First, save isn't necessary if the book is dirty, so don't... but that
means that the book has to be marked dirty after a session swap. No more
laziness.
Second, regardless of the outcome of inner_main_add_price_quotes the
session must be destroyed to remove the lock.
A couple of cleanups in QofSessionImpl::save as well: Rewrote the
descriptive comment to reflect how it really works when the backend has
gotten disconnected and removed the superfluous qof_book_set_backend
with the backend that we'd *just gotten from the book*.
Change all instances of bugzilla.gnome.org to bugs.gnucash.org, reflecting
our migration to a self-hosted bug tracker.
Inform the Translation Project Coordinator at release that this affects
translatable strings and that all message catalogs have been updated.
In addition to not begining to edit already-loaded transactions,
don't try to load splits that are already loaded. It shouldn't
be possible to load a transaction without also loading its splits.
The underlying problem was that the vendor object remained in infant state
That confused the backend code so it used an sql INSERT statement instead
of an UPDATE statement to write back the changes. As the object already
existed in the db this would fail.
The fix is to ensure the object doesn't remain in infant state during
sql loading. See the bug report for a more detailed explanation.
An extra XaccTransBeginEdit, never committed, for transactions that
the backend tried to load when they were already there. That made
the register think that something else had it open.
Instead of reporting an error and declining to load the file (XML)
or failing to enter a value (SQL) when a bad date is found in the
database, use a 0 time stamp (1970-01-01 00:00:00 UTC). Adds a warning
in SQL backends; there was one already in XML.
Provide a backup recovery function that instead of dropping primaries
and restoring backups merges the primaries and backups. This should
handle a worst-case safe-save failure where the backup tables don't
have a complete set of rows for some reason.
We don't use floats in GnuCash, we use doubles (and those as little as
possible), but dbd-sqlite3 is broken in that it stores only floats.
Simply casting floats to doubles introduces bogus additional digits
that can cause round-trip tests to fail. Instead convert floats to
doubles by multiplying by 10E6, rounding, then dividing by 10E6.
Removes a public function, GncDbiSqlConnection::table_manage_backup that
should have been private in the first place.
Better encapsulates table renames and drops with private functions and
handles cases where there exist some primary tables and some backup tables.
SQLite backend.
Release Note: This bug caused data loss if you saved your SQLite3
database to a different file or database.
The problem is that in SQLite3 (though not in MySQL or PgSQL) the
subquery ((SELECT DISTINCT guid FROM transactions)) (note the double
parentheses) returns only the first guid in the subquery's results.
Some transactions are loaded by special queries and those queries are
also used to retrieve the transaction's slots so they weren't affected.
This commit contains another round of cleanups in the
timespec to time64 conversion. There were a number of
false assumptions that time64 = 0 would be a bad date
in the xml parser. This commit corrects enough of them to
eliminate the bug. Further cleanup is probably advised but
can be done at a later stage.
First, remove the unnecessary locale push & pop on <CT_TIME64>load.
Second, the registry accesses were caused by using g_win32_get_locale
to convert the Microsoft locale strings to POSIX ones. We don't care
what kind of string we get as long as we can pass it back to setlocale,
so remove that.
Third, gnc_push/pop_locale were used only in backend/dbi in a
very limited way and did much more than was necessary, so
convert them to C++ inlines in gnc-backend-dbi.hpp that does
only what we need them to.
properly - corrupted business data
Turned out to be a pointer/value mismatch between <CT_NUMERIC>load()
and most of the setter functions, so the address was getting set
as the value.
This crash started to appear as of commit 80dbb9940b because the sequence
of split loading has changed as a result of the query optimizations.
Invoice transactions get loaded before the general transaction loading happens.
However because of this, when an invoice transaction was encountered again
during general transaction loading, it was (correctly) not created again
AND (incorrectly) not opened for subsequent editing. This caused
an assert to fail when the splits for this transaction are loaded
shortly afterwards. The solution is simply to ensure all transactions
are opened for editing during the general transaction loading call.
set/get_locale are apparently very expensive on Mingw64, and setting
the C locale for extracting a string is unnecessary.
Unfortunately the released version of libdbi still uses strtod so
setting the C locale *is* still necessary for retrieving floats
and doubles and for passing queries.
Thanks to Mkubat for the diagnosis.
Be smarter about what is path and key for each slot.
Instead of assuming a slash is always a path separator (first attempt
on unstable) or never a separator (second attempt),
track the parent path while loading kvp slots from the db
and deduce the slot's name by substracting this parent path.
Four date elements were affected: GncEntry::date, GncEntry::date_entered,
GncInvoice::opened, and GncInvoice::posted. The problem arose during the
cleansing of Timespec from the reports; the setter functions for those
elements were converted to time64 but no provision was made to the SQL
backend to pass them time64 instead of Timespec*.
This commit adds a new column type, CT_TIME64, and changes the column
types for those elements to CT_TIME64.
The import-map-bayes uses a three-part key that uses the same delimiter
as a path and the SQL backend was throwing away everything except the
account guid.
Added more slot types including import-map-bayes ones to the test xml
file to help debug this and to catch it in the future.
Also don't pass std::string.c_str() to a std::vector<std::string>
constructor, just pass the string.
transaction number or split action (requires at least GnuCash 2.5.0)
Strip leading delimiters from KVP keys when reading them from the
database. Leading delimiters are incorrectly included in databases
created with GnuCash 2.6.x.
Set a feature to prevent versions older that 2.6.20 from loading a
database from which they cannot read the dates. Ideally we would do this
only if the database is written to, but the current persistence design
includes committing back to the database during the load so the net
effect is that the flag would be set anyway.
The template avoids the need to cast to and from void*, and adds flexibility to
the targeted function's signature.
test-stuff.h defines a macro, "failure" which is used as an identifier
in the standard IO library, so I moved any inclusion of test-stuff.h to
the last include position so that "failure" wouldn't be defined before
the IO library was included.
Only string values should be quoted in queries; in particular NULL
isn't a string value and must not be quoted.
Note that this is a less than perfect solution because it doesn't use
the Database's quoting function and so doesn't escape quotes, linefeeds,
or carriage returns inside the string. That's because the SQL generating
logic is independent of the connection class and can't easily get to it.
Wherin the problem is that MySQL's TIMESTAMP has a date range of
1970-01-01 00:00:01 to 2038-01-19 03:14:07 and is unable to handle
time_t of 0. MySQL's TIMESTAMP also assumes that input is in the server's
timezone and adjusts it to UTC. GnuCash has already done that conversion.
For each subclass, getting rid of GNC_SQL_OBJECT_BACKEND_VERSION which
was a bit misguided.
Also remove the bogus test the skipped loading a table if its version
didn't match GNC_SQL_OBJECT_BACKEND_VERSION which was even more misguided.
This will avoid a ninja-build from picking up a config.h generated by the autotools build
(in the root build directory). Picking up the wrong config.h may lead to all kinds of
subtle issues if the autotools run was done with different options than the cmake run.
gtk-mac-bundler can't access the executable's rpath list so it can't
follow dependencies if they're not in $install_dir/lib from @rpath.
Autotools always sets absolute path install names so this should have no
adverse affects on other Mac builds.
Instead of using the more generic string constructor. The struct tm
constructor is several times faster and this change speeds up
transaction loading by 2x.
It is split into
- /libgnucash (for the non-gui bits)
- /gnucash (for the gui)
- /common (misc source files used by both)
- /bindings (currently only holds python bindings)
This is the first step in restructuring the code. It will need much
more fine tuning later on.
