Test Details Overview

Fix config parsing bugs and split the config tests (#5180)

* Split the config tests by the layer they exercise

The toml suite mixed tests of the table wrapper with tests of the config
classes that sit on top of it. The wrapper tests move to a tomlconfig
suite in their own file, leaving toml.cpp to the toml_config tests of
daemon_config, rpc_config and log_config.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Apply the configured RPC IPC address

The RPC config deserializer parsed process.ipc_address correctly but then
stored the RPC bind address in its place, so the key had no effect.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* Generate the RPC sample config for the active network

--generate_config rpc always used dev network constants, so the printed
default port ignored --network. The node branch already used the active
network.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* Return an empty table for a missing or non-table child

get_required_child returned the parent table when the key was missing, so
the caller went on reading the parent's keys as if they were the child's.
Both child lookups also dereferenced a null table when the key held a
plain value. Report the error and hand back an empty table instead.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* Reject negative numbers for unsigned settings

lexical_cast wraps a negative number into an unsigned target, so
`bandwidth_limit = -1` was accepted as 18446744073709551615 instead of
being reported. Values destined for an unsigned type are now checked for
a minus sign first. The stringified value is also checked for presence
before being dereferenced, which a key holding a table or an array would
previously have done on a null optional.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Derive config type descriptions from the type

Type descriptions were a list of explicit specializations, so any type
not on the list fell through to "an unknown type". size_t, which several
node settings use, is such a type on platforms where it is neither
uint64_t nor unsigned int, and addresses had no description at all. A
single if constexpr chain derives the description from the type instead,
covering every integral width and naming IPv6 addresses.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Remove leftovers from the config classes

Drop a duplicated enable_sign_hash getter, an unused qtwallet config path
and a redundant table copy in daemon_config, jsonconfig includes that no
config translation unit uses any more, and a typo in an lmdb type
annotation.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* Keep the first configuration error

The child lookups assigned the error object directly, so a structural
problem found late in a document replaced the parse failure that came
before it and the loader reported the wrong cause. They now go through
set_error_once, which follows the policy configbase already documents
for conditionally_set_error: the first error is the one that explains
the rest.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Cover non-scalar values at scalar keys

A table or an array where a setting expects a scalar is rejected with
invalid_value and the target keeps the value it had. Nothing exercised
that shape before, the tests only ever fed wrong scalars. String targets
are the case that depends on the presence check in get_config: an empty
string converts successfully, so without the check a non-scalar would be
read as an empty value instead of being reported.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

__________

Type: commit

Hash: 0e741ee22ed84f4ed6574252a9b2f442bc340aaf

Created: today

__________

gr0v1ty/nano-node:0e741ee22ed84f4ed6574252a9b2f442bc340aaf