmatheny, anca, darshan, mshneer, folly-diffs@, bil, yfeldblum, haijunz, chalfant

FB internal diff: D1960215

Signature: t1:1960215:1427920934:8abd7e94c50676b05bf7ff79800df0db1bd04266

Add socket mocks

Summary: These parallel the MockTAsync* from thrift/lib/cpp/test.  Once all the users of the thrift versions are gone they can be deleted

Test Plan: Unit tests

Reviewed By: yfeldblum@fb.com, alandau@fb.com

Subscribers: doug, net-systems@, folly-diffs@, yfeldblum, chalfant

FB internal diff: D1959979

Signature: t1:1959979:1427917909:19af219f88dd6847a064da986dd30765e29bdc99

Move AsyncSocket tests from thrift to folly

Summary: These tests belong with the code that they test.  The old tests had a couple dependencies on TSocket/TSSLSocket, so I wrote a BlockingSocket wrapper for AsyncSocket/AsyncSSLSocket

Test Plan: Ran the tests

Reviewed By: alandau@fb.com

Subscribers: doug, net-systems@, alandau, bmatheny, mshneer, folly-diffs@, yfeldblum, chalfant

FB internal diff: D1959955

Signature: t1:1959955:1427917833:73d334846cf248f8bb215f3eb5b596df7f7cee4f

Move AtomicLinkedList to folly

Summary:
Adding new atomic data structure from the mcrouter project.

Test Plan: fbconfig -r mcrouter && fbmake runtests

Reviewed By: andrii@fb.com

Subscribers: alikhtarov, folly-diffs@, yfeldblum, chalfant, achao, jpearce

FB internal diff: D1960219

Signature: t1:1960219:1427921204:78e5313ea916b8a249d32e31cc7ba1aa9e890d95

(Wangle) Swap order of Try<T> and T matching

Summary:
See the test case and D1958860.

I don't really know why I chose one over the other. This fixes a bug. So
it's better?

Test Plan: Run all the tests?

Reviewed By: yfeldblum@fb.com

Subscribers: folly-diffs@, jsedgwick, yfeldblum, chalfant, over

FB internal diff: D1960740

Signature: t1:1960740:1427927644:25093b049a1519d7c869ee7043f3caced4cc971e

Add support for writev for AsyncUDPSocket

Summary:
I have 2 local buffers and instead of allocating iobufs
on the heap, I can simply use iovec on the stack to send.

Test Plan: Unit tests and turn server.

Reviewed By: davejwatson@fb.com

Subscribers: folly-diffs@, ehrhardt, yfeldblum, chalfant, paramr

FB internal diff: D1956201

Signature: t1:1956201:1427918649:09f00399fd9e8ed3fe62be43b19adbf0a5a3f3b0

Minor fixes from documentation

Test Plan: unittests

Reviewed By: jbrewer@fb.com, ashishsharma@fb.com

Subscribers: maxime, folly-diffs@, yfeldblum, chalfant

FB internal diff: D1956101

Signature: t1:1956101:1427828321:2d828e6f2c06c6dcd6e7baa3ed5e72ee59d9b77f

improve split() docs

Summary:
From reading the docs, I assumed that casting exceptions would be caught.  Let's clarify this.

Test Plan: readingiskey

Reviewed By: yfeldblum@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant

FB internal diff: D1960186

Signature: t1:1960186:1427920471:b7aa2af38a90a986418b98294fc0f69e678f0d1e

(Wangle) Allocate lambda space inside Core instead of inside std::function

Summary:
Taking this trick that is used in the fibers library. We can keep 64
bytes of space inside `Core` to allocate the callback lambda into,
instead of having `std::function` do another `malloc`. This seems to
greatly improve the synthethic benchmark, and hopefully production
workloads, too, by reducing the number of mallocs.

64 bytes were picked because it fits all the lambdas in the futures
tests. We might want to adjust this based on production data...?

https://fb.intern.facebook.com/groups/715931878455430/permalink/837898842925399/

Test Plan: Run all the tests for all platforms, compilers, and Windtunnel.

Reviewed By: hans@fb.com

Subscribers: chalfant, meisner, folly-diffs@, jsedgwick, yfeldblum

FB internal diff: D1931620

Signature: t1:1931620:1427841595:6ec667b58980be232dfb116bc316148bb67de4fc

stop in same thread

Summary: In a couple places in ServerBootstrap there is code that needs to run in the EB, but if we are already in the EB, it's fine to run it inline.  Maybe this should be a method on EB directly?  There is runInEventBaseThreadAndWait(), but it explicitly disallows this usage.

