[<-.POSIXlt changes order of attributes (PR#9197) - R-devel

Fri, Sep 1, 2006 4:58 AM #

This is a multi-part message in MIME format.
--------------090203020600020104020707
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

Hello!

I was doing some tests with identical() and found out that [<-.POSIXlt
method changes order of attributes. This example shows that:

x <- strptime("1900-1-1", format="%Y-%m-%d")
x <- c(x)
y <- c(x, x+1)
x1 <- x
y1 <- y

attributes(x)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$class
[1] "POSIXt"  "POSIXlt"

$tzone
[1] ""     "CET"  "CEST"

identical(attributes(x), attributes(y))
TRUE

x[1] <- NA
attributes(x)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$tzone
[1] ""     "CET"  "CEST"

$class
[1] "POSIXt"  "POSIXlt"

y[1] <- NA
attributes(y)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$tzone
[1] ""     "CET"  "CEST"

$class
[1] "POSIXt"  "POSIXlt"

identical(attributes(x), attributes(x1))
FALSE

identical(attributes(y), attributes(y1))
FALSE

This can be solved with either:

Index: R/src/library/base/R/datetime.R
===================================================================
--- R/src/library/base/R/datetime.R     (revision 39045)
+++ R/src/library/base/R/datetime.R     (working copy)
@@ -713,10 +713,10 @@
 {
     if(!as.logical(length(value))) return(x)
     value <- as.POSIXlt(value)
-    cl <- oldClass(x)
+    att <- attributes(x)
     class(x) <- class(value) <- NULL
     for(n in names(x)) x[[n]][i] <- value[[n]]
-    class(x) <- cl
+    attributes(x) <- att
     x
 }

or in the same way as it is done in [.<-POSIXct method

Index: R/src/library/base/R/datetime.R
===================================================================
--- R/src/library/base/R/datetime.R     (revision 39045)
+++ R/src/library/base/R/datetime.R     (working copy)
@@ -714,9 +714,11 @@
     if(!as.logical(length(value))) return(x)
     value <- as.POSIXlt(value)
     cl <- oldClass(x)
+    tz <- attr(x, "tzone")
     class(x) <- class(value) <- NULL
     for(n in names(x)) x[[n]][i] <- value[[n]]
     class(x) <- cl
+    attr(x, "tzone") <- tz
     x
 }

I have checked both versions in r-devel with make check-all and all went
fine.

Regards, Gregor

Lep pozdrav / With regards,
    Gregor Gorjanc

----------------------------------------------------------------------
University of Ljubljana     PhD student
Biotechnical Faculty
Zootechnical Department     URI: http://www.bfro.uni-lj.si/MR/ggorjan
Groblje 3                   mail: gregor.gorjanc <at> bfro.uni-lj.si

SI-1230 Domzale             tel: +386 (0)1 72 17 861
Slovenia, Europe            fax: +386 (0)1 72 17 888

----------------------------------------------------------------------
"One must learn by doing the thing; for though you think you know it,
 you have no certainty until you try." Sophocles ~ 450 B.C.
----------------------------------------------------------------------

--------------090203020600020104020707
Content-Type: application/x-gzip;
 name="datetime.R.patchVer1.gz"
Content-Transfer-Encoding: base64
Content-Disposition: inline;
 filename="datetime.R.patchVer1.gz"

H4sICH7v90QCA2RhdGV0aW1lLlIucGF0Y2hWZXIxAKVOz0vDMBQ+m7/ieUvosmxMGYpCwdNg
qFSEQekhTbMazBJJ0tkh/u+atE5vO/guyft+vW9lGtlfQ8G8E0yr2nF3YDX3khWs4UEGtZPT
At3+fxCl9NSdM+zkXnllDSyuZheXBGVZdtr0bt2rMi0I+3YgKM+BLueLyXwG2fjmOYIPBHHU
Fp9zP9W2VYJrrKVpwwvec91JQgg4GTpncE8GdcLhhsK35fHhabXRYdQiGnmhI2l1c6e599GW
RZiHkEwhOFV3QfpjoBh1kR7+Q1zc75/X60G1tQ4bUAYM3yUzgb4sTVWVqorK5EnAT42/qccO
v8fHNkN6j+ATAfoCTtG/r/sBAAA=
--------------090203020600020104020707
Content-Type: application/x-gzip;
 name="datetime.R.patchVer2.gz"
Content-Transfer-Encoding: base64
Content-Disposition: inline;
 filename="datetime.R.patchVer2.gz"

