inst changes old dates in Clojure 1.12 alpha10

Question

UPDATE:

https://clojure.org/news/2024/05/23/clojure-1-12-alpha12

Alpha12 reverted the change that caused this issue. This discussion is therefore obsolete.

---

I just noticed that Clojure 1.12 alpha10 has changed the way old #inst dates are created. As far as I can tell, Java is trying to adjust old dates before the Gregorian calendar was introduced.

In Clojure 1.11.3

user-> #inst "1582-02-24"
#inst "1582-02-24T00:00:00.000-00:00"

In Clojure 1.12-alpha10

user=> #inst "1582-02-24"
#inst "1582-03-06T00:00:00.000-00:00"

I found this when I ran some generative tests on an old project that tested dates. I'm changing my tests to look only at this millennium, which is fine for my application. I don't think Clojure should necessarily do anything to mitigate this change. I just thought it was worth a mention in case someone else bumps into it.

Updated to Clojure 1.12 alpha 11:

user=> (use 'clojure.instant)
nil
user=> (read-instant-date "1582-02-24")
#inst "1582-03-06T00:00:00.000-00:00"
user=> (read-instant-calendar "1582-02-24")
#inst "1582-02-24T00:00:00.000+00:00"
user=> (read-instant-timestamp "1582-02-24")
#inst "1582-03-06T00:00:00.000000000-00:00"

Comments

I'm learning a bit about the history of the transition from the Julian calendar to the Gregorian, and how proleptic Gregorian dates are handled in modern Java.  I'm starting to think that #inst should avoid proleptic dates altogether but I'm sure someone else has a more informed opinion.  Here are a couple of useful StackOverflow questions:

https://stackoverflow.com/questions/23975205/why-does-converting-java-dates-before-1582-to-localdate-with-instant-give-a-diff

https://stackoverflow.com/questions/23457470/calendar-to-date-conversion-for-dates-before-15-oct-1582-gregorian-to-julian-ca

---

I think the right thing to do is to reimplement most of instant.clj to use java.time.Instant as the basis.  That seems to fix things for me.

---

Logged as https://clojure.atlassian.net/browse/CLJ-2849

Answer 1

Please disregard the below - I tested it using clojure.instant/read-instant-date directly and it seems that that's not what #inst ends up using in an plain REPL.

---

This has nothing to do with Java since changing the version of Clojure doesn't by itself change the version of Java that you're using.

A quick experiment shows that both versions parse the date in the same way - the call to (.getTime ...) returns the same -12238560000000.

So it seems that what's different is how the instances are printed. Commit 213c50e7 fixes CLJ-2803 by switching to a different printing method.

By using the old printing method manually with 1.12.0-alpha10, we can confirm the guess:

(let [fmt (java.text.SimpleDateFormat. "yyyy-MM-dd'T'HH:mm:ss.SSS-00:00")]
  (.setTimeZone fmt (java.util.TimeZone/getTimeZone "GMT"))
  (.format fmt d))
=> "1582-02-24T00:00:00.000-00:00"

Comments

How to quickly lose one's mind without bothering anyone:

    user=> (.getTime #inst "1582-02-24")
    -12237696000000
    user=> (let [d #inst "1582-02-24"] (.getTime d))
    -12237696000000
    user=> (def d #inst "1582-02-24")
    #'user/d
    user=> (.getTime d)
    -12238560000000  ;; What the hell?