Test Plan: used in D1942242, deadlocks without this, since stop() was called in the same eb

Reviewed By: yfeldblum@fb.com

Subscribers: chalfant, doug, fugalh, folly-diffs@, jsedgwick, yfeldblum

FB internal diff: D1947581

Signature: t1:1947581:1427831021:d1d08ff9a4a00501d6be60670709fcb17af04134

Move OpenSSL uninitialization to a separate function

Summary:
The current behaviour of SSLContext is subject to a bug where an instance going
out of scope at the wrong time can decrement the global refcount to 0, causing
the destructor to destroy the SSL context. Instead, this requires callers to
explicitly uninitialize the SSLContext instance and assume the risk of wrongly
destroying the global SSL context.

Test Plan:
Tested with mcrouter

Reviewed By: subodh@fb.com

Subscribers: afrind, ssl-diffs@, folly-diffs@, yfeldblum, chalfant, dihde, melitam

FB internal diff: D1949649

Tasks: 6358211

Signature: t1:1949649:1427854689:aea2dc801f63256ff64188b0f7a15121dcecee69

Get access to EventBaseManager object

Summary: as title

Test Plan: run unit tests

Reviewed By: yfeldblum@fb.com

Subscribers: ming, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1955074

MPMCQueue - FPE to invalid_argument for 0 capacity

Summary:
If an MPMCQueue was initialized with an explicit zero capacity, it would
try to compute stride for zero and trigger SIGFPE which pretty fatally
killed the process -- but in a way that required gdb to figure out the
root cause.

This replaces that with a std::invalid_argument exception, which
includes a description of the problem and should make fixing this user
error much faster.

Test Plan:
GTEST unit test added to verify behaviour; if this didn't work it would
SIGFPE and abort the test script, which is probably a good indicator
that something went wrong. :)

fbconfig -r folly && fbmake runtests

Reviewed By: ngbronson@fb.com

Subscribers: chalfant, folly-diffs@, yfeldblum, rkomorn

FB internal diff: D1930978

Signature: t1:1930978:1427740315:cc06a8b9f3c314b956ae41f813b2f904d3e979c9

Default accept function in Acceptor

Summary: Since Acceptor can be used for TCP or UDP, make an empty default TCP accept function so subclasses don't have to.

Test Plan:
built D1942242, no build errors

contbuild

Reviewed By: yfeldblum@fb.com

Subscribers: doug, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1948940

Signature: t1:1948940:1427498977:26b7bb23e3367ff6695dad86ebf0e60637815a7b

EventBase::runImediatelyOrRunInEventBaseThreadAndWait.

Summary:
[Folly] EventBase::runImediatelyOrRunInEventBaseThreadAndWait.

When you have code that you need to run in the event loop and wait for it to finish, and you yourself might be in the event loop but might be in another thread instead.

Test Plan:
Unit tests:
* `folly/io/async/test/EventBaseTest.cpp`.

Reviewed By: davejwatson@fb.com

Subscribers: folly-diffs@, yfeldblum, chalfant, agallagher, njormrod, ldbrandy, brettp, dougw

FB internal diff: D1953110

Signature: t1:1953110:1427751076:ad2c9f4d5b65ef952cba94b06be887672ee44583

Fixing predicate inlining

Summary:
as title.

Test Plan: run unit tests

Reviewed By: yfeldblum@fb.com

Subscribers: ming, fugalh, folly-diffs@, jsedgwick, yfeldblum, ridge@

FB internal diff: D1946589

Extend IO pool with external optional event_base_manager

Summary: as title

Test Plan: run unit tests

Reviewed By: yfeldblum@fb.com

Subscribers: ming, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1950610

asyncudpserversocket reuse port

Summary:
AsyncUDPSocket supports reuse port, also make it available in AsyncUDPServerSocket.
Has to happen before bind

Test Plan:
?? it builds?

