This fixes and closes#481
---
By creating this pull request, I understand that code that is AI
generated or otherwise automatically generated may be rejected without
further discussion.
I declare that I fully understand all code I pushed into this PR, and
wrote all this code myself and own the rights to this code.
for example, loading a library which returns a table of functions would
fail with an inexplicable (to the user) error about serialization. This
is now fixed. We no longer use a single result type, instead there are
void results and normal results, and the former simply never has to
worry about the result.
This was likely causing even more issues; if the `sol::object` was
populated before, it hitting the dtor in another path would, again, like
we've seen before, corrupt the lua stack.
You're supposed to .lock() instead of TOCTOU checking, of course.
Not sure what I was thinking when I built that. .lock() returns a
default constructed std::shared_ptr on error, which is `false` via
`operator bool`.
this massively improves thread safety and cleanly serializes accesses
into the lua engine's result objects where accesses before were
extremely unsafe and could access a corrupt/invalid stack.
this fixes various obscure crashes related to accessing results,
without changing any observable behavior.
between the time we check for `is_open` and the actual disconnect, the
socket could already have been disconnected by another thread (TOCTOU).
Furthermore, the disconnects can race causing a segfault or similar
issue in the asio's internals.
with this many http connnections, we were exhausting all available file
descriptors, leading to a dead server that keeps CLOSE_WAIT tcp sockets.
Because we want to retain the behavior that we keep connections open for
reuse, we instead make a pool of 8 curl instances now, shared between
all the different requests.
this was exhausting file descriptors with enough concurrent reads, from
what I can tell. Either way, spawning a new OS thread per read is not
the way.
Because this is so critical, I added unit-tests for that behavior.
this happens when, somehow, the client disconnects before we get here.
I had this happen when breaking in the debugger and continuing, which
leads to clients timing out (client-side timeouts).
By creating this pull request, I understand that code that is AI
generated or otherwise automatically generated may be rejected without
further discussion.
I declare that I fully understand all code I pushed into this PR, and
wrote all this code myself and own the rights to this code.
Removes continue in socket accept flow
---
By creating this pull request, I understand that code that is AI
generated or otherwise automatically generated may be rejected without
further discussion.
I declare that I fully understand all code I pushed into this PR, and
wrote all this code myself and own the rights to this code.
This pull request adds BEAMMP_PROVIDER_DISABLE_MP_SET variable
(true/false/1/0) to disable MP::Set for providers.
---
By creating this pull request, I understand that code that is AI
generated or otherwise automatically generated may be rejected without
further discussion.
I declare that I fully understand all code I pushed into this PR, and
wrote all this code myself and own the rights to this code.
This PR adds a boolean parameter to the MP.SendChatMessage function that
allows logging the message or not (default true).
I took an example from the ``set_function("SendNotification" ...`` to
make the default value working.
It's my first time actually doing C++, I hope it's alright!
---
By creating this pull request, I understand that code that is AI
generated or otherwise automatically generated may be rejected without
further discussion.
I declare that I fully understand all code I pushed into this PR, and
wrote all this code myself and own the rights to this code.
Lion Kortlepel
SaltySnail
Quentin Ritzler
Tixx
SaltySnail