Big issue with IOCP vs. poll variants is that the overlapped
structures have a longer lifetime than intermediate state needed
for epoll. One cannot free overlapped structures after closing a
handle/socket, like one can do with any intermediate state when using
readiness-based IO.
This results in fewer system calls and presumably more effcient code. It
also brings the epoll (and kqueue) code more in line with how the
windows IOCP code works, incidentally.
Any references exclusively held by a weak table may be collected
without the programmer needing to free references manually. A table
can be setup to have weak keys, weak values, or both.
Make more use of the built in GC code for abstracts to
be sure things are more correct. Issue before was streams could
be freed before IOCP events arrived.
This calls setvbuf to change FILE buffering. A goal is
to be able to use the existing file/* functions for blocking
IO analogous to `read` and `write` system calls.
One is a way to export symbols, the other a way to reference
API functions. Also include prebuilt dlljanet.dll and dlljanet.lib
for windows to save people the trouble of compiling janet.c themselves.
Instead of setting a flag, each interrupt increments an atomic
counter. When the interrupt is finally handled, either by scheduling
code to run on the event loop or executing some out of band code, the
user must now decrement the interrupt counter with
janet_interpreter_interrupt_handled. While this counter is non-zero, the
event loop will not enter the interpreter. This changes the API a bit but
makes it possible and easy to handle signals without race conditions
or scheduler hacks, as the runtime can ensure that high priority code is
run before re-entering possibly blocking interpreter code again.
Also included is a new function janet_schedule_soon, which prepends to
the task queue instead of appending, allowing interrupt handler to skip
ahead of all other scheduled fibers.
Lastly, also update meson default options to include the
interpreter_interrupt code and raise a runtime error if os/sigaction
is used with interpreter interrupt but that build option is not enabled.
Threaded channels _can_ be marshalled, just not for communication
between threads. This is a special case since the same abstract type
is used for both threaded and non-threaded channels.
This allows for more flexible C interop from DLLs. Users can skip the
usual extension loading mechanism and manage function pointers manually
if they need to.
commit fbb0711ae1
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 12:07:55 2023 -0500
Distinguish between subprocess when testing.
commit 676b233566
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 11:59:17 2023 -0500
Hack for qemu based testing (also should work with valgrind)
commit d7431c7cdb
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 11:54:04 2023 -0500
Revert "Test removing 32bit ptr marshalling."
This reverts commit 566b45ea44.
commit 566b45ea44
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 11:52:22 2023 -0500
Test removing 32bit ptr marshalling.
commit ff2f71d2bc
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 11:42:10 2023 -0500
Conditionally compile marshal_ptr code.
commit bd420aeb0e
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 11:38:34 2023 -0500
Add range checking to bit-shift code to prevent undefined behavior.
commit b738319f8d
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 11:17:30 2023 -0500
Remove range check on 32 bit arch since it will always pass.
commit 7248626235
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 10:56:45 2023 -0500
Quiet some build warnings.
commit 141c1de946
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 10:50:13 2023 -0500
Add marshal utilities for pointers.
commit c2d77d6720
Merge: 677b8a6fff90b81e
Author: Calvin Rose <calsrose@gmail.com>
Date: Sat Jun 24 10:40:35 2023 -0500
Merge branch 'master' into armtest
commit 677b8a6f32
Author: Ico Doornekamp <ico@zevv.nl>
Date: Mon Jun 12 21:01:26 2023 +0200
Added ARM32 test
The use cases involve user-expandable grammars.
For example, consider the IRC nickname specification.
> They SHOULD NOT contain any dot character ('.', 0x2E).
> Servers MAY have additional implementation-specific nickname restrictions.
To implement this, we can do something along these lines:
```janet
(def nickname @{:main '(some :allowed)
:allowed (! (+ :forbidden/dot :forbidden/user))
# for lax mode, (put nickname :forbidden/dot false)
:forbidden/dot "."
# to add your own requirements
# (put nickname :forbidden/user 'something)
:forbidden/user false})
```
Additionally, it's common in parsing theory to allow matches of the
empty string (epsilon). `true` essentially allows for this.
Note that this does not strictly add new functionality, you could
emulate this previously using `0` and `(! 0)` respectively, but this
should be faster and more intuitive.
The speed improvement primarily comes from `(! 0)` which is now a single
step.
Previously, `*macro-lints*` was set after the `macroexpand1` fiber was
resumed, rather than just before. And, `*macro-lints*` was never
cleared. This behavior was typically fine since the main users of
`compile` pass the same lint array repeatedly, and the first macro
expansion (somewhere in boot.janet) never produces a lint. But, when
compiling with a fresh lint array, if the first macro invocation
produced a lint, the lint was always lost.