Mutliple binds already work with reuseaddr, but load balancing doesn't work as well.  See http://lwn.net/Articles/542629/ for details

Reviewed By: naizhi@fb.com

Subscribers: doug, fugalh, folly-diffs@, jsedgwick, yfeldblum, chalfant

FB internal diff: D1948932

Signature: t1:1948932:1427473013:8c114458a0746bb2b6957db33c1834179d860c00

Future::filter(A => bool)

Test Plan: new unit tests

Reviewed By: davejwatson@fb.com

Subscribers: chalfant, bmatheny, trunkagent, exa, folly-diffs@, yfeldblum, jsedgwick

FB internal diff: D1828251

Tasks: 6166893

Signature: t1:1828251:1427475565:512463d5728482a40c1da2548a5bed91e7f92d60

Add CaptureFD for log testing (and some glog patterns)

Summary:
Without a gadget like this, it's really hard to test logging output. With it, it's really easy. I found about 50 callsites to the various functions across several projects, so folly seems appropriate.

PS The patterns are functions for two reasons:
- Static variables are a pain.
- This leaves the option of adding an optional argument, so you can grep for a particular kind of error string.

Test Plan: unit test

Reviewed By: yfeldblum@fb.com

Subscribers: folly-diffs@, yfeldblum

FB internal diff: D1933439

Signature: t1:1933439:1427345479:5b3d1c6566a026fdbccb16b382211688e327ea1a

Add EXPECT_{NO_,}_PCRE_MATCH macros

Summary: There are about 40 callsites to these across a couple of projects, and they seem generally useful.

Test Plan: ```fbconfig folly/experimental/test && fbmake runtests```

Reviewed By: yfeldblum@fb.com

Subscribers: folly-diffs@, yfeldblum

FB internal diff: D1933415

Signature: t1:1933415:1427345307:6587eea3dac74c5d841950ace3c3501d6e3dbe4a

TemporaryFile and TemporaryDirectory

Summary:
TemporaryFile, TemporaryDirectory, and ChangeToTempDir should all be moveable
objects, but not copiable.  Define default move constructors and move
assignment operators for these classes.  This will prevent copy constructor and
copy assignment operators from being implicitly defined.

Test Plan:
Used this in a new test to write a helper function which created and returned
a new TemporaryFile object using the move constructor.

Reviewed By: yfeldblum@fb.com

Subscribers: doug, net-systems@, exa, folly-diffs@, yfeldblum

FB internal diff: D1945134

Signature: t1:1945134:1427342944:3428327e797ce4b3d362f9a2d2276de6d8b96137

(Wangle) Reduce

Summary:
1] The lambda should ble able to return a `Try<T>`. Maybe? Can a `then`
return a `Try<T>` actually? Can fix this with `resultOf`. (Doubling the number of functions to 4.)

2] `initial` and `func` have to be copyable.

Test Plan: Added tests.

Reviewed By: hans@fb.com

Subscribers: trunkagent, folly-diffs@, jsedgwick, yfeldblum

FB internal diff: D1870996

Tasks: 6025252

Signature: t1:1870996:1427318511:2ae5894b79022da88990835b26d35c4520fdbd29

Bump version to 32:0

Add a pretty macro for deprecation

Summary: Make it easy for people to mark code as deprecated

Test Plan: Chose a random file and added the FOLLY_DEPRECATED macro to it

Reviewed By: yfeldblum@fb.com

Subscribers: ps, folly-diffs@, yfeldblum

FB internal diff: D1943248

Signature: t1:1943248:1427316689:b3305d134f374697e93d36bafbf0961eec1ef68a

Relax CHECK condition in stringAppendfImpl

Summary:
vsnprintf can return fewer bytes and bytes_used if
we print a string containing '\0' using a width
specifier.

Test Plan: -- ran all tests

Reviewed By: andrei.alexandrescu@fb.com

Subscribers: folly-diffs@, yfeldblum

FB internal diff: D1915035

Signature: t1:1915035:1426799341:4aaea928c4bdde1998bf66cf9e2732a53572c6e3

httpserver on serverbootstrap (2)

Summary:
Original diff D1800100, reverted in D1895181.  Issue seems to have been caused by a copy constructor issue.

Fixes: Remove copy constructor, move contructor only. Separate stop() and join(), since join() can happen only after all stop()s have completed when reusing the same IOThreadPoolExecutors.
Test Plan: build and unit tests of relevant product

Reviewed By: praveenr@fb.com

Subscribers: trunkagent, doug, fugalh, alandau, bmatheny, mshneer, folly-diffs@, jsedgwick, yfeldblum

FB internal diff: D1901718

Tasks: 6431975

Signature: t1:1901718:1426885577:46b63bfe117df207be3d273b953c100249c1a7da

Blame Revision: D1895181

Add ChangeToTempDir to TestUtil

Summary: It's convenient to be able to write tempfiles in the current directory. This provides a one-liner way of doing that in tests.

