[Rd] Another issue with Sys.timezone

[Martin Maechler]

>>>>> Martin Maechler <maechler at stat.math.ethz.ch>
>>>>>     on Mon, 16 Oct 2017 19:13:31 +0200 writes:

>>>>> Stephen Berman <stephen.berman at gmx.net>
>>>>>     on Sun, 15 Oct 2017 01:53:12 +0200 writes:

>     > (I reported the test failure mentioned below to R-help but was advised
>     > that this list is the right one to address the issue; in the meantime I
>     > investigated the matter somewhat more closely, including searching
>     > recent R-devel postings, since I haven't been following this list.)
> 
>     > Last May there were two reports here of problems with Sys.timezone, one
>     > where the zoneinfo directory is in a nonstandard location
>     > (https://stat.ethz.ch/pipermail/r-devel/2017-May/074267.html) and the
>     > other where the system lacks the file /etc/localtime
>     > (https://stat.ethz.ch/pipermail/r-devel/2017-May/074275.html).  My
>     > system exhibits a third case: it lacks /etc/timezone and does not set TZ
>     > systemwide, but it does have /etc/localtime, which is a copy of, rather
>     > than a symlink to, a file under zoneinfo.  On this system Sys.timezone()
>     > returns NA and the Sys.timezone test in reg-tests-1d fails.  However, on
>     > my system I can get the (abbreviated) timezone in R by using as.POSIXlt,
>     > e.g. as.POSIXlt(Sys.time())$zone.  If Sys.timezone took advantage of
>     > this, e.g. as below, it would be useful on such systems as mine and the
>     > regression test would pass.
> 
>     > my.Sys.timezone <- 
>     > 	function (location = TRUE) 
>     > {
>     > 	tz <- Sys.getenv("TZ", names = FALSE)
>     > 	if (!location || nzchar(tz)) 
>     > 	    return(Sys.getenv("TZ", unset = NA_character_))
>     > 	lt <- normalizePath("/etc/localtime")
>     > 	if (grepl(pat <- "^/usr/share/zoneinfo/", lt) ||
>     > 	    grepl(pat <- "^/usr/share/zoneinfo.default/", lt)) 
>     > 	    sub(pat, "", lt)
>     > 	else if (lt == "/etc/localtime")
>     > 	    if (!file.exists("/etc/timezone"))
>     > 		return(as.POSIXlt(Sys.time())$zone)
>     > 	    else if (dir.exists("/usr/share/zoneinfo") && {
>     > 		info <- file.info(normalizePath("/etc/timezone"), extra_cols = FALSE)
>     > 		(!info$isdir && info$size <= 200L)
>     > 	    } && {
>     > 		tz1 <- tryCatch(readBin("/etc/timezone", "raw", 200L), 
>     > 				error = function(e) raw(0L))
>     > 		length(tz1) > 0L && all(tz1 %in% as.raw(c(9:10, 13L, 32:126)))
>     > 	    } && {
>     > 		tz2 <- gsub("^[[:space:]]+|[[:space:]]+$", "", rawToChar(tz1))
>     > 		tzp <- file.path("/usr/share/zoneinfo", tz2)
>     > 		file.exists(tzp) && !dir.exists(tzp) &&
>     > 		    identical(file.size(normalizePath(tzp)), file.size(lt))
>     > 	    }) 
>     > 		tz2
>     > 	    else NA_character_
>     > }
> 
>     > One problem with this is that the zone component of as.POSIXlt only
>     > holds the abbreviated timezone, not the Olson name.  
> 
> Yes, indeed.  So, really only for  Sys.timezone(location = FALSE)  this
> should be given, for the default  location = TRUE   it should
> still give NA (i.e. NA_character_)  in your setup.
> 
> Interestingly, the Windows versions of Sys.timezone(location =
> FALSE) uses something like your proposal,  and I tend to think that
> -- again only for location=FALSE -- this should be used on
> on-Windows as well, at least instead of returning  NA  then.
> 
> Also for me on 3 different Linuxen (Fedora 24, F. 26, and ubuntu
> 14.04 LTS), I get
> 
>   > Sys.timezone()
>   [1] "Europe/Zurich"
>   > Sys.timezone(FALSE)
>   [1] NA
>   > 
> 
> whereas on Windows I get Europe/Berlin for the first (why on
> earth - I'm really in Zurich) and get  "CEST" ("Central European Summer Time") 
> for the 2nd one instead of NA ... simply using a smarter version
> of your proposal.   The windows source is
> in R's source at  src/library/base/R/windows/system.R :
> 
> Sys.timezone <- function(location = TRUE)
> {
>     tz <- Sys.getenv("TZ", names = FALSE)
>     if(nzchar(tz)) return(tz)
>     if(location) return(.Internal(tzone_name()))
>     z <- as.POSIXlt(Sys.time())
>     zz <- attr(z, "tzone")
>     if(length(zz) == 3L) zz[2L + z$isdst] else zz[1L]
> }
> 
> >From what I read, the last three lines also work in your setup
> where it seems zz would be of length 1, right ?
> 
> I'd really propose to use these 3 lines in the non-Windows
> version of Sys.timezone .. at the end *instead* of NA_character_
> (or a slightly safer version which gives  NA_character_ if zz is
> of length 0 {e.g. if there is no "tzone" attribute}.
> 
>     > i don't know how to
>     > get the Olson name using only R functions, but maybe it would be good
>     > enough to return the abbreviated timezone where possible, e.g. as above.
>     > (On my system I can get the Olson name of the timezone in R with a shell
>     > pipeline, e.g.: system("find /usr/share/zoneinfo/ -type f | xargs md5sum
>     > | grep $(md5sum /etc/localtime | cut -d ' ' -f 1) | head -n 1 | cut -d
>     > '/' -f 5,6"), but the last part of this is tailored to my configuration
>     > and the whole thing is not OS-neutral, so it isn't suitable for
>     > Sys.timezone.)
> 
>     > Steve Berman
> 
> Definitely not.  I still recommend you think of a more portable
> solution for the   `location = TRUE` (default) case in Sys.timezone().
> Returning the non-location form (e.g "CEST") when something like
> "Europe/Zurich" is expected is really not a good idea,
> and you are lucky that the regression test passes "accidentally" ...
> 
> Martin

In the mean time, I have committed a common version (Windows and
non-Windows) of  Sys.timezone()  to the R development sources
(aka "R-devel").

That now uses  as.POSIXlt(Sys.time())  very similarly to the
above "Windows only" case,  but __only__ for  'location=FALSE'
which is not the default.

The most current development source is always available (via
'svn' or alternatively for browsing via your web browser) from

   https://svn.r-project.org/R/trunk/src/library/base/R/datetime.R

As you say yourself, the above system("... xargs md5sum ...")
using workaround is really too platform specific  but I'd guess
there should be a less error prone way to get the long timezone
name on your system ...
If that remains "contained" (i.e. small) and works with files
and R's files tools -- e.g. file.*() ones [but not system()],
I'd consider a patch to the above source file
(sent by you to the R-devel mailing list --- or after having
 gotten an account there by asking, via bug report & patch
 attachment at https://bugs.r-project.org/ )

Best,
Martin

> 
> --
> Martin <Maechler at stat.math.ethz.ch>   http://stat.ethz.ch/~maechler
> Seminar für Statistik, ETH Zürich
> and R Core Team