This patch adds an option to configure the name and/or directory of the
debug log.
The user can specify either a relative path, in which case the path
is relative to the data directory. They can also specify an absolute
path to put the log anywhere else in the file system.
Add leveldb: prefix to leveldb debug logging lines.
leveldb debug messages come in various scary flavors such as:
2017-11-30 08:26:31 leveldb: Recovering log #26
2017-11-30 08:26:31 leveldb: Level-0 table #28: started
2017-11-30 08:26:31 leveldb: Level-0 table #28: 597 bytes OK
2017-11-30 08:26:31 leveldb: Delete type=0 #26
2017-11-30 08:26:31 leveldb: Delete type=3 #24
so it's reasonably important to mark them as coming from leveldb
internals and not from consensus validation wallet or such.
(this only affects `-debug=leveldb` or `-debug=1` otherwise you won't
see them in the first place)
PR #10286 introduced a few steps which are not robust to early shutdown
in initialization.
Stumbled upon this with #11781, not sure if there are other scenarios
that can trigger it, but it's harden against this in any case.
blockchain.cpp has low unit test coverage. This commit is intended
to start improving its code coverage to reasonable levels. One or more
follow up commits will complete the task that this commit is starting
(though the usefulness of this commit is not dependent upon later
commits).
Note that these tests were not written based upon a specification of how
GetDifficulty *should* work, but rather how it actually *does* work. As
a result, if there are any bugs in the current GetDifficulty
implementation, these unit tests serve to lock them in rather than
expose them.
-- Why has blockchain.cpp been modified if this is a unit testing change?
Since the existing GetDifficulty function relies on a global variable,
chainActive, it was not suitable for unit testing purposes. Both the
existing GetDifficulty function and the unit tests now call through to
a new, more modular version of GetDifficulty that can work on any chain,
not just chainActive.
-- Why does blockchain_tests.cpp directly include blockchain.cpp instead
of blockchain.h?
While the new GetDifficulty function's signature is arguably better than
the old one's, it still isn't great, and doesn't seem to warrant inclusion
as part of the blockchain.h API, especially since only test code is
directly using it. If a better way of exposing the new GetDifficulty
function to unit tests exists, please mention it and the commit will be
updated accordingly.
-- Why is the test fixture named blockchain_difficulty_tests rather than
blockchain_tests?
The Bitcoin Core policy for naming unit test files is to match the the
file under test ("blockchain" becomes "blockchain_tests"). While this
commit complies with that, blockchain.cpp is a massive file, such that
having all of the unit tests in one file will tend towards disorder.
Since there will be a lot more tests added to this file, the intention
is to divide up different types of tests into different test fixtures
within the same file.
ConnectBlock() relies on validation that happens in ContextualCheckBlock() and
ContextualCheckBlockHeader(). This has implications for implementing consensus
changes and handling software upgrade to ensure that nodes upgrading their
software end up enforcing all the consensus rules.
A) The changes in behavior are as follows:
1. Introduce logging category "none" as alias of "0" for
both RPC-logging and bitcoind "-debug" parameter.
2. Same as "0" is given to argument of "-debug",
if "none" or "0" is given to <include>, all other given logging
categories are ignored. The same is true for <exclude>.
(Before this PR, "0" was accepted but just be ignored itself.)
B) The changes in the help text are as follows:
1. Add a descrption about the evaluation order of <include> and
<exclude> to clarify how debug loggig categories to be set.
2. Delete text that describe restriction about libevent because
it's already allowed libevent logging to be updated during runtime.
3. Add a description for category "all", "1", "none" and "0".
4. Add "optional" to the help text of <include> and <exclude>.
5. Add missing new lines before "Argument:".
6. This RPC always returns all logging categories with status.
Fix the help text to match this behavior.
Opt-in RBF checkbox uses less technical jargon and emphasises
the fee bump functionality (at the expense of not mentioning
other uses of RBF).
The transaction confirmation screen uses copy consistent with this.
- Added `help-console` to the list of autocompletion strings
- Implemented requested changes to help message:
- Added an example that uses access-by-index `getblock(getblockhash(0) true)[tx][0]`
- Replace "bracketed syntax" to "parenthesized syntax" where applicable
- Replace "separate" with "delimit"
- Removed `<br>` and `<b>help/help-console</b>` from translation strings, since these parts don't change between languages
- Changed examples to be based off `getblock 0` so they will work even with pruned/no blockchain and `disablewallet` if copied and pasted
- Clarified syntax for queries of named/unnamed result objects.
Currently it's possible to accidentally type e.g.
bitcoin-cli -getinfo getbalance
and get an answer which can be confusing; the trialing arguments are
just ignored.
To avoid this, throw an error if the user provides arguments to
`-getinfo`.
This makes all include paths in the GUI absolute.
Many changes are involved as every single source file in
src/qt/ assumes to be able to use relative includes.
Change feebumper from a stateful class into a namespace of stateless
functions.
Having the results of feebumper calls persist in an object makes process
separation between Qt and wallet awkward, because it means the feebumper object
either has to be serialized back and forth between Qt and wallet processes
between fee bump calls, or that the feebumper object needs to stay alive in the
wallet process with an object reference passed back to Qt. It's simpler just to
have fee bumper calls return their results immediately instead of storing them
in an object with an extended lifetime.
In addition to making feebumper stateless, also:
- Move LOCK calls from Qt code to feebumper
- Move TransactionCanBeBumped implementation from Qt code to feebumper
Future commit will remove the FeeBumper class. This commit simply places
everything into a feebumper namespace, and changes the enum class name
from BumpeFeeResult to feebumper::Result.
Future PRs will completely refactor this translation unit and touch all
this code so we rename the variables to follow project stlye guidelines
in this preparation commit.
Don't use m_ prefixes for member variables since we're going to remove
the class entirely in the next commits.
Static analyzer (and humans!) will see ...
```
else if (state.m_chain_sync.m_timeout == 0 || (state.m_chain_sync.m_work_header != nullptr && ...
```
... and infer that state.m_chain_sync.m_work_header might be set to nullptr,
and thus flag `state.m_chain_sync.m_work_header->GetBlockHash().ToString()`
as a potential null pointer dereference.
This commit makes the tacit assumption (m_work_header != nullptr) explicit.
Code introduced in 5a6d00 ("Permit disconnection of outbound peers on
bad/slow chains") which was merged into master four days ago.
We should generally avoid writing to debug.log unconditionally for
inbound peers which misbehave (the peer being about to be banned
being an exception, since they cannot do this twice).
To avoid removing logs for outbound peers, a new log is added to
notify users when a new outbound peer is connected which mimics
the version print.
Tim Shimmin