There were about 20 skipped tests for RATE and PRICE marked
"This test should be fixed". This change does that by fixing
the code for those functions, validating the existing tests,
and adding new ones. XIRR and XNPV are also substantially changed.
As part of this change, the following functions also have minor changes:
- isValidFrequency
- COUPDAYBS
- COUPNUM (additional tests)
- DB
- DDB
PhpUnit reports 100% coverage for all the changed functions.
Since I was dealing with skipped tests, I also fixed
tests/PhpSpreadsheetTests/Writer/Xlsx/LocaleFloatsTest,
which was being skipped in Windows. I also delete the temporary
file which it creates.
There is now only one remaining test which is skipped -
ODS Reader is not complete enough to run some tests against it.
Unfortunately, that test is too complicated for me to deal with now.
In researching this change, I found several places in the code where special code was added for Gnumeric claiming:
- Gnumeric does not handle free-format string dates
- Gnumeric adds extra options, not available in Excel,
for the frequency parameter for functions such as YIELD
- Gnumeric rounds the results for DB and DDB to 2 decimal places
None of these claims is true, at least not on a recent version
of Gnumeric, and the code which supports these differences is removed.
There did not appear to be any tests targeted for
these supposed properties of Gnumeric.
The PRICE function needed relatively minor changes - mostly
additional tests for invalid input. The main problem with the PRICE
tests is that Excel appears to have a bug. The algorithm is published:
https://support.office.com/en-us/article/price-function-3ea9deac-8dfa-436f-a7c8-17ea02c21b0a
The results that Excel returns for basis codes 2 and 3 appear to be
incorrect in many cases. I have segregated these tests into a
new test PRICE3. The results of these tests agree with the published
algorithm, and to the results for LibreOffice and Gnumeric.
The results returned by Excel do not agree with them.
The tests which remain in the test PRICE all use basis codes other
than 2 or 3, and all agree with Excel, LibreOffice, and Gnumeric.
For the RATE function, there appears to be a problem with how the
secant method was implemented. I studied the implementation of RATE
in Python numpy, and adapted its implementation of secant method.
The results now agree with numpy, and, more important, with Excel.
XIRR, which calls XNPV, permits its dates to be earlier than the
start date, whereas XNPV does not. I dealt with this by renaming
the existing XNPV function to xnpvOrdered, adding a parameter to
indicate whether start date has to be earliest. XNPV calls the new
function with that parameter set to TRUE, and XIRR calls it with
the parameter set to FALSE. Some additional error checking was
added to xnpvOrdered, and also to XIRR. XIRR tests benefited
from increasing the value of FINANCIAL_MAX_ITERATIONS.
Finally, since this change is very test-related:
samples/Basic/13_CalculationCyclicFormulae
PhpUnit started reporting an error like "too much regression".
The test deals with an infinite cyclic formula, and allowed
the calculation engine to run for 100 cycles. The actual number of cycles
seems irrelevant for the purpose of this test. I changed it to 15,
and PhpUnit no longer complains.
I believe that both CSV Reader and Writer are 100% covered now.
There were some errors uncovered during development.
The reader specifically permits encodings other than UTF-8 to be used.
However, fgetcsv will not properly handle other encodings.
I tried replacing it with fgets/iconv/strgetcsv, but that could not
handle line breaks within a cell, even for UTF-8.
This is, I'm sure, a very rare use case.
I eventually handled it by using php://memory to hold the translated
file contents for non-UTF8. There were no tests for this situation,
and now there are (probably too many).
"Contiguous" read was not handle correctly. There is a file
in samples which uses it. It was designed to read a large sheet,
and split it into three. The first sheet was corrrect, but the
second and third were almost entirely empty. This has been corrected,
and the sample code was adapted into a formal test with assertions
to confirm that it works as designed.
I made a minor documentation change. Unlike HTML, where you never
need a BOM because you can declare the encoding in the file,
a CSV with non-ASCII characters must explicitly include a BOM
for Excel to handle it correctly. This was explained in the Reading CSV
section, but was glossed over in the Writing CSV section, which I
have updated.
Returns #N/A, unless the element searched for is at the end of the array.
