32-bit section missing transitions in 2021b
Hi, It looks like 32-bit data generated by 2021b is missing the effects of transitions below INT32_MIN. Before commit 93132d18, zones had a transition at INT32_MIN to the type in use at that time. For example, the Pacific/Port_Moresby zone used to have these two transitions: -2147483648 -> 1 (+10) 2147483647 -> 1 (+10) Starting with commit 93132d18, including in 2021b, only one transition is present: 2147483647 -> 1 (+10) Since clients are supposed to assume type 0 before the earliest transition, this almost always leads to wrong times for some period of time (or all times for extreme cases like Port Moresby). Danny -- Daniel Fischer Rastatt, Germany Oracle Global Services Germany GmbH
On 9/29/21 9:53 AM, Daniel Fischer via tz wrote:
It looks like 32-bit data generated by 2021b is missing the effects of transitions below INT32_MIN.
Thanks for reporting that regression in 2021b, which was caused by the recent changes to support TZif file truncation. I installed the attached patch, which fixed things for me. Please give it a try. This should go into 2021c which I hope to publish soon.
On Thu, 30 Sept 2021 at 23:56, Daniel Fischer via tz <tz@iana.org> wrote:
I installed the attached patch, which fixed things for me. Please give it a try.
Thank you. It looks like this patch fixes it for me too.
And has fixed problems with pytz, which I thought I was going to lose my weekend over. Thanks! -- Stuart Bishop <stuart@stuartbishop.net> http://www.stuartbishop.net/
participants (3)
-
Daniel Fischer -
Paul Eggert -
Stuart Bishop