time: Date returns a Time object with incorrect offset for years 1900-1905

[jinfengnarvar]

What version of Go are you using (go version)?

$ go version
go version go1.11.2 darwin/amd64

Does this issue reproduce with the latest release?

Don't know. I did repro this issue on https://play.golang.org/.

What operating system and processor architecture are you using (go env)?

$ go env
GOARCH="amd64"
GOBIN="/Users/jinfeng/dev/narvar/go/bin"
GOCACHE="/Users/jinfeng/Library/Caches/go-build"
GOEXE=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="darwin"
GOOS="darwin"
GOPATH="/Users/jinfeng/dev/narvar/go"
GOPROXY=""
GORACE=""
GOROOT="/usr/local/opt/go/libexec"
GOTMPDIR=""
GOTOOLDIR="/usr/local/opt/go/libexec/pkg/tool/darwin_amd64"
GCCGO="gccgo"
CC="clang"
CXX="clang++"
CGO_ENABLED="1"
GOMOD=""
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=/var/folders/x2/6yhw9qq129gffvpx90dv5gyw0000gp/T/go-build291093703=/tmp/go-build -gno-record-gcc-switches -fno-common"

What did you do?

Here is the repro code:

package main

import (
	"fmt"
	"time"
)

func main() {
	for i := 1900; i <= 1910; i++ {
		//tz := "Asia/Calcutta"
		tz := "America/New_York"
		loc, err := time.LoadLocation(tz)
		if err != nil {
			fmt.Println(i, "/", err)
			return
		}
		t := time.Date(i, time.January, 1, 0, 0, 0, 0, loc)
		tstr := t.Format(time.RFC3339)
		tzName, tzOffset := t.Zone()
		fmt.Println(tstr, "/", tzName, "/", tzOffset)
	}
}

What did you expect to see?

1900-01-01T00:00:00-05:00 / EST / -18000
1901-01-01T00:00:00-05:00 / EST / -18000
1902-01-01T00:00:00-05:00 / EST / -18000
1903-01-01T00:00:00-05:00 / EST / -18000
1904-01-01T00:00:00-05:00 / EST / -18000
1905-01-01T00:00:00-05:00 / EST / -18000
1906-01-01T00:00:00-05:00 / EST / -18000
1907-01-01T00:00:00-05:00 / EST / -18000
1908-01-01T00:00:00-05:00 / EST / -18000
1909-01-01T00:00:00-05:00 / EST / -18000
1910-01-01T00:00:00-05:00 / EST / -18000

What did you see instead?

1900-01-01T00:00:00-04:56 / LMT / -17762
1901-01-01T00:00:00-04:56 / LMT / -17762
1902-01-01T00:00:00-05:00 / EST / -18000
1903-01-01T00:00:00-05:00 / EST / -18000
1904-01-01T00:00:00-05:00 / EST / -18000
1905-01-01T00:00:00-05:00 / EST / -18000
1906-01-01T00:00:00-05:00 / EST / -18000
1907-01-01T00:00:00-05:00 / EST / -18000
1908-01-01T00:00:00-05:00 / EST / -18000
1909-01-01T00:00:00-05:00 / EST / -18000
1910-01-01T00:00:00-05:00 / EST / -18000

Similar output for "Asia/Calcutta" time zone, except the deviation lasts through 1905:

1900-01-01T00:00:00+05:53 / HMT / 21200
1901-01-01T00:00:00+05:53 / HMT / 21200
1902-01-01T00:00:00+05:21 / MMT / 19270
1903-01-01T00:00:00+05:21 / MMT / 19270
1904-01-01T00:00:00+05:21 / MMT / 19270
1905-01-01T00:00:00+05:21 / MMT / 19270
1906-01-01T00:08:50+05:30 / IST / 19800
1907-01-01T00:00:00+05:30 / IST / 19800
1908-01-01T00:00:00+05:30 / IST / 19800
1909-01-01T00:00:00+05:30 / IST / 19800
1910-01-01T00:00:00+05:30 / IST / 19800

[agnivade]

/cc @ALTree @ianlancetaylor

[gopherbot]

Change https://golang.org/cl/161202 mentions this issue: time: read 64-bit data if available

[ianlancetaylor]

This is happening because we were only reading the 32-bit version of the tzdata info, which can't represent times before December 13, 1901.

[jinfengnarvar]

@ianlancetaylor What about "Asia/Calcutta", of which the anomaly seems to continue beyond 1901.

[ianlancetaylor]

It seems to me that with CL 161202 the times printed for "Asia/Calcutta" match the information in the tzdata database.

# Reed S, Low F. The Indian Year Book 1936-37. Bennett, Coleman, pp 27-8.
# https://archive.org/details/in.ernet.dli.2015.282212
# This lists +052110 as Madras local time used in railways, and says that on
# 1906-01-01 railways and telegraphs in India switched to +0530.  Some
# municipalities retained their former time, and the time in Calcutta
# continued to depend on whether you were at the railway station or at
# government offices.  Government time was at +055320 (according to Shanks) or
# at +0554 (according to the Indian Year Book).  Railway time is more
# appropriate for our purposes, as it was better documented, it is what we do
# elsewhere (e.g., Europe/London before 1880), and after 1906 it was
# consistent in the region now identified by Asia/Kolkata.  So, use railway
# time for 1870-1941.  Shanks is our only (and dubious) source for the
# 1941-1945 data.

# Zone	NAME		GMTOFF	RULES	FORMAT	[UNTIL]
Zone	Asia/Kolkata	5:53:28 -	LMT	1854 Jun 28 # Kolkata
			5:53:20	-	HMT	1870	    # Howrah Mean Time?
			5:21:10	-	MMT	1906 Jan  1 # Madras local time
			5:30	-	IST	1941 Oct
			5:30	1:00	+0630	1942 May 15
			5:30	-	IST	1942 Sep
			5:30	1:00	+0630	1945 Oct 15
			5:30	-	IST

[jinfengnarvar]

Thanks @ianlancetaylor for investigation! And thanks @bradfitz for including the issue for next release!

In the meantime, I have my own wrapper around time.Date and limit the year to 1970 (an arbitrarily picked year that is just old enough :)

[jinfengnarvar]

@ianlancetaylor which version of go has this 64-bit reading fix been ported in? the label says go1.13, but I'm wondering if 1.12.4 has it or not, seems like my repro test works correctly now under 1.12.4.

[ianlancetaylor]

@jinfengnarvar The patch that fixed the problem is not on the 1.12 branch. I have no explanation. Perhaps the tzdata information on your system changed in some way?