it is more about memory accounting sake. At allocation time,
we make clear we re possibly reusing regions marked as reusable.
Noted also calls might not necessarily succeed at first so we do
only when necessary.
With this patch, TracePoint receives a `:fiber_switch` event for
_almost_ every fiber switch. Previously, it would not be sent when an
exception was going to be raised. Now the event should only be blockable
by an interrupt (including `Thread#raise`) or a fatal error.
Additionally, interrupts will now be checked on the return fiber
_before_ re-raising the terminating unhandled exception. And a fiber
that terminates with an unhandled exception no longer creates a pending
interrupt on its thread. The exception will be raised in the return
fiber the same way as `Fiber#raise`: using `cont.value` with `cont.argc
== -1`
I moved `rb_exc_raise` from `fiber_store` to the end of `fiber_switch`
after _all_ of the other cleanup code: `fiber_stack_release`,
`th->blocking` increment, `RUBY_VM_CHECK_INTS`, and `EXEC_EVENT_HOOK`.
It seems to me that skipping those other cleanup steps may have also
resulted in other bugs.
* Rename `rb_scheduler` to `rb_fiber_scheduler`.
* Use public interface if available.
* Use `rb_check_funcall` where possible.
* Don't use `unblock` unless the fiber was non-blocking.
* Document Fiber's method related to scheduling;
* Extend Fiber's class docs with concepts of non-blocking
fibers;
* Introduce "imaginary" (documentation-only) class
Fiber::SchedulerInterface to properly document how
scheduler's methods should look.
This automatically choosess whether to use transfer on a transferring
fiber or resume on a yielding fiber. If the fiber is resuming, it
raises a FiberError.
This has been a TODO since 79df14c04b. While adcf0316d1 covered the
root_fiber of the initial thread, it didn't cover root_fibers of other
threads. Now it's hooked properly in rb_threadptr_root_fiber_setup.
With regards to "XXX: Is this mjit_cont `mjit_cont_free`d?", when
rb_threadptr_root_fiber_release is called, although I'm not sure when
th->root_fiber is truthy, fiber_free seems to call cont_free and
mjit_cont_free. So mjit_conts of root_fibers seem to be freed properly.
To make some kind of Ractor related extensions, some functions
should be exposed.
* include/ruby/thread_native.h
* rb_native_mutex_*
* rb_native_cond_*
* include/ruby/ractor.h
* RB_OBJ_SHAREABLE_P(obj)
* rb_ractor_shareable_p(obj)
* rb_ractor_std*()
* rb_cRactor
and rm ractor_pub.h
and rename srcdir/ractor.h to srcdir/ractor_core.h
(to avoid conflict with include/ruby/ractor.h)
Using Fiber#transfer with Fiber#resume for a same Fiber is
limited (once Fiber#transfer is called for a fiber, the fiber
can not be resumed more). This restriction was introduced to
protect the resume/yield chain, but we realized that it is too much
to protect the chain. Instead of the current restriction, we
introduce some other protections.
(1) can not transfer to the resuming fiber.
(2) can not transfer to the yielding fiber.
(3) can not resume transferred fiber.
(4) can not yield from not-resumed fiber.
[Bug #17221]
Also at the end of a transferred fiber, it had continued on root fiber.
However, if the root fiber resumed a fiber (and that fiber can resumed
another fiber), this behavior also breaks the resume/yield chain.
So at the end of a transferred fiber, switch to the edge of resume
chain from root fiber.
For example, root fiber resumed f1 and f1 resumed f2, transferred to
f3 and f3 terminated, then continue from the fiber f2 (it was continued
from root fiber without this patch).
If a fiber and thread are collected at the same time, the thread might
get collected first and the pointer on the fiber will go bad. I don't
think we need to check whether or not this is the main fiber in order to
release its stack
* Enables Mutex to be used as synchronization between multiple Fibers
of the same Thread.
* With a Fiber scheduler we can yield to another Fiber on contended
Mutex#lock instead of blocking the entire thread.
* This also makes the behavior of Mutex consistent across CRuby, JRuby and TruffleRuby.
* [Feature #16792]
This commit introduces Ractor mechanism to run Ruby program in
parallel. See doc/ractor.md for more details about Ractor.
