From 4ef6ab5fdddb4016c4b3015ff764c262bd616a91 Mon Sep 17 00:00:00 2001
From: badasahog <52379863+badasahog@users.noreply.github.com>
Date: Wed, 10 Sep 2025 04:16:33 -0400
Subject: [PATCH 1/3] Data desegregation + minor formatting improvements +
 constness improvements

---
 src/assign.c | 49 ++++++++++++++++++++++---------------------------
 1 file changed, 22 insertions(+), 27 deletions(-)

diff --git a/src/assign.c b/src/assign.c
index eee00cc0f5..6c48a88828 100644
--- a/src/assign.c
+++ b/src/assign.c
@@ -2,21 +2,19 @@
 
 static void finalizer(SEXP p)
 {
-  SEXP x;
-  R_len_t n, l, tl;
   if(!R_ExternalPtrAddr(p)) internal_error(__func__, "didn't receive an ExternalPtr"); // # nocov
   p = R_ExternalPtrTag(p);
   if (!isString(p)) internal_error(__func__, "ExternalPtr doesn't see names in tag"); // # nocov
-  l = LENGTH(p);
-  tl = TRUELENGTH(p);
+  const R_len_t l = LENGTH(p);
+  const R_len_t tl = TRUELENGTH(p);
   if (l<0 || tl<l) internal_error(__func__, "l=%d, tl=%d", l, tl); // # nocov
-  n = tl-l;
+  const R_len_t n = tl-l;
   if (n==0) {
     // gc's ReleaseLargeFreeVectors() will have reduced R_LargeVallocSize by the correct amount
     // already, so nothing to do (but almost never the case).
     return;
   }
-  x = PROTECT(allocVector(INTSXP, 50));  // 50 so it's big enough to be on LargeVector heap. See NodeClassSize in memory.c:allocVector
+  SEXP x = PROTECT(allocVector(INTSXP, 50));  // 50 so it's big enough to be on LargeVector heap. See NodeClassSize in memory.c:allocVector
                                          // INTSXP rather than VECSXP so that GC doesn't inspect contents after LENGTH (thanks to Karl Miller, Jul 2015)
   SETLENGTH(x,50+n*2*sizeof(void *)/4);  // 1*n for the names, 1*n for the VECSXP itself (both are over allocated).
   UNPROTECT(1);
@@ -25,11 +23,7 @@ static void finalizer(SEXP p)
 
 void setselfref(SEXP x) {
   if(!INHERITS(x, char_datatable))  return; // #5286
-  SEXP p;
-  // Store pointer to itself so we can detect if the object has been copied. See
-  // ?copy for why copies are not just inefficient but cause a problem for over-allocated data.tables.
-  // Called from C only, not R level, so returns void.
-  setAttrib(x, SelfRefSymbol, p=R_MakeExternalPtr(
+  SEXP p=R_MakeExternalPtr(
     R_NilValue,                  // for identical() to return TRUE. identical() doesn't look at tag and prot
     PROTECT(getAttrib(x, R_NamesSymbol)),  // to detect if names has been replaced and its tl lost, e.g. setattr(DT,"names",...)
     PROTECT(R_MakeExternalPtr(   // to avoid an infinite loop in object.size(), if prot=x here
@@ -37,7 +31,11 @@ void setselfref(SEXP x) {
       R_NilValue,                // this tag and prot currently unused
       R_NilValue
     ))
-  ));
+  );
+  // Store pointer to itself so we can detect if the object has been copied. See
+  // ?copy for why copies are not just inefficient but cause a problem for over-allocated data.tables.
+  // Called from C only, not R level, so returns void.
+  setAttrib(x, SelfRefSymbol, p);
   R_RegisterCFinalizerEx(p, finalizer, FALSE);
   UNPROTECT(2);
 
