Segmentation fault while converting invalid date.

[amerocu]

Constructing some wrong Dates I have managed to reach a segmentation fault error. While I guess building correct dates is a user duty, it seems to me a bit too much to have segmentation fault for these cases.

Attached an example to reproduce.

import Chronos
import Data.Int

validDate =
    Date
      (Year 2025)
      (Month 0) -- valid month January
      (DayOfMonth 10)

validDay = dateToDay validDate
-- λ> validDay
-- Day {getDay = 60685}

brokenDate =
    Date
      (Year 2025)
      (Month (-1)) -- clearly not a valid month
      (DayOfMonth 10)

bangDay = dateToDay brokenDate
-- λ> bangDay
-- Day {getDay = Segmentation fault (core dumped)

alsoBrokenDate =
    Date
      (Year 2025)
      (Month 13) -- also clearly not a valid month
      (DayOfMonth 10)

alsoBangDay = dateToDay alsoBrokenDate
-- λ> alsoBangDay
-- Day {getDay = Segmentation fault (core dumped)

brokenButNotQuiteDate =
    Date
      (Year 2025)
      (Month 12) -- not a valid month 0..11
      (DayOfMonth 10)

notBangButShoudlBangDay = dateToDay brokenButNotQuiteDate
-- λ> notBangButShoudlBangDay
-- Day {getDay = 61040}

what = (dayToDate . dateToDay) brokenButNotQuiteDate
-- λ> what
-- Date { dateYear = Year {getYear = 2025}
--      , dateMonth = Month {getMonth = 11}
--      , dateDay = DayOfMonth {getDayOfMonth = 31}
--      }

As s bonus the roundtrip for Dates on the thirteenth month just does not make sense.

[mikeplus64]

Shopping around for a higher performance time library and can reproduce this with the provided examples. Somewhat terrifying!

The error seems to happen inside monthDateToDayOfYear:

  >>> import Chronos
  >>> brokenDate = Date (Year 2025) (Month (-1)) (DayOfMonth 10)
  >>> let (Date y m d) = brokenDate in (monthDateToDayOfYear (isLeapYear y) (MonthDate m d))
DayOfYear {getDayOfYear = [1]    80668 segmentation fault (core dumped) 

Looking through, it's almost certainly caseMonth. It uses unsafeIndex, assuming I guess that Month is actually a correct month, and blows up when not.

[andrewthad]

I would take a PR that adds bounds checking before indexing into the arrays in these cases. The result would still be a program crash, but at least it would be an informative error message instead of a segfault.