Over time, it has cost us more developer cycles than it has gained.
It has hidden intermittant bugs, and allowed cruft to accumulate:
when we eventually tried to figure out what was going wrong, the
actual change which caused it was now stale and forgotten.
This was a particular bane during the connectd rewrite, and I
worked through some issues which had occurred before, but were not
more likely.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
So this was quite a journey:
- We want relative depdendencies (using the `path` argument) whenever
developing locally. Otherwise we would have to install each
dependency every time we change a single character, which
undoubtedly would cause us to waste time trying to debug an issue
just because we forgot to install.
- When publishing however we want to rely on the version number,
since the repo context gets lost upon publishing, and path
dependencies cause failures.
The solution then it seems is to use `dev-dependencies` (not that
surprising once you find it) with relative paths, so that `poetry
install` uses these over the normal dependencies (no idea how they
dedup them) and use `dependencies` when publishing. The paths are
still in there when publishing, but `pip install` ignores them.
I checked that `poetry install` from an unrelated project doesn't
accidentally use the path dependencies, even when adding them as
dev-dependencies. This should hopefully also allow installing them
as a repo link, though I can't test that right now.
Mostly comments and docs: some places are actually paths, which
I have avoided changing. We may migrate them slowly, particularly
when they're user-visible.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Turns out that the pyln-proto dependency in the bolt packages is only
needed for testing, not for production. Making it a dev-dependency
means it isn't considered in resolution anymore.
Since the bolt, testing and client packages are to be used outside
from the project we can't use relative dependencies either, so make
then dependent on the version on PyPI. This also means we had to push
a couple of updated to PyPI.
Changelog-None
I thought about fixing them up, but really these should be in
lnprototest anyway. Turns out they're from the spec, so we should
actually fix them up there.
I moved the vector files into contrib/pyln-proto, since that still
needs them.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Mainly fixing type annotations, but some real fixes:
1. GossmapHalfchannel.from_str() should be a classmethod.
2. update_channel had weird, unusable default values (fields can't be NULL,
since we use it below).
[ There was one more occurence where isinstance should be used above
type() == xyz comparison. -- MS ]
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This would lead to errors about missing dependencies when attempting
to install using `pyhon setup.py install`. This is because the
`setup.py` file effectively is the manifest file used to discover
which dependencies are needed, so when using it to detect dependencies
we obviously don't have them yet.
See https://packaging.python.org/guides/single-sourcing-package-version/
We weren't writing out the length of a nested subtype's
dynamicarraylenght, now we do. The trick is to iterate through the
fields on a subtype (since the length field is added separately)
and to also iterate down through the otherfield values as we 'descend'
We were getting a number of incompatibility warning due to the
dependencies being expressed too rigidly. This losens the requirement
definitions to being compatible with a known good version, and while
we're at it we also bump all outdated requirements.
- Adds bitarray filling so mesages of any length can be encoded, instead of forcing the encoding to be of messages with length multiple of 5.
- Adds checks for encoding / decoding and raises expections if the inputs are not as expected.
- Flags functions that are supposed to be internal as "private".
Currently ``LightningConnection.remote_pubkey`` is set to ``None`` if the node is not the handshake initiator. This sets it to ``rs`` in act three from the receiver side
Since pyln-bolt* specify the 0.8.4 version which we didn't upload, and the
requirements.txt specify ==0.8.4, we need to backfill that version, even if we
could just bump it directly to 0.9.1.