@@ -107,8 +105,7 @@ Moved out of ?setkey Details section in 1.12.2 (Mar 2019). Revisit this w.r.t. t
 */
 
 static int _selfrefok(SEXP x, Rboolean checkNames, Rboolean verbose) {
-  SEXP v, p, tag, prot, names;
-  v = getAttrib(x, SelfRefSymbol);
+  SEXP v = getAttrib(x, SelfRefSymbol);
   if (v==R_NilValue || TYPEOF(v)!=EXTPTRSXP) {
     // .internal.selfref missing is expected and normal for i) a pre v1.7.8 data.table loaded
     //  from disk, and ii) every time a new data.table is over-allocated for the first time.
@@ -117,20 +114,20 @@ static int _selfrefok(SEXP x, Rboolean checkNames, Rboolean verbose) {
     //  In both cases the selfref is not ok.
     return 0;
   }
-  p = R_ExternalPtrAddr(v);
+  SEXP p = R_ExternalPtrAddr(v);
   if (p==NULL) {
     if (verbose) Rprintf(_("The data.table internal attributes of this table are invalid. This is expected and normal for a data.table loaded from disk. Please remember to always setDT() immediately after loading to prevent unexpected behavior. If this table was not loaded from disk or you've already run setDT(), please report to the data.table issue tracker.\n"));
     return -1;
   }
   if (!isNull(p)) internal_error(__func__, ".internal.selfref ptr is neither NULL nor R_NilValue"); // # nocov
-  tag = R_ExternalPtrTag(v);
+  SEXP tag = R_ExternalPtrTag(v);
   if (!(isNull(tag) || isString(tag))) internal_error(__func__, ".internal.selfref tag is neither NULL nor a character vector"); // # nocov
-  names = getAttrib(x, R_NamesSymbol);
+  SEXP names = getAttrib(x, R_NamesSymbol);
   if (names!=tag && isString(names) && !ALTREP(names))  // !ALTREP for #4734
     SET_TRUELENGTH(names, LENGTH(names));
     // R copied this vector not data.table; it's not actually over-allocated. It looks over-allocated
     // because R copies the original vector's tl over despite allocating length.
-  prot = R_ExternalPtrProtected(v);
+  SEXP prot = R_ExternalPtrProtected(v);
   if (TYPEOF(prot) != EXTPTRSXP)   // Very rare. Was error(_(".internal.selfref prot is not itself an extptr")).
     return 0;                      // # nocov ; see http://stackoverflow.com/questions/15342227/getting-a-random-internal-selfref-error-in-data-table-for-r
   if (x!=R_ExternalPtrAddr(prot) && !ALTREP(x))
@@ -195,7 +192,7 @@ static SEXP shallow(SEXP dt, SEXP cols, R_len_t n)
   SET_TRUELENGTH(newdt,n);
   setselfref(newdt);
   UNPROTECT(protecti);
-  return(newdt);
+  return newdt;
 }
 
 // Wrapped in a function so the same message is issued for the data.frame case at the R level
@@ -242,14 +239,12 @@ SEXP setdt_nrows(SEXP x)
 
 SEXP alloccol(SEXP dt, R_len_t n, Rboolean verbose)
 {
-  SEXP names, klass;   // klass not class at request of pydatatable because class is reserved word in C++, PR #3129
-  R_len_t l, tl;
   if (isNull(dt)) error(_("alloccol has been passed a NULL dt"));
   if (TYPEOF(dt) != VECSXP) error(_("dt passed to alloccol isn't type VECSXP"));
-  klass = getAttrib(dt, R_ClassSymbol);
+  SEXP klass = getAttrib(dt, R_ClassSymbol);// klass not class at request of pydatatable because class is reserved word in C++, PR #3129
   if (isNull(klass)) error(_("dt passed to alloccol has no class attribute. Please report result of traceback() to data.table issue tracker."));
-  l = LENGTH(dt);
-  names = getAttrib(dt,R_NamesSymbol);
+  const R_len_t l = LENGTH(dt);
+  SEXP names = getAttrib(dt,R_NamesSymbol);
   // names may be NULL when null.data.table() passes list() to alloccol for example.
   // So, careful to use length() on names, not LENGTH().
   if (length(names)!=l) internal_error(__func__, "length of names (%d) is not length of dt (%d)", length(names),l); // # nocov
@@ -260,7 +255,7 @@ SEXP alloccol(SEXP dt, R_len_t n, Rboolean verbose)
   // if (TRUELENGTH(getAttrib(dt,R_NamesSymbol))!=tl)
   //    internal_error(__func__, "tl of dt passes checks, but tl of names (%d) != tl of dt (%d)", tl, TRUELENGTH(getAttrib(dt,R_NamesSymbol))); // # nocov
 