When suspended in `ev/give` or `ev/take`, closing the channel should
cause the result of `ev/give` or `ev/take` to be `nil`.
When suspended in `ev/select`, closing the channel should cause the
result of `ev/select` to be `[:close ch]`.
The results were flipped before.
When there is a bad arity function passed in to the fiber
constructor, return NULL so the runtime can choose what to do.
This is not the prettiest API but does work, and gives better error
messages for instance in the compiler.
Coming from commit 77189b6e66, relating
to changes in source mapping debug info, this caused a segfault when
traversing a stack frame where the birth_pc was incredibly large due
to wrap around. This fix prevents the wrap around and does saturating
subtraction to 0.
This removes unnecessary movn, movf, lds, and a few other instructions.
Any instructions that has not side effects and writes to a slot that
isn't used can be removed. A number of other optimizations can follow
from this:
- Implement the def-aliasing-var optimization better
- This function can be iterated as a fix point until no more
instructions are removed.
- If we implement slot renaming, then we no longer need to free slots
and can simplify the initial code generation a lot.
error: bad slot #1, expected string|symbol|keyword|buffer, got ...
error: bad slot #1, expected a string, symbol, keyword or buffer, got ...
bad s64 initializer: "donkey"
can not convert string "donkey" to s64
Otherwise attempts to use it on some platforms cause the following error
`error: initializer element is not a compile-time constant`
when attempting to use the corresponding `JANET_REG`.
By take and releasing locks twice per channel in the case where nothing
is reading, there was an opportunity for ev/select to hang in the
multithreaded case. Also silence valgrind/helgrind errors.
Like getcstring, but operates on a byteview.
When writing bindings (i.e what capi.c is primarily used for), it's
common to want to accept a buffer *or* a string rather than just
a string.
For this, a byteview is perfect (and why not accept keywords while
you're at it?).
However, there's no built-in function for getting a cstring out of
a byteview, this adds one.
This also reformulates getcstring to be an edge-case of getcbytes
(simply adding an explicit check for stringness).
This turns splices that are ignored into compiler errors. Other
alternatives here should also be considered, for example making this
a compiler warning rather than an error. For example, the latest
spork as of a3ee63c137ee3234987dbbca71b566994ff8ae8c has an error of this
kind, but the resulting program does work correctly.
Also disallow splice propagation - code of the
form (+ 1 (do ;[2 3 4]) 5).
These now have semantic menaings that are pretty difficult to
work around. Code that tries to maniuplate user8 and user9 signals
right now may be affected
When peg/replace or peg/replace-all are given a function to serve as the text
replacement, any captures produced by the PEG are passed as additional
arguments to that function.
Functions will be invoked with the matched text, and their result will be
coerced to a string and used as the new replacement text.
This also allows passing non-function, non-byteviewable values, which will be
converted into strings during replacement (only once, and only if at least
one match is found).
there was a request to improve the error message, but the whole function
has non-informative errors. (both functions, actually, since the code is
duplicated)
as such, instead of catching it directly, address the assumption that
led to the SIGSEGV and let it be caught by the functions themselves,
thus reusing existing error messages (which can then be improved
separately).
When there is no format to be found after a %, get_fmt_mapping returns
NULL. It then gets called against strlen, which is a typical SEGV.
Check for NULL aginst mapping, which signals a null format being
specified.
These functions are designed to make it easier to properly preserve the
sourcemap and tuple type in macros. This commit also modifies the threading
macros to make use of these functions.
While the old behavior was reasonable, it is not spelled out anywhere
in the documentation and was incidental rather than intentional.
Parameters of the same name of the function should probably take
precedence on name collision, following the principle of least surprise.
Comparison between different bracket and normal tuples
will now take into account the delimiter type. This solves strange
non-locality issues in the compiler due to this false equality, and is
more consistent with Janet's otherwise strong equality philosophy.
buffer/blit is difficult to use, and while buffer/push is the easiet
buffer manipulation function to use it only appends to the buffer.
buffer/push-at lets users manipulate buffers at any index - useful
for buffers used as an in-memory databases, for example.
Added underlying buffer support for buffer instances that cannot
reallocated underlying memory - useful for (small) memory mapped
files and other FFI utilties.
Doesn't really impart (much) file systtem information when used, and
can be used for a lot of things where file functions are used to process
in a stream.