The problem is in Calculation.php line 4231:
if (!is_array($functionCall)) {
foreach ($args as &$arg) {
$arg = Functions::flattenSingleValue($arg);
}
unset($arg);
}
I believe this code is intended to handle functions where PhpSpreadsheet just passes
the call on to PHP without implementing the code on its own, e.g. for atan or acos.
In the bug report, the following code fails:
$flat_rate = "=MATCH(6,{4,5,6,2}, 0)";
$sheet->getCell('A1')->setValue($flat_rate);
The expected value is 3, but the actual result is "#N/A".
The reason for this result is that the parser replaces the braces with calls
to the MKMATRIX internal function, whose value for functioncall was:
'self::MKMATRIX'. Since this isn't an array, the flattening code is executed,
and the unintended result occurs. The fix is to change the definition for
functioncall in that case to [__CLASS__, 'mkMatrix'], avoiding the flattening.
However, there is also another part to this bug. The flattening should be
returning the first entry in the array, but is in fact returning the last.
This explains why the bug report specified "unless ... end of the array".
I confirmed that Excel does use the first item in the array rather than the last,
e.g. =atan({1,2,3}) entered into a cell will return atan(1), not atan(3).
The problem here is that flattenSingleValue, which says in its comments that
it is supposed to be returning the first item, uses array_pop rather than array_shift.
I have changed that as well. The same mistake was also present in
Cell.php function getCalculatedValue. The correct behavior can be verified
by entering =minverse({-2.5,1.5;2,-1}) into an Excel cell'
Excel flattens the result ({2,3;4,5}) to 2, and so should PhpSpreadsheet.
Fixes#1271Closes#1332
`$highestRow = $this->getHighestDataRow();` was calculated after `$this->getCellCollection()->removeRow($pRow + $r);` - this is the root reason for incorrect rows removal because removing last row will change '$this->getHighestDataRow()' value, but removing row from the middle will not change it. So, removing last row causes incorrect `$highestRow` value that is used for wiping out empty rows from the bottom of the table:
```php
for ($r = 0; $r < $pNumRows; ++$r) {
$this->getCellCollection()->removeRow($highestRow);
--$highestRow;
}
```
To prevent this incorrect behavior I've moved highest row calculation before row removal.
But this still doesn't solve another problem when trying remove non existing rows: in this case the code above will remove `$pNumRows` rows from below of the table, e.g. if `$highestRow=4` and `$pNumRows=6`, than rows 4, 3, 2, 1, 0, -1 will be deleted. Obviously, this is not good, that is why I've added `$removedRowsCounter` to fix this issue.
And finally, moved Exception to early if statement to get away from unnecessary 'if-else'.
Fixes#1364Closes#1365
Indentation in the xml leaves spaces in style string even after
replacing newlines. Replacing the spaces ensures no spaces in keys
of the resulting style-array
Fixes#1347
The `setRange` method of the `Xlsx/AutoFilter` class expects a filter
range format like "A1:E10". The returned value from
`$this->worksheetXml->autoFilter['ref']` could contain "$" and returning
a value like "$A$1:$E$10".
Fixes#687Fixes#1325Closes#1326
When freeze pane is in use on a worksheet, PhpSpreadsheet saves to Xlsx in such
a way that the active cell is always set to the top left cell below the freeze
pane. I find it difficult to understand why:
1. You have given users the setSelectedCells function, but then choose to
ignore it.
2. Excel itself does not act in this manner.
3. PHPExcel did not act in this manner.
4. PhpSpreadsheet when writing to Xls does not act in this manner.
This is especially emphasized because the one test in FreezePaneTest which
would expose the difference is the only test in that member which is
not made for both Xls and Xlsx.
5. It is *really* useful to be able to open a spreadsheet anywhere, even when
it has header rows.
Closes#1323
* Changes to WEEKNUM and YEARFRAC
The optional second parameter for WEEKNUM can take any of 10 values
(1, 2, 11-17, and 21), but currently only 1 and 2 are supported.
This change adds support for the other 8 possibilities.
YEARFRAC in Excel does not require that end date be before start date,
but PhpSpreadsheet was returning an error in that situation.
YEARFRAC third parameter (method) of 1 was not correctly implemented.
I was able to find a description of the algorithm, and documented
that location in the code, and implemented according to that spec.
