There are no feature-dependent fields left in the spec. (I've also
made a PR for the spec to remove the tooling for it there).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We need to check that the key is valid for two reasons:
1) towire_ext_key() aborts if the key is invalid
2) fromwire_ext_key() doesn't check the parsed key for validity
Since bip32_key_get_fingerprint() fails if the key is invalid, we can
call it first to guarantee the key is valid before serializing.
By using fuzzer instrumentation for dependencies, we get more coverage
signal during fuzzing. This is useful when the fuzzer must figure out
how to take certain branches in a dependency.
In our case, the fuzz-bip32 target was failing to create a data buffer
that successfully passed fromwire_ext_key() parsing because the fuzzer
couldn't see what was happening inside libwally-core.
Without this, a stuck test (such as before the previous commit, where a plugin crashed
when running a command) simply gets timed out by the full CI timeout.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
When we release too fast, the plugin crashes:
```
thread 'tokio-runtime-worker' panicked at 'called Result::unwrap() on an Err value: SendError(())', plugins/examples/cln-plugin-reentrant.rs:31:27
note: run with RUST_BACKTRACE=1 environment variable to display a backtrace
```
This happens with CI under VALGRIND!
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This is a regression that we introduced in this release
due to some dirty parts of our codebase.
For historical reasons (I think), we were using a `json_add_sat_only`
procedure defined in `peer_control.c`. So when @rustyrussell removed the _msat,
we thought that all the fields were reflecting the new behavior, but
we were wrong.
This PR fixes this bug and also removes the additional function
from `peer_control.c`. This way, we can be sure that there is no other part
of our codebase that uses this method (except for other `json_add` methods).
Link: https://github.com/ElementsProject/lightning/issues/6244
Reported-by: @hMsats
Signed-off-by: Vincenzo Palazzo <vincenzopalazzodev@gmail.com>
We were using per-type overrides which caused some asymmetries, where
conversions could end up dropping fields as we went along. Essentially
each conversion would need to override a superset of the previous one,
which then caused issues when attempting to close the loop. By
overriding on the model level we ensure that all representations are
equivalent and convertible into one another, at the expense of
overriding a bit more aggressively, which should be fine anyway.
We use overrides that omit fields in some cases, which makes the
conversion lossy. This also means that until we complete the mapping
we can't reconvert back.
gcc 13 add an extra check for the enum in the definition
of a method. In our case the code was failing with the
following error, and the compiler is right, our definition
is different from the implementation.
```
$ make
CC: cc -DBINTOPKGLIBEXECDIR="../libexec/c-lightning" -Wall -Wundef -Wmissing-prototypes -Wmissing-declarations -Wstrict-prototypes -Wold-style-definition -Werror -Wno-maybe-uninitialized -Wshadow=local -std=gnu11 -g -fstack-protector-strong -Og -I ccan -I external/libwally-core/include/ -I external/libwally-core/src/secp256k1/include/ -I external/jsmn/ -I external/libbacktrace/ -I external/gheap/ -I external/x86_64-redhat-linux/libbacktrace-build -I external/libsodium/src/libsodium/include -I external/libsodium/src/libsodium/include/sodium -I external/x86_64-redhat-linux/libsodium-build/src/libsodium/include -I . -I/usr/local/include -DSHACHAIN_BITS=48 -DJSMN_PARENT_LINKS -DCOMPAT_V052=1 -DCOMPAT_V060=1 -DCOMPAT_V061=1 -DCOMPAT_V062=1 -DCOMPAT_V070=1 -DCOMPAT_V072=1 -DCOMPAT_V073=1 -DCOMPAT_V080=1 -DCOMPAT_V081=1 -DCOMPAT_V082=1 -DCOMPAT_V090=1 -DCOMPAT_V0100=1 -DCOMPAT_V0121=1 -DBUILD_ELEMENTS=1 -c -o
LD: cc -Og config.vars -Lexternal/x86_64-redhat-linux -lwallycore -lsecp256k1 -ljsmn -lbacktrace -lsodium -L/usr/local/include -lm -lgmp -lsqlite3 -lz -o
cc plugins/spender/multifundchannel.c
plugins/spender/multifundchannel.c:71:6: error: conflicting types for ‘fail_destination_msg’ due to enum/integer mismatch; have ‘void(struct multifundchannel_destination *, enum jsonrpc_errcode, const char *)’ [-Werror=enum-int-mismatch]
71 | void fail_destination_msg(struct multifundchannel_destination *dest,
| ^~~~~~~~~~~~~~~~~~~~
In file included from plugins/spender/multifundchannel.c:13:
./plugins/spender/multifundchannel.h:263:6: note: previous declaration of ‘fail_destination_msg’ with type ‘void(struct multifundchannel_destination *, int, const char *)’
263 | void fail_destination_msg(struct multifundchannel_destination *dest,
| ^~~~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
make: *** [Makefile:307: plugins/spender/multifundchannel.o] Error 1
```
The gcc 13 is not released yet, but fedora beta is out for public testing,
so it is useful fix this error in this release candidate cycle.
Changelog-Fixed: Build: Compilation with upcoming gcc 13
Reported-by: @grubles
Link: https://github.com/ElementsProject/lightning/issues/6175
Signed-off-by: Vincenzo Palazzo <vincenzopalazzodev@gmail.com>
After the first iteration of the loop, we call memmem with a buflen that
points past the end of buf.
In practice we probably never read the uninitialized memory since we
guarantee the buffer ends with "\r\n", and since most/all libc
implementations probably read the haystack sequentially. But maybe
there's some libc with a crazy optimization out there. It's good to use
an accurate buflen just in case.
Discovered this while running some unit tests with MSan.
Avoids failing the test with the pip warning:
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv
reported by: @ksedgwic
Changelog-None
Fixes a bug in installer registration where executable is evaluated
before entrypoints and other details are added.
***RECKLESS STDERR***
Traceback (most recent call last):
File lightning/tools/reckless, line 382, in <module>
INSTALLERS['nodejs'].add_entrypoint('{name}')
KeyError: 'nodejs'
Reported by @ksedgwic
Changelog-None
When enabling or disabling a plugin, the entrypoint is inferred
from the user provided name. A canonical name should be used, which
the installer entrypoint formats help to determine (this generally strips
the file extension if one is provided.)
Also adds a timeout when testing a plugin. Previously the behavior
of pyln-client was relied upon to exit if not communicating with
lightningd, however, this behavior is not universal.
Changlelog-Changed: reckless now installs node.js plugins
Also removes support for pip editable install using pyproject.toml
`pip install -e .` This was a fallback method when a requirements
file was not present, but was hacky and often failed anyway.
reckless: remove installation via pyproject.toml
This method relied on pip install in editable mode (hacky) and often
failed to complete anyhow. We should instead encourage a requirements
file to be created/used for user installation.
Rusty Russell
Matt Morehouse
fanquake
Amin Bashiri
daywalker90
Shahana Farooqui
Vincenzo Palazzo
Christian Decker
Alex Myers
Greg Sanders