Correction to comment in prev commit (and removing it here):
spec.loader.exec_module does not spawn new threads, it simply
executes the module in the current thread.
I got confused but turns out "load_plugin" itself is sometimes
not called from the main thread. Specifically (e.g.), since the recent
wizard rewrite, in the qt gui, the wizard loads the hww plugins
from a new thread.
This now better explains the macos hww crashes: they had started
appearing because we upgraded hidapi (which made it more sensitive
to having to import from main thread) AND scanning(->importing) from
the wizard no longer happened on the main thread after the rewrite.
Plugins should be thread-safe in terms of where they are imported from.
Let's log the importer thread's name (added here), to help recognise
related threading issues.
- risk_of_burning_coins_as_fees is turned into a private (helper) method, only called by check_sighash.
UIs should only care about check_sighash.
- check_sighash returns instance of new class "TxSighashDanger" instead of tuple
- made warning levels more fine-grained (FEE_WARNING_SKIPCONFIRM vs FEE_WARNING_NEEDCONFIRM)
- this became more complicated than I had hoped for but I think it is worth it to ~merge
check_sighash and risk_of_burning_coins_as_fees into one.
related https://github.com/spesmilo/electrum/pull/8699
I think it's nicer for the network status to be rightmost, which was the case for a long time.
(Though plugins are still added after (via multiple different mechanisms, see 2fa, hww, revealer).
We could perhaps add a VLine separator, or add them to the left instead.)
follow-up 90d1d587e8
follow-up https://github.com/spesmilo/electrum/pull/8651
We cannot do direct comparison of path-like strings without normalisation.
E.g.
```
relative_path() cp1. path='c:\\users\\user\\appdata\\roaming\\electrum\\wallets\\9dk', wallets=C:\Users\User\AppData\Roaming\Electrum\wallets
relative_path() cp2. commonpath='c:\\users\\user\\appdata\\roaming\\electrum\\wallets'
relative_path() cp3. new_path='9dk'
```
In the db, the 'seed_type' field could be present both at the top-level and inside keystores.
Note:
- both fields had usages
- the top-level field was added in 2.8 re "initial segwit support" (3a64ec0f2e)
- there was no db upgrade for it, so older files are missing it
- if present, valid values can be electrum types but also
other types supported by the wizard, e.g. "bip39"
- the keystore-level field was added in 4.1 (7b7bba2299)
- there was a db upgrade when it was introduced, so old files also have it
- if present, valid values can only be electrum types
- there is not much value in the top-level one having a non-electrum value,
and those values were never used by other parts of the code
- note that when creating a standard wallet from a bip39 seed, the seed is discarded.
Only the derived xprv and the derivation path are kept. If we changed this and also kept the seed,
e.g. to display it to the user, then it would make sense to save the seed type (e.g. "bip39").
However storing that seed_type would make more sense at the keystore level (not top-level).
We delete the top-level 'seed_type' field.
```
{
"keystore": {
"seed_type": "segwit",
...
},
"seed_type": "segwit",
...
}
```
```
32.40 | E | gui.qt.exception_window.Exception_Hook | exception caught by crash reporter
Traceback (most recent call last):
File "/home/user/wspace/electrum/electrum/gui/qt/wizard/wallet.py", line 709, in <lambda>
button.clicked.connect(lambda: Bip39RecoveryDialog(self, get_account_xpub, on_account_select))
File "/home/user/wspace/electrum/electrum/gui/qt/bip39_recovery_dialog.py", line 40, in __init__
fut = asyncio.run_coroutine_threadsafe(coro, network.asyncio_loop)
AttributeError: 'NoneType' object has no attribute 'asyncio_loop'
```
Ah ok, I give up for now... the prev does not really work.
The prior commit works on Windows but not on macOS.
On Windows, it would package all plugins as code and only as code.
On MacOS however, it would not package any plugins at all. And with this commit,
where I mark the plugins folder to be packaged as *data*, it packages all plugins as *both* code and data.
Not sure why.
Let's just package all plugins as both code and data, and ignore the code instances explicitly...
