* Optimize packet construction functions
Some of the functions that construct packets in
connection.cpp are using a const reference to get the raw
packet data to package and others use a value passed
parameter to do that. The ones that use the value passed
parameter suffer from performance hit as the rather bulky
packet data gets a temporary copy when the parameter is
passed before it lands at its final destination inside the
newly constructed packet. The unnecessary temporary copy
hurts quite badly as the underlying class (SharedBuffer)
actually allocates the space for the data in the heap.
Fix the performance hit by converting all of these value
passed parameters to const references. I believe that this
is what the author of the relevant code actually intended
to do as there is a couple of packet construction helper
functions that already use a const reference to get the
raw data.
* Optimize packet sender thread class
Most of the data sending methods of the packet sender thread
class use a value passed parameter for the packet data to be
sent. This causes the rather bulky data to be allocated on
the heap and copied, slowing the packet sending down. Convert
these parameters to const references to avoid the performance
hit.
* Optimize packet receiver thread class
The packet receiver and processor thread class has many
methods (mostly packet handlers) that receive the packed data
by value. This causes a performance hit that is actually
worse than the one caused by the packet sender methods
because the packet is first handed to the processPacket
method which looks at the packet type stored in the header
and then delegates the actual handling to one of the
handlers. Both, processPacket and all the handlers get the
packet data by value, leading to at least two unnecessary
copies of the data (with malloc and all the slow bells and
whistles of bulky classes).
As there already is a few methods that use a const reference
parameter for the packet data, convert all this value passed
packets to const references.
6125 is the time of first full light according to 'get_node_light()',
and the time of first full light visually when basic shaders are on.
This is the optimum default new world start time, taking all possible
games into account.
The previous time assumed a game similar to Minetest Game. Games
should set this setting themselves according to their needs.
It turns out there is no need to return the new value and
preserve the old one in random_turn, the procedure can be
made to modify the value in-place. This saves quite a bunch
of parameter and return value copying.
Previously, when basic shaders were enabled, the function
time_to_daynight_ratio() returned values jumping between 149 and 150
between times 4375 and 4625, and values jumping between 999 and 1000
between times 6125 and 6375, (and the corresponding times at sunset)
due to tiny float errors in the interpolation code.
This caused the light level returned by blend_light() to jump between
14 and 15, which became noticeable recently as those light levels were
given different visual brightnesses.
Add early returns to avoid the problematic interpolation, and to
avoid unnecessary running of the loop.
Makes the liquid waving shader per-nodedef like waving leaves/plants,
instead of being applied to all liquids.
Like the waving leaves/plants shaders, the liquid waving shader can
also be applied to meshes and nodeboxes.
Derived from a PR by t0ny2.
Like randomwalk caves, preserve nodes that have 'is_ground_content = false',
to avoid dungeons that generate out beyond the edge of a mapchunk destroying
nodes added by mods in 'register_on_generated()'.
Issue discovered by, and original PR by, argyle77.
This change permits to use up-to-date compilers, clang-tidy and
clang-format
It also refactor the tidy/format step to drop the binary selection from
scripts and perform it directly in travis
Shorter, simpler, clearer and more consistent with other mapgens,
while preserving functionality.
Base terrain shape is unchanged.
With the 'vary river depth' option disabled, river surface level
is unchanged.
Behaviour of the 4 heat/humidity/river depth options is very
slightly changed due to bugfixes and code cleanup (the mapgen is
'unstable').
Apply heat and humidity gradients above water_level instead of
above y = 0.
I removed the MapNode constructor which takes a nodename and gives the node's id or CONTENT_IGNORE
The code which used this constructor (two places) now handles the situation of not registered nodes correctly:
* minetest.set_node and similar functions make minetest crash when a not registered node is passed
* reverting a node with rollback aborts if the node is not registered
Use "append" method to construct the various game paths
instead of wasteful string concatenation. Additionally, use a
temporary to extract and reuse a result of a few common
subexpressions to further reduce the overhead.
The "what" parameter is being passed by value, most likely by
accident as the type is "const std::string". Convert it to a
reference by adding the missing "&".
* Fix color command line parameter ignorance
* coloured log: Support detecting the tty on windows
* Print an error message when setting something invalid as color mode instead of silently using mode never
* Revert "coloured log: Support detecting the tty on windows"
This reverts commit 4c9fc6366487ac0e6799e181796ca594797bb6f8.
It didn't work for travis and belongs to a separate PR
* Allow adjusting the log color with an environment variable
If --color is not passed to minetest, is used to decide on the log colorization.
Minetest settings can not be used instead of an environment variable because logs may appear before loading them.
* fix empty if body
This loop makes multiple passes over m_stack (type std::list) in order to remove all elements with a specified value. Replacing the loop with a call to std::list::remove does the same job, but in only one pass.