See ticket [Feature #17100] to see the implementation details
and discussions.
[Feature #17100]
This commit does not complete the implementation. You can find
many bugs on using Ractor. Also the specification will be changed
so that this feature is experimental. You will see a warning when
you make the first Ractor with `Ractor.new`.
I hope this feature can help programmers from thread-safety issues.
This removes the warnings added in 2.7, and changes the behavior
so that a final positional hash is not treated as keywords or
vice-versa.
To handle the arg_setup_block splat case correctly with keyword
arguments, we need to check if we are taking a keyword hash.
That case didn't have a test, but it affects real-world code,
so add a test for it.
This removes rb_empty_keyword_given_p() and related code, as
that is not needed in Ruby 3. The empty keyword case is the
same as the no keyword case in Ruby 3.
This changes rb_scan_args to implement keyword argument
separation for C functions when the : character is used.
For backwards compatibility, it returns a duped hash.
This is a bad idea for performance, but not duping the hash
breaks at least Enumerator::ArithmeticSequence#inspect.
Instead of having RB_PASS_CALLED_KEYWORDS be a number,
simplify the code by just making it be rb_keyword_given_p().
Saves comitters' daily life by avoid #include-ing everything from
internal.h to make each file do so instead. This would significantly
speed up incremental builds.
We take the following inclusion order in this changeset:
1. "ruby/config.h", where _GNU_SOURCE is defined (must be the very
first thing among everything).
2. RUBY_EXTCONF_H if any.
3. Standard C headers, sorted alphabetically.
4. Other system headers, maybe guarded by #ifdef
5. Everything else, sorted alphabetically.
Exceptions are those win32-related headers, which tend not be self-
containing (headers have inclusion order dependencies).
Fiber#transfer prevents calling Fiber#resume on the receiver of the
transfer method, not the fiber calling transfer.
Transfering back to a fiber does not allow later calling resume on
the fiber. Once transfer has been called on a fiber, you can never
call resume on the fiber.
Calling resume on a transferred fiber is not a double resume error,
it is a different FiberError (cannot resume transferred Fiber).
For details on the differences between transferred fibers and
regular fibers, see Sasada-san's RubyKaigi 2017 presentation (in
short, Fiber#transfer is for coroutine, Fiber#resume is for
semi-coroutine).
If a fiber is invoked with transfer method (such as "f.transfer"),
then the invoked fiber ("f") is labeled as "transferred" and this
fiber can not be invoked with Fiber#resume. This patch adds
transferred attribute for "Fiber#to_s" (and inspect).
Fiber#transfer previously made it impossible to resume the fiber
if it was transferred to (no resuming the target of Fiber#transfer).
However, the documentation specifies that you cannot resume a fiber
that has transferred to another fiber (no resuming the source of
Fiber#transfer), unless control is transferred back.
Fix the code by setting the transferred flag on the current/source
fiber, and unsetting the transferred flag on the target fiber.
Fixes [Bug #9664]
Fixes [Bug #12555]
I've been compiling with:
```
set -lx cflags '-std=c99 -Werror=pedantic -pedantic-errors'
```
But compilation would fail with the following:
```
cont.c:296:90: error: format specifies type 'void *' but the argument has type 'struct fiber_pool_stack *' [-Werror,-Wformat-pedantic]
if (DEBUG) fprintf(stderr, "fiber_pool_stack_alloca(%p): %"PRIuSIZE"/%"PRIuSIZE"\n", stack, offset, stack->available);
~~ ^~~~~
cont.c:467:24: error: format specifies type 'void *' but the argument has type 'struct fiber_pool *' [-Werror,-Wformat-pedantic]
count, fiber_pool, fiber_pool->used, fiber_pool->count, size, fiber_pool->vm_stack_size);
^~~~~~~~~~
cont.c:588:83: error: format specifies type 'void *' but the argument has type 'struct fiber_pool_vacancy *' [-Werror,-Wformat-pedantic]
if (DEBUG) fprintf(stderr, "fiber_pool_stack_acquire: %p used=%"PRIuSIZE"\n", fiber_pool->vacancies, fiber_pool->used);
~~ ^~~~~~~~~~~~~~~~~~~~~
cont.c:736:76: error: format specifies type 'void *' but the argument has type 'rb_fiber_t *' (aka 'struct rb_fiber_struct *')
[-Werror,-Wformat-pedantic]
if (DEBUG) fprintf(stderr, "fiber_stack_release: %p, stack.base=%p\n", fiber, fiber->stack.base);
```
This commit just fixes the pedantic errors
The kw_splat flag is whether the original call passes keyword or not.
