AppVeyor provides vcpkg and a great guide explaining how to use it and cache the installed packages: https://www.appveyor.com/docs/lang/cpp/#vc-packaging-tool
However, we're currently relying on Azure Pipelines for the Windows builds and keeping one on AppVeyor would be a waste of CI resources.
This commit also changes the x86 build so that it uses the "amd64_x86" toolchain, for potentially better build performance.
The architecture is now appended to the installers.
The purpose of the configuration was to build installers and continually verify that MSVC 2008 could still build the project.
MSVC 2008 was a requirement because we wanted to maintain support for very old versions of Windows (9x).
During the past few years we encountered many annoying limitations due to that requirement, such as the lack of support for designated initializers.
We ended up deciding to only keep support for operating systems older than 7 in the stable repository.
As for the installers, we now build them on Azure Pipelines (currently with MSVC 2019).
The reason why we don't build these two targets is that they're not used.
More specifically: they require proper configuration to work correctly, which is currently missing.
While vpninstall may be worth salvaging, vpnweb is definitely a relict of the past because it relies on ActiveX.
vpndrvinst is the name of the target and thus the default output name, let's use it.
`vpndrvinst.exe" also sounds less fishy than "driver_installer.exe"...
BuildUtil compiles the project as 32 bit and 64 bit, before building the installer package.
64 bit binaries have the "_x64" suffix and are added to the package alongside the 32 bit ones (that have no suffix).
The CMake project compiles the binaries for a single architecture and they have no suffix.
We decided that providing two separate installers is the best solution.
As for the binaries with the "_ia64" suffix: they never existed during the this repository's lifespan.
The MSBuild project built the binary into "src/bin/hamcore", causing it to be added to "hamcore.se2".
As hinted by the name of the file ("vpnsetup_nosign.exe"), it is not signed by BuildUtil, possibly to save time (the setup package is signed).
The CMake project builds the binary in the same directory as the other ones, allowing the setup to install them without the need to build a package.
Previously, the file needed to be present in order for the setup to work.
This commit removes the requirement so that the setup can be ran from the build directory without the need to copy the file (which is now removed from the repository).
The inclusion of the headers is probably a very old leftover, from when OpenSSL was not encapsulated into Mayaqua yet.
In fact, there was a "HAM_C" (defined in vpndrvinst.c) definition check in Mayaqua/Encrypt.h preventing the redefinition of OpenSSL types.
When "VPN_EXE" is defined, Mayaqua.h defines WinMain(), which handles arguments in a special way.
This commit passes "WIN32" to add_executable(), so that WinMain() is used as entry point instead of main().
The use of main() instead of WinMain() was causing service mode not to work due to the "/service" argument being discarded.
Our CMake project used to forcefully create and use two different build directories: "build" and "tmp".
This commit changes the behavior so that only the build directory CMake is ran in is used.
The "configure" script now runs CMake in "build" by default, instead of "tmp".
strtok() and wcstok() are considered unsafe functions.
A segmentation fault caused by the use of strtok() was recently reported.
Co-authored-by: Takuho NAKANO <takotakot@users.noreply.github.com>
The SSTP implementation doesn't provide packet identification, because it's not required: the protocol is identified by the HTTP header it sends to the server.
"ClientOption", as the name implies, is only used in a client context.
The issue was introduced in 235bd07e67. Before that, an unrelated check prevented UnixVLanSetState() from being called in a server context.
