fix(7407): added check for x1a in fread.c to avoid segfault

NEWS.md

@@ -28,6 +28,8 @@
 
 4. `sum(<int64 column>)` by group is correct with missing entries and GForce activated ([#7571](https://github.com/Rdatatable/data.table/issues/7571)). Thanks to @rweberc for the report and @manmita for the fix. The issue was caused by a faulty early `break` that spilled between groups, and resulted in silently incorrect results!
 
+5. `fread(text=)` could segfault when reading text input ending with a `\x1a` (ASCII SUB) character after a long line, [#7407](https://github.com/Rdatatable/data.table/issues/7407) which is solved by adding check for eof. Thanks @aitap for the report and @manmita for the fix.  
+
 ### Notes
 
 1. {data.table} now depends on R 3.5.0 (2018).

inst/tests/tests.Rraw

@@ -22031,3 +22031,7 @@ if (test_bit64) local({
     merged$gforce_mean, merged$true_mean
   )
 })
+
+# 7407 Test for fread() handling \x1A (ASCII SUB) at end of input
+txt = paste0("foo\n", strrep("a", 4096 * 100), "\x1A")
+test(2359.1, nchar(fread(txt)$foo), 409600L)

src/fread.c

@@ -349,9 +349,9 @@ static inline bool end_of_field(const char *ch)
   // We use eol() because that looks at eol_one_r inside it w.r.t. \r
   // \0 (maybe more than one) before eof are part of field and do not end it; eol() returns false for \0 but the ch==eof will return true for the \0 at eof.
   // Comment characters terminate a field immediately and take precedence over separators.
-  return *ch == sep || ((uint8_t)*ch <= 13 && (ch == eof || eol(&ch))) || (commentChar && *ch == commentChar);
   if (*ch == sep) return true;
-  if ((uint8_t)*ch <= 13 && (ch == eof || eol(&ch))) return true;
+  if (ch == eof) return true;
+  if ((uint8_t)*ch <= 13 && eol(&ch)) return true;
   if (!commentChar) return false;
   return *ch == commentChar;
 }