This sends the following header to a remote media server:
Referer: minetest://<server_name>:port
This was verified with CTF and the Minetest Public Remove Media
server. If the servername was a plain IPv6 address it will
contain `:` characters and will be encapsulated in `[]` to
be a valid URI.
The getS16NoEx() handler will return true unless there is a
`[num_emerge_threads]` line in the `minetest.conf` at which
point the excption handler part is reached. Due to the fact that
`defaultsettings.cpp` has a default value set for this setting,
that never will happen.
Because of this, the code will never check the number of threads on
the system, and keep `nthreads = 0`. If that happens, the value is
changed to `1` and only 1 emerge thread will be used.
The default should be set to `1` instead, due to the potential unsafe
consequences for the standard sqlite map files, but that should be a
separate commit that also adds documentation for that setting. This
commit focuses on removing this `hiding` bug instead.
* Drop the ID mapper, use a big u64 instead. This will permit to resync server ids properly with the manager code
* Modernize some code parts (std::unordered_map, auto)
* generate id on client part on U32_MAX + 1 ids, lower are for server ids
The reverted commit 968ce9af598024ec71e9ffb2d15c3997a13ad754
is suspected (through the use of bisection) of causing network slowdowns.
Revert for now as we are close to release.
Store the rotation in the node as a 4x4 transformation matrix internally (through IDummyTransformationSceneNode), which allows more manipulations without losing precision or having gimbal lock issues.
Network rotation is still transmitted as Eulers, though, not as matrix. But it will stay this way in 5.0.
Previously, when using 'place on vmanip' to add a schematic to a
lua voxelmanip, if part of the schematic was outside the voxelmanip
volume, the outside part would often appear in a strange place
elsewhere inside the voxelmanip instead of being trimmed off.
This was due to the out-of-bounds check checking the index.
A position outside the voxelmanip can have an index that satisfies
'0 <= index <= voxelmanip volume', causing the node to be placed
at a strange position inside the voxelmanip.
Use 'vm->m_area.contains(pos)' instead.
Move index calculation to later in the code to optimise.
If a formspec is submitted from a form fields handling
callback of another form (or "formspec shown from another
formspec"), the fields submitted for it can get
rejected by the form exploit mitigation subsystem with a
message like "'zorman2000' submitted formspec
('formspec_error:form2') but server hasn't sent formspec to
client, possible exploitation attempt" being sent to logs.
This was already reported as #7374 and a change was made
that fixed the simple testcase included with that bug
report but the bug still kept lurking around and popping
out in more complicated scenarios like the advtrains TSS
route programming UI.
Deep investigation of the problem revealed that this
sequence of events is entirely possible and leads to the
bug:
1. Server: show form1
2. Client *shows form1*
3. Client: submits form1
4. Server: show form2
5. Client: says form1 closed
6. Client *shows form2*
7. Client: submits form2
What happens inside the code is that when the server in
step 4 sends form2, the registry of opened forms is
updated to reflect the fact that form2 is now the valid
form for the client to submit. Then when in step 5 client
says "form1 was closed", the exploit mitigation subsystem
code deletes the registry entry for the client without
bothering to check whether the form client says was
closed just now is indeed the form that is recorded in
that entry as the valid form. Then later, in step 7 the
client tries to submit its valid form fields, these will
be rejected because the entry is missing.
It turns out the procedure where the broken code resides
already gets the form name so a simple "if" around the
offending piece of code fixes the whole thing. And
advtrains TSS agrees with that.
Reserve space for the list of games in findWorldSubgame. The
performance gain is pretty much negligible but this change
also gets rid of a performance warning by CLANG TIDY.
This patch will make distinguishable mods in modpacks possible in the future
`nil` checks are required to provide backwards-compatibility for fresh configured worlds
The craft definition handling code that collects the names of
the craftable nodes suffers from vector reallocation
performance hits, slowing down instances with lots of
crafting recipes (VanessaE's DreamBuilder and most public
server some to my mind when thinking about this). As in each
instance the size of the resulting vector is already known,
add a reserve() call before the offending loops to allocate
the needed chunk of memory within the result vector in one
go, getting rid of the overhead.
The pathfinder needs quite a bunch of items to add to the
resulting list. It turns out the amount of the space needed
for the finalized path is known in advance so preallocate it
to avoid a burst of reallocation calls each time something
needs to look for a path.
* Fix a crash on Android with Align2Npot2
glGetString can be NULL. If stored in a string it triggers a SIGSEGV.
Instead do a basic strstr and verify the pointer
* Better Align2Npot2 check (+ performance)
* Fix various bugs (Anticheat, Lua helpers)
Anticheat: Use camera position instead of player position for shoot line calculations
Lua helpers: Increase 'i' to not overwrite earlier added table values
* Remove lag compensation
* * 1.5 for larger selection boxes
