Upon investigating the extremely slow MSVC build times in #80513, I noticed
that while Godot policy is to never use exceptions, we weren't enforcing it
with compiler flags, and thus still included exception handling code and
stack unwinding.
This is wasteful on multiple aspects:
- Binary size: Around 20% binary size reduction with exceptions disabled
for both MSVC and GCC binaries.
- Compile time:
* More than 50% build time reduction with MSVC.
* 10% to 25% build time reduction with GCC + LTO.
- Performance: Possibly, needs to be benchmarked.
Since users may want to re-enable exceptions in their own thirdparty code
or the libraries they compile with Godot, this behavior can be toggled with
the `disable_exceptions` SCons option, which defaults to true.
- `_DEBUG` is MSVC specific so it didn't make much sense to define for
Android and iOS builds.
- iOS was the only platform to define `DEBUG`. We don't use it anywhere
outside thirdparty code, which we usually don't intend to debug, so it
seems better to be consistent with other platforms.
- Consistently define `NDEBUG` to disable assert behavior in both `release`
and `release_debug` targets. This used to be set for `release` for all
platforms, and `release_debug` for Android and iOS only.
- Due to the above, I removed the only use we made of `assert()` in Godot
code, which was only implemented for Unix anyway, should have been
`DEV_ENABLED`, and is in PoolAllocator which we don't actually use.
- The denoise and recast modules keep defining `NDEBUG` even for the `debug`
target as we don't want OIDN and Embree asserting all over the place.
(cherry picked from commit 26e9145c26)
This makes it possible to retrieve all relevant versioning info used to
generate `core/version_generated.gen.h` in the buildsystem.
Notably it makes the custom logic parsing the `GODOT_VERSION_STATUS`
environment variable to override status easy to reuse.
(cherry picked from commit d432fe38a9)
Adds support for LTO on macOS and Android.
Disable LTO by default on iOS even if `production=yes` is set.
Also add `linker` option to `server` platform missed in #63283.
Refactor code handling old arguments to make it simpler (breaks compat,
but is explicit enough about it and scripts are easy to fix).
This doesn't change the behavior when `--jobs`/`-j` is specified as a
command-line argument or in `SCONSFLAGS`.
The SCons hack used to know if `num_jobs` was set by the user is derived
from the MongoDB setup.
We use `os.cpu_count()` for portability (available since Python 3.4).
With 4 CPUs or less, we use the max. With more than 4 we use max - 1 to
preserve some bandwidth for the user's other programs.
(cherry picked from commit ea21122575)
We shouldn't presume that future compilers will not have false positives or
find new occurrences of this warning, which would break compiling old versions
of the engine without passing custom `CXXFLAGS`.
Follow-up to #60652.
(cherry picked from commit e55d30768a)
This reverts #53828 which had caused a significant drop in incremental
rebuild time for debug builds (from 10s to 23s on my laptop).
The "faster but unsafe" options are re-added, as well as adding
`max_drift=60` which we didn't use previously.
These options speed up SCons' own processing of the codebase to decide
what to build/rebuild (i.e. the first step before actually calling the
compiler). This will therefore not make much difference for scratch
builds, and is mostly useful for incremental rebuilds (including "null"
rebuilds with no change).
These options are enabled automatically for `debug` builds, unless
`fast_unsafe=no` is passed.
They are disabled by default for `release` and `release_debug` builds,
unless `fast_unsafe=yes` is passed.
(cherry picked from commit d4553c5126)
Debug builds are considerably slower than release builds or even
release_debug builds. `target=debug` is still the default SCons
target option, so unsuspecting users may be compiling unoptimized
debug builds for their personal use.
Change the entire navigation system.
Remove editor prefix from nav mesh generator class. It is now used for baking
at runtime as well.
Navigation supports obstacle avoidance now with the RVO2 library.
Nav system will also automatically link all nav meshes together to form one
overall complete nav map.
The Godot Project Leadership Committe has decided to update the sponsoring
tiers so that the Platinum Sponsorship no longer includes a logo on the
editor splash screen.
This lets us reclaim the editor splash screen space for community-related
content instead of sponsors (e.g. a different community-designed splash
screen for each stable branch?).
Also removes two Platinum Sponsors whose sponsorship has expired earlier this
year.
(cherry picked from commit c283fce698)
When disabling specific modules, misspellings can occur. Additionally,
when switching between the `3.x` and `master` branches frequently,
it's possible to forget about renamed modules such as `lightmapper_cpu`
versus `lightmapper_rd`.