H4sICCXw90QCA2RhdGV0aW1lLlIucGF0Y2hWZXIyAKWOwUoDMRCGz+Ypxp4S0jQuVqSisOCp
UFRWBGHZQ7qbrsE0kSSt6YLvrtmIBxF6cC5J/vm+mSxNJ+MVVNy7lmu1dsId+Fp4ySveiSCD
2spZhW7+X4gxdmzPCXZyr7yyBs4XZ/MLgiilx6V3616V6aG1bweCyhLYZTGfLoCmoyigLBGk
Uht8KvxM2161QmMtTR9e8F7onSSEgJNh5wyOJNNjDtcMvpSH+8flsw7fbO63OjWt7m618D5p
NMVhGJ0QHI5TmITBGjn5UTKZiHzPA9P77mm1ytTGOmxAGTBiKxNOINa1aZpaNYkcnTH4a2r+
xe/9qReGzEcEHwjQJ/Ktnoj7AQAA
--------------090203020600020104020707--

Brian Ripley

Fri, Sep 1, 2006 8:18 AM #

Please point us to the documentation that says attributes are ordered.
E.g. R-lang.texi says

  All objects except @code{NULL} can have one or more attributes attached
  to them.  Attributes are stored as a list where all elements are named.

(although in fact they are stored in a pairlist).

I know of tens of functions that change the order of attributes, and 
did look at teaching identical() that they are not ordered.  But 
since they are stored as a pairlist, it would be quite expensive.
(Given that attributes seem to be growing in use, another internal storage 
mechanism is becoming more appropriate.)

BTW, all.equal() does not assume attributes are ordered.

On Fri, 1 Sep 2006, gregor.gorjanc at bfro.uni-lj.si wrote:

This is a multi-part message in MIME format.
--------------090203020600020104020707
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

Hello!

I was doing some tests with identical() and found out that [<-.POSIXlt
method changes order of attributes. This example shows that:

x <- strptime("1900-1-1", format="%Y-%m-%d")
x <- c(x)
y <- c(x, x+1)
x1 <- x
y1 <- y

attributes(x)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$class
[1] "POSIXt"  "POSIXlt"

$tzone
[1] ""     "CET"  "CEST"

identical(attributes(x), attributes(y))
TRUE

x[1] <- NA
attributes(x)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$tzone
[1] ""     "CET"  "CEST"

$class
[1] "POSIXt"  "POSIXlt"

y[1] <- NA
attributes(y)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$tzone
[1] ""     "CET"  "CEST"

$class
[1] "POSIXt"  "POSIXlt"

identical(attributes(x), attributes(x1))
FALSE

identical(attributes(y), attributes(y1))
FALSE

This can be solved with either:

Index: R/src/library/base/R/datetime.R
===================================================================
--- R/src/library/base/R/datetime.R     (revision 39045)
+++ R/src/library/base/R/datetime.R     (working copy)
@@ -713,10 +713,10 @@
 {
     if(!as.logical(length(value))) return(x)
     value <- as.POSIXlt(value)
-    cl <- oldClass(x)
+    att <- attributes(x)
     class(x) <- class(value) <- NULL
     for(n in names(x)) x[[n]][i] <- value[[n]]
-    class(x) <- cl
+    attributes(x) <- att
     x
 }

or in the same way as it is done in [.<-POSIXct method

Index: R/src/library/base/R/datetime.R
===================================================================
--- R/src/library/base/R/datetime.R     (revision 39045)
+++ R/src/library/base/R/datetime.R     (working copy)
@@ -714,9 +714,11 @@
     if(!as.logical(length(value))) return(x)
     value <- as.POSIXlt(value)
     cl <- oldClass(x)
+    tz <- attr(x, "tzone")
     class(x) <- class(value) <- NULL
     for(n in names(x)) x[[n]][i] <- value[[n]]
     class(x) <- cl
+    attr(x, "tzone") <- tz
     x
 }

I have checked both versions in r-devel with make check-all and all went
fine.

Regards, Gregor

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Brian Ripley

Sat, Sep 2, 2006 9:41 AM #

There is a bug in identical(), though:

[1] TRUE

[1] TRUE

so identical() is not even looking at the names of the attributes but only 
checking their values.  Oops ....

On Fri, 1 Sep 2006, Prof Brian Ripley wrote:

Please point us to the documentation that says attributes are ordered.
E.g. R-lang.texi says

  All objects except @code{NULL} can have one or more attributes attached
  to them.  Attributes are stored as a list where all elements are named.

(although in fact they are stored in a pairlist).

I know of tens of functions that change the order of attributes, and 
did look at teaching identical() that they are not ordered.  But 
since they are stored as a pairlist, it would be quite expensive.
(Given that attributes seem to be growing in use, another internal storage 
mechanism is becoming more appropriate.)

