If installed on the system curl will try linking to it dynamically,
which we definitely don't want. If not installed, nghttp2 support would
get disabled, which is now always the case with this commit.
cmake 3.24 defaults to ignoring all archive timestamps, which completely
breaks autotools packages (which rely on timestamps to know when they
need to be autoreconf'd).
Override this stupid default.
openssl is a miserable dependency to fight with, especially for
iOS/Android, and we use it for very little:
- it gets (mis)used to write base64 data to a file in wallet2 for things
like multisig data. However this mode is *not* enabled by default,
basically completely unknown, completely unused, only exists in the
cli wallet, and is just dumb. (Honestly the justification given in
the PR is that "Windows users might want it", presupposing that there
exists Windows users who are capable of generating a multisig wallet
in a CLI-only application and yet are incapable of dealing with binary
files).
- it's a dependency of unbound (in order to do dnssec, I believe).
Unbound itself is fairly useless for Oxen, so I've removed it too:
- it does OpenAlias lookups, which are a Monero thing that has never
been used outside Monero, doesn't work reliably (because it fails
if the result isn't DNSSEC validated) and is pointless when we
have ONS.
- it does DNS resolution on seed nodes, but we have never set seed
nodes by name and just use seed node IPs instead (which seems a
bit better anyway since the DNS lookup leaks some metadata).
- it *was* being used for sha256, but an earlier commit in this PR
already replaced that with libsodium (entirely coincidentally).
- for static deps, it enables HTTPS support for the wallet. However
only the CLI wallet actually supports this (the GUI and mobile wallets
don't), and since oxend hasn't support this for a while I have strong
doubts it is being used anywhere. (Plus wallet3 will do everything
encrypted using zmq/libsodium, so doesn't need this to be secure).
Note here that it is *only* removed by this commit for static builds:
if doing a system build that links to libcurl supporting HTTPS then
HTTPS support will still work.
Libexpat is also gone because it was only there for libunbound.
- bump libusb version
- build hidapi using cmake (autoconf is deprecated, and doesn't properly
set up the required linking on Windows)
- pass through toolchain
- zmq/win32 compilation fix
This communicates with the Ledger over TCP, which is what the ledger
emulator requires.
To use, specify:
--hw-device LedgerTCP --hw-device-address localhost:9999
to the wallet command-line arguments.
bootstrap.sh seems completely broken in boost 1.76.0 when you want to
use something other than the default compiler: it doesn't respect CXX
anymore, and if you give it --cxx to specify the CXX compiler it
produces a broken project-config.bjam file that prevents the project
from building.
Just skip that crap and build and use b2 ourselves.
Trezor support currently doesn't work (Trezor's hardware wallet is
hard-coded for a Monero net-id), and this saves needing to muck around
with protobuf.
BUILD_GUI_DEPS was dumb: it should have just been a non-"all" target (as
it is now) that can be explicitly built, rather than misusing a cmake
option.
It was doing one useful thing, however, of turning on FORCE_USE_HEAP --
but that was wrong too, because that needs to be enabled when doing an
android build (and probably IOS as well).
- Various static deps updated needed for properly cross-compiling
libraries for android.
- disable LTO because it doesn't work reliably with the android NDK.
- allow building without miniupnpc, and default to not building it under
android.
- make sure we build translation tools on the native arch when
cross-compiling for android.
- don't build ncurses, libusb, hidapi when doing an android build
Converts all use of boost::filesystem to std::filesystem.
For macos and potentially other exotic systems where std::filesystem
isn't available, we use ghc::filesystem instead (which is a drop-in
replacement for std::filesystem, unlike boost::filesystem).
This also greatly changes how we handle filenames internally by holding
them in filesystem::path objects as soon as possible (using
fs::u8path()), rather than strings, which avoids a ton of issues around
unicode filenames. As a result this lets us drop the boost::locale
dependency on Windows along with a bunch of messy Windows ifdef code,
and avoids the need for doing gross boost locale codecvt calls.
Jason Rhinelander
Thomas Winget