Some types of methods (e.g., bmethod and sym_proc) drops the
information. This change tries to propagate the flag to the final
callee, as far as I can.
After 5e86b005c0, I now think ANYARGS is
dangerous and should be extinct. This commit deletes ANYARGS from
rb_proc_new / rb_fiber_new, and applies RB_BLOCK_CALL_FUNC_ARGLIST
wherever necessary.
After 5e86b005c0, I now think ANYARGS is
dangerous and should be extinct. This commit deletes ANYARGS from
rb_ensure, which also revealed many arity / type mismatches.
https://github.com/ruby/ruby/pull/2170#issuecomment-489880700
Documentation is for those who don't know, remember, or understand (to any degree) the language, it should attempt to be clear above all other things. The example given is needlessly unclear because if you use a block it's common for arguments to be reused on every entry to the block. In Fiber's case this is not so.
First time round 10 goes in, 12 comes out.
Second time round 14 goes in, 14 comes out… was that because 14 is 12 + 2 or because it's "the return value of the call to Fiber.yield". It's the latter because it says so but why does the example need to make anyone think the former?
Using different numbers makes it immediately clear what's happening whether the description is there or not.
Renaming this function. "No pin" leaks some implementation details. We
just want users to know that if they mark this object, the reference may
move and they'll need to update the reference accordingly.
After calling `fiber_pool_vacancy_reset`, `vacancy->stack` and `stack` are
no longer in sync. Therefore, `fiber_pool_stack_free(&vacancy->stack)` can
do the wrong thing and clobber the vacancy data.
Additionally, when testing using VM_CHECK_MODE > 0, use MADV_DONTNEED if
possible, to catch issues w.r.t. clobbered vacancy data earlier.
fiber->cont.saved_ec.cfp should be initialized by NULL
because no vm_stack is allocated. However, cont_init()
captures current Fiber's cfp for continuation, so it should
only initialize fibers.
`cont_init` didn't initialize `cont->saved_ec.cfp`. Calling `cont_mark`
would result in an invalid `cfp` in `rb_execution_context_mark`. Because
fibers lazy-initialize the stack, fibers that are created but not resumed
could cause this problem to occur.
If `mmap` fails to allocate memory, try half the size, and so on.
Limit FIBER_POOL_ALLOCATION_MAXIMUM_SIZE to 1024 stacks. In typical
configurations this limits the memory mapped region to ~128MB per
allocation.
We use COROUTINE_LIMITED_ADDRESS_SPACE to select platforms where address
space is 32-bits or less. Fiber pool implementation enables more book
keeping, and reduces upper limits, in order to minimise address space
utilisation.
`madvise(free)` and similar operations are good because they avoid swap
usage by clearing the dirty bit on memory pages which are mapped but no
longer needed. However, there is some performance penalty if there is no
memory pressure. Therefore, we do it by default, but it can be avoided.
On 32-bit platforms, expanding the fiber pool by a large amount may fail,
even if a smaller amount may succeed. We limit the maximum size of a single
allocation to maximise the number of fibers that can be allocated.
Additionally, we implement the book-keeping required to free allocations
when their usage falls to zero.
Replace previous stack cache with fiber pool cache. The fiber pool
allocates many stacks in a single memory region. Stack allocation
becomes O(log N) and fiber creation is amortized O(1). Around 10x
performance improvement was measured in micro-benchmarks.
nagachika
Nobuyoshi Nakada
Samuel Williams
Ryuta Kamizono
David CARLIER
Kazuhiro NISHIYAMA
nicholas a. evans
Marcus Stollsteimer
Marc-Andre Lafortune
zverok
Delton Ding
Takashi Kokubun
Koichi Sasada
Aaron Patterson
Benoit Daloze
Jeremy Evans
卜部昌平
Lourens Naudé
Yusuke Endoh
Iain Barnett
git