Harden validation and preprocessing edge cases

[pedrohcgs]

Summary

This PR performs a deeper defensive hardening pass across the package's public estimation, aggregation, bootstrap, simulation, preprocessing, plotting, and exported helper surfaces. The goal is to make malformed inputs fail early with package-level diagnostics, prevent non-finite data from leaking into estimator internals, and lock down slow/fast path parity under adversarial inputs.

The release version remains 2.5.1.

Root Cause

Several user-facing controls and helper entry points were only partially validated before being used in if (...) branches, formula/model-frame evaluation, bootstrap setup, aggregation logic, data.table indexing, or low-level matrix operations. That allowed malformed inputs to produce raw errors such as the condition has length > 1, missing value where TRUE/FALSE needed, ambiguous subsetting failures, silent vector recycling, or nonsensical bootstrap results.

A second class of failures came from rows whose raw data or evaluated formula terms were non-finite. These rows could previously reach overlap/rank checks or estimator code, producing avoidable NA cells or lower-level errors. Both preprocessing paths now apply the same missing/non-finite filtering discipline before estimation.

Changes

• Added shared validators for formulas, scalar column names, column-name vectors, finite numeric scalars/vectors, probability scalars, and non-negative whole-number controls.
• Hardened att_gt(), pre_process_did(), and pre_process_did2() against malformed formulas, malformed column-name arguments, missing gname rows, non-finite outcomes, non-finite weights, and non-finite evaluated covariates.
• Made slow and fast preprocessing agree on missing/non-finite row filtering and warning text.
• Preserved gname = Inf never-treated units in the finite-data filter. complete_finite_cases() gained a finite_exclude argument, and both pre_process_did() and pre_process_did2() pass gname to it. Inf is a documented never-treated code ("group status 0 or Inf"); the new finite filter would otherwise have silently dropped every never-treated unit — warning under control_group = "notyettreated" and, worse, returning plausible-looking but wrong ATTs with no error under control_group = "nevertreated". Excluded columns still get the NA/NaN check via complete.cases(); only legitimate Inf is preserved, restoring bit-identical parity with gname = 0.
• Hardened exported helpers: trimmer(), process_attgt(), mboot(), and test.mboot() now reject malformed direct inputs before raw indexing, recycling, or matrix errors.
• Hardened build_sim_dataset() so modified simulation parameter lists cannot silently recycle wrong-length vectors or propagate NA/Inf simulation parameters.
• Hardened aggte() object validation, balance_e validation, and corrupted influence-function dimension checks.
• Added a mutation-safety regression to ensure att_gt() does not modify caller-owned data.frame or data.table inputs in either implementation.
• Added model-matrix regressions for matrix-valued transformed formula terms and non-finite evaluated terms.
• Updated robustness tests so transformed non-finite covariates are expected to be filtered before 2x2 estimator internals.
• Added regression tests that gname = Inf produces ATT and influence functions identical to gname = 0 across both code paths and both control groups, plus a complete_finite_cases() unit test confirming NA/NaN gname is still dropped while Inf is preserved.

Validation

• Rscript -e "devtools::test(filter='error-handling')": 0 fail, 0 warn, 0 skip, 276 pass
• Rscript -e "devtools::test(filter='pretest-vectorization|conditional-did-pretest')": 0 fail, 0 warn, 0 skip, 43 pass
• Rscript -e "devtools::test(filter='mboot|cluster|inference|pretest-vectorization')": 0 fail, 0 warn, 7 skip, 137 pass
• Rscript -e "devtools::test(filter='faster-mode-consistency|modelmatrix-hoist|robustness-guards|aggte|att_gt|edge-cases|slowpath-precompute|compute-inffunc|mutation-safety')": 0 fail, 12 expected warnings, 0 skip, 1050 pass
• Randomized fast/slow parity stress with formulas, weights, panel/repeated cross sections, control groups, base periods, and missing/non-finite contamination: 97 successful slow/fast comparisons, 43 rejected by both paths, 0 one-sided errors, 0 mismatches
• Rscript -e "devtools::test()": 0 fail, 0 warn, 8 skip, 1784 pass
• Rscript -e "devtools::check(document = FALSE, args = c('--no-manual'), error_on = 'never')": 0 errors, 0 warnings, 0 notes
• DESCRIPTION: Version: 2.5.1

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.