There were two issues here, the first one was that the copied doclink
was pointing to the original doclink which lead to a double-free. The
second is the setting of the doclink on the copied invoice needs to be
done with gncInvoiceSetDocLink otherwise on closure the doclink value
will disappear.
- avoid hashing Account* twice. one call to operator[] is sufficient.
- don't need use std::move with better constructors
- sanity check - can't have a budget with num_periods = 0
- remove unused headers
- don't need to create/destroy GHashTable for each feature query
- plugs leak: g_hash_table_unref (features_used) not always called properly
- to check 1 feature, don't need to traverse whole GHashTable
Instead of using static_assert. This prevents the compiler from even
trying and avoids weird compilation errors when testing types for
instantiating other templates.
The compiler complains that there's no matching
gnc_register_number_range_option for GncOptionDB*, which without this
commit is true because the explicit templates are for GncOptionDBPtr&.
Note that the original template definition is for GncOptionDB* and
that the header-defined inlines that take GncOptionDBPtr& call the
GncOptionDB* version.
Because when loading the value the split in the xml backend doesn't yet
have a parent so the code tried to convert to GNC_COMMODITY_MAX_FRACTION
and if the numerator was larger than 10^10 that would overflow.
To fix it this changes the "don't know" response to get_currency_denom
and get_commodity_denom to GNC_DENOM_AUTO which will normally leave
the denominator alone.
The root cause of which is that on recent releases of GLib (recent
meaning 2.66 in the current Debian stable!) g_type_instance_get_private
looks in the wrong place for the private data. When running the script
in question it returned NULL and since the code didn't check for a valid
pointer, it crashed.
So this change replaces all calls to g_type_instance_get_private with
the function [type_prefix]_get_instance_private() added in glib-2.36
except for two register2 files that have been removed from master; those
are ignored to avoid unnecessary merge conflicts.
tests TXN_TYPE_NONE in utest-Transaction.c
testing TXN_TYPE_INVOICE, TXN_TYPE_PAYMENT, and TXN_TYPE_LINK will
require valid posted invoices, so, are best tested in utest-Invoice.c
When a top level bayesian entry is deleted, the book is not marked
dirty and so the save button is not highlighted. Deleting individual
entries did mark book dirty.
Fixed in gnc_account_delete_all_bayes_maps by wrapping the deletion
with xaccAccountBeginEdit/CommitEdit block.
