Cross-compiling NetHack with Visual Studio from an x64 platform to an ARM64
target presents some new build challenges.
In the current nethack.sln solution, the build attempts to execute the
several just-built tools during the build of various subprojects.
For example, when cross-compiling on a typical Windows 11 x64 machine
to build a target ARM64 Windows 11 package, the build process tries to
run the following just-built target tools:
(under a Debug build)
"$(ToolsDir)\Debug\ARM64\uudecode.exe"
"$(ToolsDir)\Debug\ARM64\makedefs.exe"
"$(ToolsDir)\Debug\ARM64\tilemap.exe"
"$(ToolsDir)\Debug\ARM64\tile2bmp.exe"
"$(ToolsDir)\Debug\ARM64\dlb.exe"
(under a Release build)
"$(ToolsDir)\Release\ARM64\uudecode.exe"
"$(ToolsDir)\Release\ARM64\makedefs.exe"
"$(ToolsDir)\Release\ARM64\tilemap.exe"
"$(ToolsDir)\Release\ARM64\tile2bmp.exe"
"$(ToolsDir)\Release\ARM64\dlb.exe"
Those fail to execute successfully on Intel x64 (or x86) since they
are actually ARM64 executables, and the build attempts to execute
them on the host Intel x64 hardware.
The situation is a little different if the cross-compile is carried out
on a Windows 11 ARM64 machine (such as SnapDragon).
On an ARM64 machine, the cross-compile to build a target Intel x64
Windows 11 package, tries to execute the following:
(under a Debug build)
"$(ToolsDir)\Debug\x64\uudecode.exe"
"$(ToolsDir)\Debug\x64\makedefs.exe"
"$(ToolsDir)\Debug\x64\tilemap.exe"
"$(ToolsDir)\Debug\x64\tile2bmp.exe"
"$(ToolsDir)\Debug\x64\dlb.exe"
(under a Release build)
"$(ToolsDir)\Release\x64\uudecode.exe"
"$(ToolsDir)\Release\x64\makedefs.exe"
"$(ToolsDir)\Release\x64\tilemap.exe"
"$(ToolsDir)\Release\x64\tile2bmp.exe"
"$(ToolsDir)\Release\x64\dlb.exe"
Those actual do succeed in executing on ARM64, because of the
"prism emulation" that is available on Windows 11 ARM64 operating
systems to allow x64 and x86 executables to run.
The following change adds some detection to build environment, leading
to the definition of a "HostTools" macro that leads to the native host
tools for those steps.
There is a catch:
It means that the native build of the interim tools for Windows 11 must
be executed prior to attempting a cross-compile build to a non-native
target. That ensures that the native x64 interim uudecode, makedefs,
tilemap, tile2bmp and dlb tools are available on the disk for use by
a subsequent cross-compile.
This change consistently switches to the use of the host-native
interim tools for uudecode, makedefs, tilemap, tile2bmp and dlb tools.
Technically, this change would not be strictly necessary on an
ARM64-hosted build that was targeting x64 or x86, because of the
available prism-emulation that allows ARM64, x64, and x86 images to
execute, but this maintains consistency of the build process on
either platform. It is also likely that the native host versions execute
more quickly than versions requiring the prism emulation, although
that isn't really a concern for a NetHack build.
The use of native host uudecode, makedefs, tilemap, tile2bmp and
dlb tools is done with the Unix-hosted cross-compiles to other
target platforms as well (on Linux or macOS).
The previous code made incorrect assumptions about Qt6's file hierarchy. Reuse automatic configuration (with pkg-config) for Qt5, and replace '5' by '6'.
Also fix a typo in multiw-2.370.
If using hints file sys/unix/hints/linux.370 or sys/unix/hints/macOS.370
allow the majority of the boilerplate compile switches to reside in a
compiler response file, instead of on the command line.
Include one of the following on your make command line:
make response=1
or
make resp=1
It can be combined with other make command line options. See
sys/unix/README-hints for further information about those.
The response files that it uses are:
CC (clang or gcc) src/nethack_cc.rsp
CXx (clang++ or g++) src/nethack_cxx.rsp
Note: I think the reduced clutter should actually become the default,
and the override should be noresponse=1 to NOT use it, but I'm
not sure how others feel, so for now, it requires
make resp=1
Feedback on whether that should become the default or not
is welcome.
Tested on Linux with gcc-15 and on Linux with clang-20.
I haven't had a chance to test it on macOS yet.
OpenSUSE Tumbleweed ncurses 6.5 requires the one in
/usr/include/ncursesw/curses.h, if ncursesw is being used.
Otherwise, several needed function prototypes are not there.
Fixes#1427
nhmall