[KEXI] [Bug 420534] New: Report uses incorrect record/row in group footer
jordi fita i mas
bugzilla_noreply at kde.org
Sat Apr 25 01:47:47 BST 2020
https://bugs.kde.org/show_bug.cgi?id=420534
Bug ID: 420534
Summary: Report uses incorrect record/row in group footer
Product: KEXI
Version: 3.2.0
Platform: Compiled Sources
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: Reports and Printouts
Assignee: kexi-bugs at kde.org
Reporter: jfita at infoblitz.com
CC: adam at piggz.co.uk, inksi at fables.co.za
Target Milestone: ---
Created attachment 127843
--> https://bugs.kde.org/attachment.cgi?id=127843&action=edit
Sample KEXI database with a table and a report based of on that table, with a
group using the first column that shows different values in header and footer
of same group.
SUMMARY
I believe this is actually a bug in KDb, but it is most visible when doing a
report that has data sourced fields in a group footer: when there is a single
group the QVariant returned by KexiDBReportDataSource is invalid; when there is
more than one group it uses the first row of the next group.
STEPS TO REPRODUCE
1. Create a new file-based KEXI project.
2. Add a new table with two fields, it does not matter their data type.
3. Add some rows to that table and repeat some values in one of the two
columns.
4. Add a new report with a section grouped by the column that has repeated
values; mark at least the “Show group footer” option.
5. Add a field in the group header and set its data source to the field for
grouping.
6. Add the same field but on the group footer.
6. Switch to edit mode.
OBSERVED RESULT
The fields in the header and footer of *the same group* do not have the same
values, in fact the footer has the same value as the header for the *next*
group; the last group’s footer has no value.
EXPECTED RESULT
The header and footer fields should have the same value because they are in the
same group and use the grouping field as data source.
(I’ve attached a sample database that shows this behavior).
SOFTWARE/OS VERSIONS
Linux/KDE Plasma: openSUSE Leap 15.2
KDE Plasma Version: N/A
KDE Frameworks Version: 5.69.0
Qt Version: 5.12.2
ADDITIONAL INFORMATION
Is i said, i believe that i narrowed the problem down to KDb, but i do not know
how to fix it.
This is what i found.
In KReportPreRenderePrivate::renderDetailSection, KReport does something along
the lines of (pseudocode):
dataSource->moveFirst();
if (group->hasHeader()) {
renderSection(group->header());
}
while (!dataSource->eof()) {
renderSection(group->detail());
dataSource->moveNext();
}
if (group->hasFooter()) {
if (dataSource->movePrevious()) {
renderSection(group->footer());
}
}
The problem is that dataSource->movePrevious() returns true but does NOT move
back one record.
The only KReportDataSource-derivated class i could find in KEXI is
KexiDBReportDataSource and its movePrevious is implemented as such:
bool KexiDBReportDataSource::movePrevious()
{
if ( d->cursor ) return d->cursor->movePrev();
return false;
}
I have verified that d->cursor, a pointer to KDbCursor, is not null and that
its movePrev() return true.
KDbCursor::movePrev only works if the cursor is buffered, and
KexiDBReportDataSource makes sure of it. The rellevant part of movePrev is:
if (m_afterLast && (m_records_in_buf > 0)) {
drv_bufferMovePointerTo(m_records_in_buf - 1);
m_at = m_records_in_buf;
d->atBuffer = true; //now current record is stored in the buffer
d->validRecord = true;
m_afterLast = false;
return true;
}
It seems that buffering should be implemented by drivers and
drv_bufferMovePointerTo tells the driver’s cursor to move its pointer to the
last record in its cache.
However, i’ve seen that the driver for PostgreSQL does not implement buffering
and SqlCursor does the following:
void SqliteCursor::drv_bufferMovePointerTo(qint64 at)
{
d->curr_coldata = d->records.at(at);
}
It seems that d->records is the “raw” record data from the database —the cache
buffer—, and buffered cursor call drv_appendCurrentRecordToBuffer() to record
the current record to that buffer. However, SqlCursor’s method, even though
i’ve checked it is called, never stores anything inside records because the
first thing it does is:
void SqliteCursor::drv_appendCurrentRecordToBuffer()
{
if (!d->curr_coldata)
return;
…
The *only* time i could find that curr_coldata is set is precisely within
drv_bufferMovePointerTo, that can only set it to null because records has never
been filled.
Moveover, even if somehow curr_coldata were set, i do not see where it is being
used and SqliteCursor::value always calls SQLite3’s function to retrieve the
data from the prepared query. Once sqlite3_step returns SQLITE_DONE after the
last row, the prepared statement can not be used, SqliteCursor::value returns
QVariant() and that is why the last group header does not display anything.
I do not know what do do now.
It seems to me that SQLite’s driver buffering does nothing, PostgreSQL’s driver
explicitly says it so in its comments, and the only driver that it *might* work
—i can not test it right now— is with MySQL’s as it uses mysql_data_seek() and
mysql_fetch_row() inside drv_bufferMovePointerTo to retrieve the data from
MySQL’s own buffer.
--
You are receiving this mail because:
You are the assignee for the bug.
More information about the Kexi-bugs
mailing list