-  tl = TRUELENGTH(dt);
+  const R_len_t tl = TRUELENGTH(dt);
   // R <= 2.13.2 and we didn't catch uninitialized tl somehow
   if (tl<0) internal_error(__func__, "tl of class is marked but tl<0"); // # nocov
   if (tl>0 && tl<l) internal_error(__func__, "tl (%d) < l (%d) but tl of class is marked", tl, l); // # nocov
@@ -268,7 +263,7 @@ SEXP alloccol(SEXP dt, R_len_t n, Rboolean verbose)
   if (n>tl) return(shallow(dt,R_NilValue,n)); // usual case (increasing alloc)
   if (n<tl && verbose) Rprintf(_("Attempt to reduce allocation from %d to %d ignored. Can only increase allocation via shallow copy. Please do not use DT[...]<- or DT$someCol<-. Use := inside DT[...] instead."),tl,n);
         // otherwise the finalizer can't clear up the Large Vector heap
-  return(dt);
+  return dt;
 }
 
 int checkOverAlloc(SEXP x)
@@ -1265,7 +1260,7 @@ SEXP allocNAVectorLike(SEXP x, R_len_t n) {
   copyMostAttrib(x, v);
   writeNA(v, 0, n, false);
   UNPROTECT(1);
-  return(v);
+  return v;
 }
 
 static SEXP *saveds=NULL;

From e6c2489bf98ce7e51da1cd96f9413c72e85eb863 Mon Sep 17 00:00:00 2001
From: Benjamin Schwendinger <52290390+ben-schwen@users.noreply.github.com>
Date: Mon, 1 Jun 2026 09:14:45 +0200
Subject: [PATCH 2/3] restore translation

---
 src/assign.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/assign.c b/src/assign.c
index b31151b51c..f31f65215f 100644
--- a/src/assign.c
+++ b/src/assign.c
@@ -177,7 +177,7 @@ SEXP setdt_nrows(SEXP x)
 SEXP alloccol(SEXP dt, R_len_t n, Rboolean verbose)
 {
   if (isNull(dt)) error(_("alloccol has been passed a NULL dt"));
-  if (TYPEOF(dt) != VECSXP) error(_("dt passed to alloccol isn't type VECSXP"));
+  if (TYPEOF(dt) != VECSXP) error(_("dt passed to %s isn't type VECSXP"), "alloccol");
   SEXP klass = getAttrib(dt, R_ClassSymbol);// klass not class at request of pydatatable because class is reserved word in C++, PR #3129
   if (isNull(klass)) error(_("dt passed to alloccol has no class attribute. Please report result of traceback() to data.table issue tracker."));
   const R_len_t l = LENGTH(dt);

From 153815d4af2f55e080060316b23c0b681c42fe6c Mon Sep 17 00:00:00 2001
From: Benjamin Schwendinger <52290390+ben-schwen@users.noreply.github.com>
Date: Mon, 1 Jun 2026 09:15:27 +0200
Subject: [PATCH 3/3] dont use TRUELENGTH

---
 src/assign.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/assign.c b/src/assign.c
index f31f65215f..57f0e6eb9b 100644
--- a/src/assign.c
+++ b/src/assign.c
@@ -192,7 +192,7 @@ SEXP alloccol(SEXP dt, R_len_t n, Rboolean verbose)
   // if (TRUELENGTH(getAttrib(dt,R_NamesSymbol))!=tl)
   //    internal_error(__func__, "tl of dt passes checks, but tl of names (%d) != tl of dt (%d)", tl, TRUELENGTH(getAttrib(dt,R_NamesSymbol))); // # nocov
 
-  const R_len_t tl = TRUELENGTH(dt);
+  const R_len_t tl = R_maxLength(dt);
   // R <= 2.13.2 and we didn't catch uninitialized tl somehow
   if (tl<0) internal_error(__func__, "tl of class is marked but tl<0"); // # nocov
   if (tl>0 && tl<l) internal_error(__func__, "tl (%d) < l (%d) but tl of class is marked", tl, l); // # nocov