BTW, all.equal() does not assume attributes are ordered.


On Fri, 1 Sep 2006, gregor.gorjanc at bfro.uni-lj.si wrote:

This is a multi-part message in MIME format.
--------------090203020600020104020707
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

Hello!

I was doing some tests with identical() and found out that [<-.POSIXlt
method changes order of attributes. This example shows that:

x <- strptime("1900-1-1", format="%Y-%m-%d")
x <- c(x)
y <- c(x, x+1)
x1 <- x
y1 <- y

attributes(x)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$class
[1] "POSIXt"  "POSIXlt"

$tzone
[1] ""     "CET"  "CEST"

identical(attributes(x), attributes(y))
TRUE

x[1] <- NA
attributes(x)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$tzone
[1] ""     "CET"  "CEST"

$class
[1] "POSIXt"  "POSIXlt"

y[1] <- NA
attributes(y)
$names
[1] "sec"   "min"   "hour"  "mday"  "mon"   "year"  "wday"  "yday"  "isdst"

$tzone
[1] ""     "CET"  "CEST"

$class
[1] "POSIXt"  "POSIXlt"

identical(attributes(x), attributes(x1))
FALSE

identical(attributes(y), attributes(y1))
FALSE

This can be solved with either:

Index: R/src/library/base/R/datetime.R
===================================================================
--- R/src/library/base/R/datetime.R     (revision 39045)
+++ R/src/library/base/R/datetime.R     (working copy)
@@ -713,10 +713,10 @@
 {
     if(!as.logical(length(value))) return(x)
     value <- as.POSIXlt(value)
-    cl <- oldClass(x)
+    att <- attributes(x)
     class(x) <- class(value) <- NULL
     for(n in names(x)) x[[n]][i] <- value[[n]]
-    class(x) <- cl
+    attributes(x) <- att
     x
 }

or in the same way as it is done in [.<-POSIXct method

Index: R/src/library/base/R/datetime.R
===================================================================
--- R/src/library/base/R/datetime.R     (revision 39045)
+++ R/src/library/base/R/datetime.R     (working copy)
@@ -714,9 +714,11 @@
     if(!as.logical(length(value))) return(x)
     value <- as.POSIXlt(value)
     cl <- oldClass(x)
+    tz <- attr(x, "tzone")
     class(x) <- class(value) <- NULL
     for(n in names(x)) x[[n]][i] <- value[[n]]
     class(x) <- cl
+    attr(x, "tzone") <- tz
     x
 }

I have checked both versions in r-devel with make check-all and all went
fine.

Regards, Gregor

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Gregor Gorjanc

Sat, Sep 2, 2006 1:02 PM #

Hello,

Prof Brian Ripley wrote:

I agree that there is no need for attributes to be always in the same
order as this is tedious to achieve as you noted above, but identical()
seems to care about this.

Which internal storage mechanism are you talking about here?

Lep pozdrav / With regards,
    Gregor Gorjanc

----------------------------------------------------------------------
University of Ljubljana     PhD student
Biotechnical Faculty
Zootechnical Department     URI: http://www.bfro.uni-lj.si/MR/ggorjan
Groblje 3                   mail: gregor.gorjanc <at> bfro.uni-lj.si

SI-1230 Domzale             tel: +386 (0)1 72 17 861
Slovenia, Europe            fax: +386 (0)1 72 17 888

----------------------------------------------------------------------
"One must learn by doing the thing; for though you think you know it,
 you have no certainty until you try." Sophocles ~ 450 B.C.

Gregor Gorjanc

Sat, Sep 2, 2006 1:05 PM #

Hello,

Prof Brian Ripley wrote:

Aha, so this is the problem then. Thank you for tracking down the problem!

Lep pozdrav / With regards,
    Gregor Gorjanc

----------------------------------------------------------------------
University of Ljubljana     PhD student
Biotechnical Faculty
Zootechnical Department     URI: http://www.bfro.uni-lj.si/MR/ggorjan
Groblje 3                   mail: gregor.gorjanc <at> bfro.uni-lj.si

SI-1230 Domzale             tel: +386 (0)1 72 17 861
Slovenia, Europe            fax: +386 (0)1 72 17 888

----------------------------------------------------------------------
"One must learn by doing the thing; for though you think you know it,
 you have no certainty until you try." Sophocles ~ 450 B.C.

Brian Ripley

Sat, Sep 2, 2006 10:07 PM #

On Sat, 2 Sep 2006, Gregor Gorjanc wrote:

The one mentioned at

	'Attributes are stored as a list'

(but in fact a pairlist). We may want to change to e.g. a hash table.

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595