Fixed silent ignore of legacy predict flags (#12211)

[Sanidhyavijay24]

This PR addresses #12211.
It prevents predict.xgboost from silently ignoring legacy boolean flags (like predcontrib = TRUE). These flags are now mapped to the modern type argument with a deprecation warning. Additionally, any unknown arguments in ... will now trigger an error to prevent typos from causing silent failures.

[Copilot]

Pull request overview

This PR updates the R predict.xgboost S3 method to prevent silent ignoring of legacy boolean prediction flags (e.g., predcontrib = TRUE) by mapping them to the modern type argument with a deprecation warning, and to error on unsupported arguments passed via ... to avoid silent typos.

Changes:

• Map legacy predict flags provided via ... (e.g., predcontrib) to type = "<...>" and emit a deprecation warning.
• Error out when any unsupported arguments are provided in ....
• Add test coverage for the new warning/error behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
R-package/R/xgboost.R	Adds legacy-flag mapping logic for predict.xgboost(...) and rejects unknown ... args.
R-package/tests/testthat/test_xgboost.R	Adds a regression test for legacy flag mapping + unknown ... argument errors.

Comments suppressed due to low confidence (3)

R-package/R/xgboost.R:1379

• The legacy-argument mapping only inspects the first matched legacy flag (found_legacy[1]). This can fail to map a TRUE flag if an earlier legacy flag is present but FALSE (e.g., outputmargin = FALSE, predcontrib = TRUE), and it also silently ignores/overrides cases where multiple legacy flags are provided. Consider computing which legacy flags are isTRUE(), erroring on >1 TRUE (or conflicts with an explicit non-default 'type'), and then mapping the single TRUE flag deterministically.

    found_legacy <- intersect(names(dots), names(mapping))
    
    if (length(found_legacy) > 0) {
      legacy_arg <- found_legacy[1]
      if (isTRUE(dots[[legacy_arg]])) {
        type <- mapping[[legacy_arg]]
        warning(
          sprintf(
            "Argument '%s' is deprecated. Please use type = '%s' instead.",
            legacy_arg, type
          ),
          call. = FALSE
        )
      }
      dots <- dots[setdiff(names(dots), names(mapping))]
    }

R-package/R/xgboost.R:1363

• approxcontrib is mapped to type = 'contrib' here, but predict.xgboost never passes approxcontrib through to predict.xgb.Booster, so approxcontrib=TRUE will be silently ignored (and will compute exact SHAP instead of the requested approximation). Either add support for approxcontrib in predict.xgboost (and pass it through), or remove this mapping and error/warn with guidance (e.g., use predict.xgb.Booster for approxcontrib).

    mapping <- list(
      outputmargin = "raw",
      predleaf = "leaf",
      predcontrib = "contrib",
      approxcontrib = "contrib",
      predinteraction = "interaction"
    )

R-package/R/xgboost.R:1385

• If '...' contains unnamed arguments, names(dots) will include "" and the error message will render as an empty entry ("not supported ()."), which is hard to debug. Consider detecting unnamed elements and printing a placeholder like "" or including positional indices in the message.

    if (length(dots) > 0) {
      stop(
        "predict.xgboost: arguments in '...' are not supported (",
        paste(names(dots), collapse = ", "), ")."
      )

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.