Test Plan: unit test

Reviewed By: yfeldblum@fb.com

Subscribers: simpkins, folly-diffs@, yfeldblum

FB internal diff: D1931918

Signature: t1:1931918:1426913634:f988fb3c5061f5909e309dcaf42742d9b2ed18c6

Make folly::Singleton DFATAL as late as possible

Summary: We want to print all destructor stack-traces if Singleton destruction happens eventually (even after destroyInstances() was done).

Test Plan:
unit test

Reviewed By: chip@fb.com

Subscribers: folly-diffs@, yfeldblum

FB internal diff: D1929881

Signature: t1:1929881:1426878091:5993d544c0f84bdbf8bcdd420ef01fb7bf88ac3f

Udp Acceptor

Summary:
major changes:

1) ServerSocketFactory and AsyncSocketBase to abstract the differences between UDP and TCP async socket.  Could possibly push some of this to the sockets themselves eventually
2) pipeline() is a pipeline between accept/receive of a UDP message, and before sending it to workers.  Default impl for TCP is to fan out to worker threads.  This is the same as Netty.  Since we don't know if the data is a TCP socket or a UDP message, it's a void*, which sucks (netty uses Object msg, so it isn't any different).

Test Plan: Added lots of new tests.  Doesn't test any data passing yet though, just connects/simple receipt of UDP message.

Reviewed By: hans@fb.com

Subscribers: alandau, bmatheny, mshneer, jsedgwick, yfeldblum, trunkagent, doug, fugalh, folly-diffs@

FB internal diff: D1736670

Tasks: 5788116

Signature: t1:1736670:1424372992:e109450604ed905004bd40dfbb508b5808332c15

fix service memory leak

Summary:
This code is only used in a test so far.

Basically it looks like Pipeline has some dependency issues - any addBack() pipelines have to be deleted *after* the main pipeline is destroyed.  Ideally if the pipeline is already closed, the destruction order wouldn't matter.  There is currently no removeBack() call either.   This is a quick fix for the test to just delete stuff in the right order, we can discuss a better solution when @jsedgwick returns

Test Plan: fbconfig --sanitize=address --with-project-version=clang:3.5 --clang folly/wangle/service && fbmake runtests

Reviewed By: hans@fb.com

Subscribers: trunkagent, doug, fugalh, folly-diffs@, yfeldblum, jsedgwick

FB internal diff: D1903200

Tasks: 6415578

Signature: t1:1903200:1426528298:e109dcc0ec586a505a26cd95a6f20434d22cbd37

Add comment about djb2 hash

Summary:
In my tests, using djb2 was much worse for any sizable workload. Just use SpookyHashV2.

Test Plan: comment only

Reviewed By: tudorb@fb.com

Subscribers: trunkagent, folly-diffs@, yfeldblum

FB internal diff: D1906439

Signature: t1:1906439:1426117575:f88b7b28b578092c1716433e21150d423d4e94a7

SocketAddress::anyAddress should be a method

Summary:
Calling getaddrinfo from a static constructor is not the most
reliable thing to do. Stop doing it.

Test Plan: existing tests

Reviewed By: davejwatson@fb.com

Subscribers: dancol, mzlee, shikong, kmdent, fma, bmatheny, benyluo, ranjeeth, folly-diffs@, jsedgwick, yfeldblum

FB internal diff: D1924578

Tasks: 6534662

Signature: t1:1924578:1426718586:d73bc2f001095e66fa77fde9c027af050cc26d2a

(Wangle) Clean up move constructors

Summary: I thought this might fix #6120972. But it doesn't. Still a bit of a cleanup in my opinion.

Test Plan: Run all the tests?

Reviewed By: hans@fb.com

Subscribers: trunkagent, folly-diffs@, jsedgwick, yfeldblum

FB internal diff: D1907259

Signature: t1:1907259:1426613567:9e33fe7e9e8a36ba006d4aee604086a56f128893

Support 'min ms to log'

Summary:
Sometimes we only want to log if enough time has passed, e.g. when logging
inside of a loop doing work to detect if things are slow. Add an option for
that.

