We were working under the assumption that XLSX's inline strings only had a single value node (`<t>`). This is incorrect.
To get the actual value of an inline string node, we need to concatenate the value of all its child nodes.
This commit introduces Row and Cell entities, that will replace the arrays passed in previously.
It also adds support for Cell styling (instead of Row styling only).
This commit adds support for dates using the 1904 calendar (starting 1904-01-01 00:00:00).
It also fixes some issues with the dates in 1900 calendar (which now correctly start at 1899-12-30 00:00:00).
Finally, it is now possible to have negative timestamps, representing dates before the base date (and up to 0000-01-01 00:00:00), as per the SpreadsheetML specs. Note that some versions of Excel don't support negative dates...
* Refactor readers to get a proper DI
Similar to what was done with writers, readers also needed to be updated to match the new way of doing things.
This commits promotes a better DI (factories, injection through constructors).
* Escapers should not be singletons
Instead, they should be proper object that can be injected where needed.
When converting an XMLReader node to a SimpleXMLElement, the conversion would automatically decode the XML entities. This resulted in a double decode.
For example: "&#34;" was converted to """ when imported into a SimpleXMLElement and was again converted into " (quote).
This commit changes the way the XLSX Shared Strings file is processed. It also changes the unescaping logic for both XLSX and ODS.
Finally, it removes any usage of the SimpleXML library (yay!).
In some cases, reading an XLSX file produce E_WARNING from the max()
call in the fillMissingArray() method. This commit fix the problem
by handling empty rows.
- To determine if a style should apply a date format, the presence of "applyNumberFormat" attribute on the "cellXfs" section of styles.xml is now optional. We only look at the "numFmtId" attribute (but early return if "applyNumberFormat" is set to "0").
- The format code can contain lowercase AND now uppercase characters as its pattern.
- "General" format code used as a custom format is now supported. It seems to be used by a bunch of programs...
When a cell contains multiple text nodes, the cell value is currently obtained by concatenating the value of each text node.
Instead, values should still be concatenated but a space should be added in between.
When reading spreadsheets, Spout should be able to return formatted dates, as shown when opened with Excel for instance.
It currently only returns DateTime/DateInterval objects, making it impossible to read + write, as the Writer does not accept objects.
The value passed into the format() function is coming from an XML file and has never been coerced.
Therefore, when checking is_int($value), the check always returns false - because it's a string.
Changing the check fixes the issue and Spout now correctly parses large numbers.
Although Excel has a Date type, older Excel versions use numeric values to store dates.
The value represents the number of days since Jan 1st, 1900.
The only way to tell if the value is a number or a date is to look at the styles.xml and check if the cell has date formatting.
Instead of the hasNext() / next() syntax, readers now implements the PHP iterator pattern.
It allows readers to be used with a foreach() loop.
All readers now share the same structure (CSV is treated as having exactly one sheet):
- one concrete Reader
- one SheetIterator, exposed by the Reader
- one or more Sheets, returned at every iteration
- one RowIterator, exposed by the Sheet
Introducing the concept of sheets for CSV may be kind of confusing but it makes Spout way more consistent.
Also, this confusion may be resolved by creating a wrapper around the readers if needed.
-- This commit does not delete the old files, not change the folder structure for Writers. This will be done in another commit.
