Should setnames() produce duplicate column names without warning?

[sindribaldur]

IRIS  <- as.data.table(iris)
setnames(IRIS, "Petal.Length", "Sepal.Length") # No warning (but maybe there should be?)
IRIS[, Sepal.Length] # Only first occurrence is returned
setnames(IRIS, "Sepal.Length", "abc.Length") #  Warning: Only first occurrence is changed

Has been discussed at length here. Without conclusion?

[jangorecki]

There is a comment

dups allowed without warning; warn if and when the dup causes an ambiguity

in

data.table/R/data.table.R

Line 2460 in 6f17d18

# if (anyDuplicated(new)) warning("Some duplicates exist in 'new': ", brackify(new[duplicated(new)])) # dups allowed without warning; warn if and when the dup causes an ambiguity

is there any ambiguity here?
not in the first expression, second also seems OK, and third... should we rename both columns to a new name?

---

rather than micromanaging that here I would prefer to have global option to stop if there are dups in column names, at any stage.

[venom1204]

Hi @jangorecki,
I am planning to implement a fix for this issue based on your suggestion to use a centralized approach rather than micromanaging duplicate checks across different functions.
Proposed Plan:

• onLoad.R: Register a new global option datatable.warn.duplicate.names (defaulting to FALSE). This ensures backward compatibility for users who rely on the current behavior.
• utils.R: Add an internal helper warn_if_duplicate_names(). This centralizes the logic so we can easily extend the warning to rbindlist, merge, or setDT in the future without duplicating code.
• data.table.R: Update setnames() to construct a "preview" of the resulting names and call the utility helper before committing changes.

if this sounds good i will be happy to proceed with the pr.

[ben-schwen]

My 2 cents on this.

Maybe we can use something like datatable.unique.names and then options like c("warn", "error", "rename", "off"), where

• "warn" should return a warning
• `"error" should return an error
• "rename" should fix it, e.g. with make.names(, unique=TRUE)
• "off" should do nothing

And we can gradually escalate from warning to error and eventually even to rename since one of data.tables "promises" is that it just works

[venom1204]

Got it, that makes sense. I’ll implement the policy-based approach (off, warn, error, rename) as suggested and open a PR soon. Thanks for the feedback!

My 2 cents on this.

Maybe we can use something like datatable.unique.names and then options like c("warn", "error", "rename", "off"), where

• "warn" should return a warning
• `"error" should return an error
• "rename" should fix it, e.g. with make.names(, unique=TRUE)
• "off" should do nothing

And we can gradually escalate from warning to error and eventually even to rename since one of data.tables "promises" is that it just works