Test Plan:
build and unit tests of folly/experimental

Reviewed By: lesha@fb.com

Subscribers: trunkagent, folly-diffs@, yfeldblum

FB internal diff: D1909055

Signature: t1:1909055:1426627731:6d3ad94fd71967ec89ffebc90cd2907dd1631d36

Fix logFormat

Summary: Ooops during port ;)

Test Plan: unit test

Reviewed By: agoder@fb.com

Subscribers: trunkagent, folly-diffs@, yfeldblum

FB internal diff: D1905639

Signature: t1:1905639:1426171227:a6e415046a85f098ac090f069476204fb0367c42

Bump version to 31:0

Fixed README download link to get the correct GTest version

Summary: Signed-off-by: Nicholas Ormrod <njormrod@fb.com>

Test Plan: Clicked link, gtest started downloading

Reviewed By: markisaa@fb.com

Subscribers: sdwilsh, folly-diffs@, yfeldblum

FB internal diff: D1901849

Tasks: 6255850

Signature: t1:1901849:1426097071:e887ed8749c70d5e15877082ac0175ce3b56647e

BackoffTimeout disallow copy, move, default ctor

Summary:
BackoffTimeout disallow copy construction, copy assignment, move
construction, move assignment, and default construction.

Closes https://github.com/facebook/folly/pull/146

Test Plan: runtests

Reviewed By: ldbrandy@fb.com

Subscribers: lins, anca, folly-diffs@, yfeldblum

FB internal diff: D1912521

Signature: t1:1912521:1426268413:6edb2d8203d1efd1e825816f03be115ac46e55ca

We might avoid some temporaries

Summary:
It seems we might avoid temporaries.  To do so we pass arguments
directly to constructors with the help of emplace_back member.

Test Plan:
all folly/tests, make check for 37 tests, passed.

Closes https://github.com/facebook/folly/pull/150

Reviewed By: ldbrandy@fb.com

Subscribers: lins, anca, folly-diffs@, yfeldblum

FB internal diff: D1912505

Signature: t1:1912505:1426268158:086882bb53f3d79c3f3b7b2c362318f09b5ee3be

AsyncServerSocket disallow copy, move, and default ctor

Summary:
AsyncServerSocket disallow copy construction, copy assignment,
move construction, move assignment, and default
construction.

Closes https://github.com/facebook/folly/pull/148

Test Plan: all folly/tests, make check for 37 tests, passed.

Reviewed By: ldbrandy@fb.com

Subscribers: lins, anca, folly-diffs@, yfeldblum

FB internal diff: D1912488

Signature: t1:1912488:1426267982:a9690ce9429016f89630f0ded64b7be9ff80c39d

fix typo: add backslash to Makefile.am

Summary: backslash

Test Plan: tests

Reviewed By: aap@fb.com

Subscribers: folly-diffs@, yfeldblum

FB internal diff: D1920872

Signature: t1:1920872:1426624366:c984fecb5bd0c292d5a8309fb320a48872fff6db

Blame Revision: D1890511

Add StringKeyedUnorderedMap to Makefile

Test Plan:
unit tests

Reviewed By: bwatling@fb.com

Subscribers: stepan, folly-diffs@, yfeldblum, trunkagent, vikas, alikhtarov

FB internal diff: D1890511

Tasks: 6311387

Signature: t1:1890511:1425924980:71af7c12a09a86b280dc3c9a9ef0756bf56bf639

(folly osx) test for the zip, not the dir

Summary: Apparently `configure` creates this subdirectory and then the curl and unzip commands don't run, and then `make check` fails. Instead, do it if the zip file doesn't exist.

Test Plan:
git clone https://github.com/facebook/folly.git
cd folly/folly
/path/to/this/bootstrap-osx-homebrew.sh
make
make check

Reviewed By: andrei.alexandrescu@fb.com

Subscribers: exa, folly-diffs@, yfeldblum, davejwatson

FB internal diff: D1882472

Signature: t1:1882472:1425920470:bdab25967d3b8e52c690d5627168077c65829b92

exceptionStr(exception_wrapper)

Summary: for consistency with exception and exception_ptr functions provided in String.h

Test Plan: unit

Reviewed By: vloh@fb.com

Subscribers: trunkagent, folly-diffs@, yfeldblum

FB internal diff: D1905087

Tasks: 5961362

Signature: t1:1905087:1426101740:670ceab5140250bbecdd1247025d3d70b7774690