Found via `codespell -q 3 -S *.po,./po,*.min.js,./ChangeLog*,./NEWS,./borrowed,./doc/README*,./AUTHORS,./libgnucash/tax/us/txf-de*,./data/accounts -L ans,ba,cas,dragable,gae,iff,iif,mut,nd,numer,parm,parms,startd,stoll`
Refactor try_gz_open to return a std::pair<FILE*, thread>. That removes
the need for the threads hash-table and wait_for_gzip().
The cause of the gcc12 error was that we were using the thread's pipe's
FILE* as the key to the hash-table and had to close it before calling
wait_for_gzip(file) to remove the thread from the hash-table. gcc12
considers that a use-after-free. It happens to be wrong, but removing the
need for it results in a cleaner implementation as well as silencing the
warning.
significantly
Modify the DEBUG and PINFO macros to return unless qof_log_check is
true. Replace almost all direct calls to g_debug and g_message with
DEBUG and PINFO respectively.
Track the highest logging level sent to qof_log_set_level to provide a
short-circuit return in qof_log_check. Remove setting GNC_MOD_TESTS to
QOF_LOG_DEBUG so that the short-circuit threshold isn't defeated by
always being DEBUG.
Net result: 33% improvement in xml load times.
Moving all of the error-handling to it for better clarity and
simplicity. Drop the unnecessary stat call, open will fail with
EEXIST if there's already a lock file. Provide specific messages
for file system failures.
Prevent WebKit's fork from getting a copy of the pipe's file descriptors
and so keeping the pipe from closing.
Thanks to Simon Arlott for recognizing the FD duplication.
This was an effort of somewhat dubious value to detect if a process
on another ocmputer had a hsrd-link to a lockfile on an NFS share.
NFS is rarely used now and SMB doesn't support hard links so this check
adds complexity with no real value.
m_lockfd is not initialised. If the file is locked then it will not be set
before session_end and close() will be called on an uninitialised int.
Initialise it to -1 in the class definition.
Consistently use -1 instead of "< 0" or "< 1" as the definition of invalid.
Always set it to -1 after closing it.
The root accounts start with a non-zero editlevel because BeginEdit is
called for them during loading but not committed after loading.
If the book is then closed without performing any further edits that would
require a commit, the Account book_end process does nothing because the
root account is still being edited and so none of the accounts are freed.
Move all of the #include <glib> to before the extern "C" blocks
so that the include guards will protect against headers inside
the extern "C" block also including glib.h.
to mysql database.
It's more general, any save-as to a different backend type, i.e.
xml->sql or sql->xml, left the previous book locked.
Ensure that the session is ended and the book unlocked in the
backend's destructor. This also fixes a memory leak in the SQL
backend because the GncSqlConnection wasn't being deleted.
A "Feature" of MYSQL is that it allows C-style backslash escapes
in string constants and replaces them with the actual character
(e.g. \n is converted to 0x0a). This causes round trip problems
if the escape is one of the allowed ones and a MYSQL error if it
isn't.
Disable the feature so that MYSQL follows the SQL standard.
Price Database dropping user:price-editor entries.
A wider problem: QofInstance was unmarking itself dirty as long as the
backend raised an error and unconditionally marking itself non-infant.
This matters because the SQL backend depends on infant status to decide
whether to do an insert or update query; either will fail if the infant
status is wrong.
The price editor dialog clones a price having only its commodity set and
GncSqlPriceBackend declines to save if the currency isn't set. Since the
instance marked itself non-infant even though it wasn't saved subsequent
commits tried to use an update query and since the price wasn't in the
table that inevitably failed.
Requiring that QofBackend::commit should doing the marking-clean
required implementing it in the backends where it wasn't already.
When compiling with -Werror, GNUCash fails to build due to various error: missing sentinel in function call.
This patch resolves the issue by passing nullptr instead of C-style casting NULL.
when database already exists.
Crashed because of dereferencing GncSqlBackend::m_conn when it's nullptr
after the DB create failed. Protect all such dereferences.
For clarity. In so doing found the backend behavior a bit inconsistent
so it's modified to do what the enum values indicate.
In the course of changing the various calls I found some implementation
errors in the back end and corrected them.
It's more descriptive and less likely to be confused with the book:id value
in XML files that is the book's GUID.
Also changed the QofSessionImpl::begin new_uri parameter from std::string to
const char*. There's no point in allocating a string just to call
new_uri.c_str() all over the place.
