Bugs with partial name matching during partial replacement (PR#9299)

This is a rather interesting, but I don't think it is a bug - it is
just things that "you are not supposed to do"

... you are assuming
a certain evaluation order of the 4 "$" operators in
" D$ABC[D$M] = D$V[D$M] " as in:

temp1 <- D$M                 # 2nd and 4th
temp2 <- D$V[temp1]          # 3rd
D$ABC[temp1] = temp2         # 1st

What R did was this:

temp4 <- D$ABC         # make reference, expand to D$ABCD , 1st
temp1 <- D$M           # 2nd, and 4th
temp2 <- D$V[temp1]    # 3rd

temp4[temp1] <- temp2  # oh dear, it looks as if we are
D$ABC <- temp4         # trying to write to a reference,
                       # better make a copy instead

R is doing the 4 $'s roughly from left to right, if you have some ideas
how R works inside. (I am not saying this behavior is a "good" thing,
but at least it is consistent). Basically it is a very bad habit to
write code that depends on evaluation order of operators at the same
precendence.

The difference in behavior in the two case is probably due to
coercion, (and also how lazy R does make-a-reference versus "oops, you
seems to try to write to a reference so I better copy it") but
I'll leave you to think about what order R is doing the combination of
the 4 $'s and coercing between types... Basically writing code that
depends on evaluation order is a bad idea.

c.f. this bit of C code:

i =0;
++i = ++i + ++i;

what value do you think "i" should be?

amaliy1 at uic.edu wrote:

Hello,

First the version info:
platform       powerpc-apple-darwin8.6.0
arch           powerpc
os             darwin8.6.0
system         powerpc, darwin8.6.0
status
major          2
minor          3.1
year           2006
month          06
day            01
svn rev        38247
language       R
version.string Version 2.3.1 (2006-06-01)

I have encountered some unusual behavior when trying to create new
columns in a data frame that have names that would generate a partial
match with an existing column with a longer name.  It is my
understanding that replacement operations shouldn't have partial
matching, but it is not clear to me whether this applies only when
the named column exists and not for new assignments.

The first example:

D = data.frame(M=c(T,T,F,F,F,T,F,T,F,F,T,T,T),V=I(sprintf("ZZ%02d",

1:13)),ABCD=13:1)

D

        M    V ABCD
1   TRUE ZZ01   13
2   TRUE ZZ02   12
3  FALSE ZZ03   11
4  FALSE ZZ04   10
5  FALSE ZZ05    9
6   TRUE ZZ06    8
7  FALSE ZZ07    7
8   TRUE ZZ08    6
9  FALSE ZZ09    5
10 FALSE ZZ10    4
11  TRUE ZZ11    3
12  TRUE ZZ12    2
13  TRUE ZZ13    1

D$CBA[D$M] = D$V[D$M]
D

        M    V ABCD  CBA
1   TRUE ZZ01   13 ZZ01
2   TRUE ZZ02   12 ZZ02
3  FALSE ZZ03   11 <NA>
4  FALSE ZZ04   10 <NA>
5  FALSE ZZ05    9 <NA>
6   TRUE ZZ06    8 ZZ06
7  FALSE ZZ07    7 <NA>
8   TRUE ZZ08    6 ZZ08
9  FALSE ZZ09    5 <NA>
10 FALSE ZZ10    4 <NA>
11  TRUE ZZ11    3 ZZ11
12  TRUE ZZ12    2 ZZ12
13  TRUE ZZ13    1 ZZ13

D$ABC[D$M] = D$V[D$M]
D

        M    V ABCD  CBA  ABC
1   TRUE ZZ01   13 ZZ01 ZZ01
2   TRUE ZZ02   12 ZZ02 ZZ02
3  FALSE ZZ03   11 <NA>   11
4  FALSE ZZ04   10 <NA>   10
5  FALSE ZZ05    9 <NA>    9
6   TRUE ZZ06    8 ZZ06 ZZ06
7  FALSE ZZ07    7 <NA>    7
8   TRUE ZZ08    6 ZZ08 ZZ08
9  FALSE ZZ09    5 <NA>    5
10 FALSE ZZ10    4 <NA>    4
11  TRUE ZZ11    3 ZZ11 ZZ11
12  TRUE ZZ12    2 ZZ12 ZZ12
13  TRUE ZZ13    1 ZZ13 ZZ13

I expected ABC to equal CBA with NA values in rows not assigned, but
instead it appears that an extraction from D$ABCD and coercion to
string is being performed in the process of creating D$ABC.

Here is something I believe is definitely a bug:

D = data.frame(M=c(T,T,F,F,F,T,F,T,F,F,T,T,T),V=1:13,ABCD=13:1)
D

        M  V ABCD
1   TRUE  1   13
2   TRUE  2   12
3  FALSE  3   11
4  FALSE  4   10
5  FALSE  5    9
6   TRUE  6    8
7  FALSE  7    7
8   TRUE  8    6
9  FALSE  9    5
10 FALSE 10    4
11  TRUE 11    3
12  TRUE 12    2
13  TRUE 13    1

D$CBA[D$M] = D$V[D$M]
D

        M  V ABCD CBA
1   TRUE  1   13   1
2   TRUE  2   12   2
3  FALSE  3   11  NA
4  FALSE  4   10  NA
5  FALSE  5    9  NA
6   TRUE  6    8   6
7  FALSE  7    7  NA
8   TRUE  8    6   8
9  FALSE  9    5  NA
10 FALSE 10    4  NA
11  TRUE 11    3  11
12  TRUE 12    2  12
13  TRUE 13    1  13

D$ABC[D$M] = D$V[D$M]
D

        M  V ABCD CBA ABC
1   TRUE  1    1   1   1
2   TRUE  2    2   2   2
3  FALSE  3   11  NA  11
4  FALSE  4   10  NA  10
5  FALSE  5    9  NA   9
6   TRUE  6    6   6   6
7  FALSE  7    7  NA   7
8   TRUE  8    8   8   8
9  FALSE  9    5  NA   5
10 FALSE 10    4  NA   4
11  TRUE 11   11  11  11
12  TRUE 12   12  12  12
13  TRUE 13   13  13  13

ABC is created as before with valued from ABCD in the unassigned
rows, but ABCD is being modified as well.  The only difference form
the previous example is that V is now just a numeric column.

Anil Maliyekkel

On Mon, 16 Oct 2006, hin-tak.leung at cimr.cam.ac.uk wrote:

This is a rather interesting, but I don't think it is a bug - it is
just things that "you are not supposed to do"

It was a bug. It has been fixed in R 2.4.0. Unfortunately, since you 
didn't quote the PR# of the original bug in the subject line you have 
just filed a new bug report for it.

    -thomas

Thomas Lumley wrote:

On Mon, 16 Oct 2006, hin-tak.leung at cimr.cam.ac.uk wrote:

This is a rather interesting, but I don't think it is a bug - it is
just things that "you are not supposed to do"

It was a bug. It has been fixed in R 2.4.0. Unfortunately, since you
didn't quote the PR# of the original bug in the subject line you have
just filed a new bug report for it.

    -thomas

I am sorry (about both the duplicate report and the comment).

I don't know what happened with the duplication - but as you see
I replied to both r-devel and r-bugs in the same e-mail, and the
one arriving back via r-devel did have the right subject with
the PR#9202 at the end, and the other copy via r-bugs didn't and
got a new one. It seems somewhere during in the round-trip
over-long subject lines get truncated and/or a new-line inserted in
the middle. May worth investigating...