Respect integer64= selection when out-of-sample type bumps happen in fread() (#7032)

* fixed #2749

* updated news

* added tests

* changed readInt64As to an argument instead of extern object

* Style fix

* Revert internal variable to static

---------

Co-authored-by: Ivan K <[email protected]>

Author: badasahog

NEWS.md

@@ -40,6 +40,8 @@
 
 10. On a heavily loaded machine, a `forder` thread could try to perform a zero-length copy from a null pointer, which was de-facto harmless but is against the C standard and was caught by additional CRAN checks, [#7051](https://github.com/Rdatatable/data.table/issues/7051). Thanks to @helske for the report and @aitap for the PR.
 
+11. Out of sample type bumps now respect `integer64=` selection, [#7032](https://github.com/Rdatatable/data.table/pull/7032).
+
 ### NOTES
 
 1. Continued work to remove non-API C functions, [#6180](https://github.com/Rdatatable/data.table/issues/6180). Thanks Ivan Krylov for the PRs and for writing a clear and concise guide about the R API: https://aitap.codeberg.page/R-api/.

inst/tests/tests.Rraw

@@ -3070,6 +3070,16 @@ if (test_bit64) test(1017.1, fread(f), copy(DT)[,A:=as.integer64(A)])
 test(1017.2, fread(f, integer64="character"), DT)
 unlink(f)
 
+DT = data.table(a=seq(10000), b="100")
+DT[111, b := "1000000000000"]
+f = tempfile()
+fwrite(DT, f)
+
+test(1017.3, fread(f, integer64="numeric"), fread(f, colClasses=c("integer", "numeric")))
+test(1017.4, fread(f, integer64="character"), fread(f, colClasses=c("integer", "character")))
+
+unlink(f)
+
 # ERANGE errno handled, #106 #4165
 test(1018.1, identical(fread("1.46761e-313\n"), data.table(V1=1.46761e-313)))
 test(1018.2, identical(fread("1.46761e+313\n"), data.table(V1=1.46761e+313)))

src/fread.c

@@ -2300,7 +2300,7 @@ int freadMain(freadMainArgs _args) {
   nth = imin(nJumps, nth);
 
   if (verbose) DTPRINT(_("[11] Read the data\n"));
-  while(true){  // we'll return here to reread any columns with out-of-sample type exceptions, or dirty jumps
+  for (;;) {  // we'll return here to reread any columns with out-of-sample type exceptions, or dirty jumps
     restartTeam = false;
     if (verbose)
       DTPRINT("  jumps=[%d..%d), chunk_size=%zu, total_size=%td\n", jump0, nJumps, chunkBytes, eof-pos); // # notranslate
@@ -2510,7 +2510,12 @@ int freadMain(freadMainArgs _args) {
               // check that the new type is sufficient for the rest of the column (and any other columns also in out-of-sample bump status) to be
               // sure a single re-read will definitely work.
               while (++absType<CT_STRING && disabled_parsers[absType]) {};
-              thisType = TOGGLE_BUMP(absType);
+
+              if(args.readInt64As != CT_INT64 && absType == CT_INT64)
+                thisType = TOGGLE_BUMP(args.readInt64As);
+              else
+                thisType = TOGGLE_BUMP(absType);
+
               tch = fieldStart;
             }
 

src/fread.h

@@ -167,6 +167,9 @@ typedef struct freadMainArgs
   // should datetime with no Z or UTZ-offset be read as UTC?
   bool noTZasUTC;
 
+  // Integer64 remap
+  colType readInt64As;
+
   char _padding[1];
 
   // Any additional implementation-specific parameters.

src/freadR.c

@@ -177,6 +177,8 @@ SEXP freadR(
     readInt64As = CT_FLOAT64;
   } else STOP(_("Invalid value integer64='%s'. Must be 'integer64', 'character', 'double' or 'numeric'"), tt);
 
+  args.readInt64As = readInt64As;
+
   colClassesSxp = colClassesArg;
 
   selectSxp = selectArg;