SolrJ doesn't need a DateUtil class, particularly since we're on Java 8 and can simply use new Date(Instant.parse(d).toEpochMilli()); for parsing and DateTimeFormatter.ISO_INSTANT.format(d.toInstant()) for formatting. Yes, they are threadsafe. I propose that we deprecate DateUtil from SolrJ, or perhaps outright remove it from SolrJ for Solr 6. The only SolrJ calls into this class are to essentially use it to format or parse in the ISO standard format.
I also think we should move it to the "extraction" (SolrCell) module and name it something like ExtractionDateUtil. See, this class has a parse method taking a list of formats, and there's a static list of them taken from HttpClient's DateUtil. DateUtil's original commit was
SOLR-284 to be used by SolrCell, and SolrCell wants this feature. So I think it should move there.
There are a few other uses:
- Morphlines uses it, but morphlines depends on the extraction module so it could just as well access it if we move it there.
- The ValueAugmenterFactory (a doc transformer). I really doubt whoever added it realized that DateUtil.parseDate would try a bunch of formats instead of only supporting the ISO canonical format. So I think we should just remove this reference.
- DateFormatUtil.parseMathLenient falls back on this, and this method is in turn called by just one caller – DateValueSourceParser, registered as ms. I don't think we need leniency in use of this function query; values given to ms should be computer generated in the ISO format.
edit: added ms().