PHPExcel had a (failing) test to assert the result of
YEARFRAC("1960-12-19", "2008-06-28", 1). This test had been dropped
from PhpSpreadsheet, and is now restored; several new tests have
been added, and verified against Excel.
* Add YEARFRAC Tests
Scrutinizer reported a very mysterious failure with no details.
project.metric_change("scrutinizer.test_coverage", < 0),
without even a link to explain what it is reporting.
It is possible that it was a complaint about code coverage.
If so, I have added some tests which will, I hope, eliminate the problem.
* Make Array Constant
Responding to review from Mark Baker.
* Merge with PR 1362 Bugfix 1161
Travis CI reported problem with Calculation.php (which is not part
of this change).
That was changed in master a few days ago
(delete some unused code).
Perhaps the lack of that change is the problem here.
Merging it manually.
Support for the CONTAINSBLANKS conditional style was added a while ago.
However, that support was on write only; any cells which used
CONTAINSBLANKS on a file being read would drop that style.
I am also adding support for NOTCONTAINSBLANKS, on read and write.
* Handle Error in Formula Processing Better for Xls
When there is an error writing a formula to an Xls file,
which seems to happen when a defined name is part of a formula,
the cell is currently left blank. A better result would be to
write the calculated value of the formula.
* Making Changes Suggested in Review
Per comment from Mark Baker:
1. Made return codes from writeFormula into constants.
2. Skipped redundant call to getCellValue when possible.
3. Added support for bool type, adding additional tests
for bool and string.
Per comment from PowerKiki:
1. Used standardized convention for assigning file name in test.
Since before this change, save would throw Exception, I kept
the unlink for the file in tearDown.
* Initial unit test for locale floats
This will require potential modification of the TravisCI environment to support other locales
* var_dump to check output on TravisCI
* Fix assertions for double/float and with/without line reference
* Style in unit test
* Handle ConditionalStyle NumberFormat When Reading Xlsx File
ReadStyle in Reader/Xlsx/Styles.php expects numberFormat to be a string.
However, when reading conditional style in Xlsx file, NumberFormat
is actually a SimpleXMLElement, so is not handled correctly.
While testing this change, it turned out that reader always expects
that there is a "SharedString" portion of the XML, which is not
true for spreadsheets with no string data, which causes a
run-time message.
Likewise, when conditional number format is not one of the built-in
formats, a run-time message is issued because 'isset' is used
to determine existence rather than 'array_key_exists'.
The new workbook added to the testing data demonstrates both those
problems (prior to the code changes).
* Move Comment to Resolve Conflict
Github reports conflict involving placement of one comment statement.
* Respond to Scrutinizer Style Suggestion
Change detection for empty SimpleXMLElement.
Prior to 1.10, all numeric values where read as floats. In 1.10
numeric values are read using 0 + x, which relies on PHP type
juggling rules. As a result, float(0.0) is written as string('0'),
then read back as int(0). This fix causes the writer to retain the
the decimal for float values such that a reader can differentiate
floats from ints.
Closes#1262
CALCULATION_REGEXP_CELLREF is not sufficiently robust.
It treats some perfectly legal defined names, e.g. A1A, as cell refs.
When the Xlsx Writer tries to save a worksheet which uses such a name
in a formula in a cell, it throws an exception.
The new DefinedNameConfusedForCellTest is a simple demonstration.
The Regexp has been changed to ensure the name starts on a Word boundary,
and to make sure it is not followed by a word character or period.
This fixes the problem, and does not appear to cause any regression
problems in the test suite.
Closes#1263
Fix: Return #NUM! if values and dates contain a different number of values
Fix: Return #NUM! if there is not at least one positive cash flow and one negative cash flow
Fix: Return #NUM! if any number in dates precedes the starting date
Fix: Return #NUM! if a result that works cannot be found after max iteration tries
Fix: Correct DocBlocks for XIRR & XNPV
Add: Validate XIRR with unit tests
Closes#1177
In cells with formulas containing conditions like `=IFSUM(A1:A3;"";B1:B3)`
to sum cells from range A1:A3 with empty value in range B1:B3, the function
`Functions::ifCondition()` create in this case the code `=""""` instead of
`=""`, so it didn't work.
Closes#1206
oleibman