Most of the tests for org/apache/pdfbox/util/DateConverter.java in src/test/java/org/apache/pdfbox/util/TestDateUtil.java have been commented out. DateConverter was broken.
The attached patch fixes the problems. Extensive comments document the problems. Here's a copy:
/* the former version of DateConverter had these bugs:
- - In toISO8601 the conversion from millis to minutes was with 1000/1000;
- should have been 1000/60.
- - PDFBox-402 was not completely implemented. The calendar fields in the
- POTENTIAL_FORMATS are shared among threads. Hence we must create new
- SimpleThreadFormats for each test. (Or synchronize somehow).
- - Some formats with hh did not have an a field. I changed them to HH.
- these questionable features:
- - A timezone with neither plus sign nor minus is assumed to be minus.
- This seems wrong, but I have not changed it.
- - toCalendar() returned a value in the default Locale.
- PDF files do not have locales (I think) and even if they do
- there is no reason to assume the Java default.
- I have switched to Locale.ENGLISH which was already assumed
- in the date formats and toString.
- and these infelicities:
- - Constants 60 and 1000 appeared.
- - zeroAppend was not used where applicable.
- In one case it was inapplicable only because TimeZone.getOffset
- was suspected of returning a long. It does not.
- - Manually computed constants were used to in date.substring
- thus reducing flexibility and maintainability.
- - The TimeZone name reported by toCalendar was always "Unknown"
- It is easy enough to compute a name.
- - Time zones were not accepted with most of the alternate parsing formats.
- The new code allows a timezone after any format.