(cherry picked from commit 69b2d3f791)
They haven't really helped save much time on incremental rebuilds, and they do
cause potential issues with build correctness (and possibly even one of the cause
for overly eager incremental rebuilds).
(cherry picked from commit 0433d0f54d)
Emscripten is LLVM-based so we want to follow the same logic. But we can't just
put it as a match in `methods.using_clang()` as that would mess with the
compiler version detection logic used to restrict old GCC and Clang releases.
(cherry picked from commit 34421683eb)
It's raised for us on many comparators implemented to be able to store a struct
in `Set` or `Map` (who rely on `operator<` internally). In the cases I reviewed
we don't actually care about the ordering and we use the struct's function
pointers as that's the only distinctive data available.
(cherry picked from commit 802810c371)
This fixes a regression from #46774 where `env["ENV"]` would miss some
important env variables on Windows, such as `SystemRoot`, `PATHEXT`, etc.
So we go back to the previous setup (letting SCons initialize `env["ENV"]`
as it sees fit for the host OS) but use `PrependENVPath` instead of
`AppendENVPath` to preserve the intended fix from #46774.
Fixes#46790 for 3.2.
We constructed the SCons environment without taking any (shell) environment
variables into account, and then appended a few, but too late. This would
cause variables like `env[CXX]` not to be properly expanded to respect a
non-standard `PATH`.
With this fix, setting:
```
PATH=$GODOT_SDK/bin:$PATH
```
will now properly use `$GODOT_SDK/bin/gcc` if available over `/usr/bin/gcc`.
(cherry picked from commit 5d217a9441)
The `dev=yes` and `production=yes` options work as aliases to set a number of
options, while still aiming to allow overriding specific options if the user
wishes so. (E.g. `production=yes use_lto=no` should work to enable production
defaults *but* disable LTO.)
That wasn't working as `ARGUMENTS.get()` returns a string and not a boolean as
expected by `BoolVariable`, and this wasn't flagged as a bug... So added a
helper method using SCons' `BoolVariable._text2bool` to do the conversion
manually.
This adds `custom_modules_recursive` which allows to detect and collect
all nested C++ modules which may reside in any directory specified by
`custom_modules` option.
The detection logic is made to be more strict because `SCSub` may be
used for organizing hierarchical builds within a module itself, so the
existence of `register_types.h` and `config.py` is checked as well
(these are all required for a C++ module to be compiled by Godot).
For performance reasons, built-in modules are not checked recursively,
and there's no benefit of doing so in the first place.
It's now possible to specify a directory path pointing to a *single*
module, as it may contain nested modules which are detected recursively.
(cherry picked from commit a3c2c1e18a)
This is meant for users making custom builds to match the options used on
optimized, official builds.
This enables, on the platforms which support them:
- `use_static_cpp=yes` (portable binaries for Linux and Windows)
- `use_lto=yes` (link time optimizations - note: requires a lot of RAM!)
- `debug_symbols=no` (no debug symbols, smaller binaries)
Also abort when using MSVC with `production=yes`, as:
- It cannot optimize the GDScript VM like GCC or Clang do, leading to
significant performance drops.
- Its LTO support is unreliable, at least used to trigger crashes last
we tried it extensively.
All options can still be overridden if specified, and the `dev=yes` option
was changed to also support overrides.
(cherry picked from commit db26871210)
Fixes a pre-existing bug that #44433 exposed.
It's pretty hacky, but we use `platform` in `env` both as an optional command
line option (instead it can be autodetected, or passed via the `p` alias, and
on Linux it might be overridden if you pass one of the convenience alias
values), and as the reference value for what platform we're building on.
Thus we override `env_base["platform"]` with the autodetected or validated
platform, but any call to `opts.Update(env_base)` overrides it with the
original command line option... causing e.g. #44448.
The proper fix would be to refactor all this so that we don't reuse
`env["platform"]` for platform detection (it could instead be e.g.
`env.platform` as a member variable which holds the validated value),
but for now I'm tapering over the immediate breakage.
Fixes#44448 and other breakages induced by #44433.
(cherry picked from commit 8f660393fe)
Otherwise we can get situations where platform-specific opts with the same name
can override each other depending on the order at which platforms are parsed,
as was the case with `use_static_cpp` in Linux/Windows.
Fixes#44304.
This also has the added benefit that the `scons --help` output will now only
include the options which are relevant for the selected (or detected) platform.
(cherry picked from commit 0f84d8dc49)
