fix fread(..., nrows=0) infinite loop (#5869)

* fix fread infinite loop nrows=0

Author: ben-schwen

NEWS.md

@@ -299,7 +299,7 @@
 
 2. `print(DT, trunc.cols=TRUE)` and the corresponding `datatable.print.trunc.cols` option (new feature 3 in v1.13.0) could incorrectly display an extra column, [#4266](https://github.com/Rdatatable/data.table/issues/4266). Thanks to @tdhock for the bug report and @MichaelChirico for the PR.
 
-3. `fread(..., nrows=0L)` now works as intended and the same as `nrows=0`; i.e. returning the column names and typed empty columns determined by the large sample, [#4686](https://github.com/Rdatatable/data.table/issues/4686), [#4029](https://github.com/Rdatatable/data.table/issues/4029). Thanks to @hongyuanjia and @michaelpaulhirsch for reporting, and Benjamin Schwendinger for the PR.
+3. `fread(..., nrows=0L)` now works as intended and the same as `nrows=0`; i.e. returning the column names and typed empty columns determined by the large sample, [#4686](https://github.com/Rdatatable/data.table/issues/4686), [#4029](https://github.com/Rdatatable/data.table/issues/4029). Thanks to @hongyuanjia and @michaelpaulhirsch for reporting, and Benjamin Schwendinger for the PR. Also thanks to @HughParsonage for testing dev and reporting a bug which was fixed before release.
 
 4. Passing `.SD` to `frankv()` with `ties.method='random'` or with `na.last=NA` failed with `.SD is locked`, [#4429](https://github.com/Rdatatable/data.table/issues/4429). Thanks @smarches for the report.
 

inst/tests/tests.Rraw

@@ -13284,6 +13284,12 @@ test(1958.19, fread(text=txt, nrows=0L, verbose=TRUE), data.table(A=integer(), B
 txt = paste(c("A,B\n1,2\n", rep("3,4\n",5100), "3,4.1\n", rep("5,6\n",4900)), collapse="")
 test(1958.20, fread(text=txt, nrows=0L, verbose=TRUE), data.table(A=integer(), B=integer()),
               output="Sampled 1049 rows.*at 11 jump points")
+# invalid head position for nrows=0 #5868
+# following example has to be big enough to trigger a jump and enough blanks so we get an invalid header position after a jump (at least n=599101 for following two cols)
+test(1958.21, fread(sprintf('%s%s\n', 'a,b', paste0('\n', sample(c('', '3'), 6e5L, TRUE), sample(c(',a',',b',',c'), 6e5L, TRUE), collapse="")), nrows=0L),
+  data.table(a=integer(), b=character()))
+test(1958.22, fread(sprintf('%s%s\n', 'a,b', paste0('\n', sample(c('', '3'), 6e5L, TRUE), sample(c(',a',',b',',c'), 6e5L, TRUE), collapse="")), nrows=0L, nThread=1L),
+  data.table(a=integer(), b=character()))
 
 # Skip should work with all types of newlines #3006
 eols = c("\n", "\r\n", "\r", "\n\r")

src/fread.c

@@ -2431,7 +2431,7 @@ int freadMain(freadMainArgs _args) {
         if (stopTeam) {             // A previous thread stopped while I was waiting my turn to enter ordered
           myNrow = 0;               // # nocov; discard my buffer
         }
-        else if (headPos!=thisJumpStart) {
+        else if (headPos!=thisJumpStart && nrowLimit>0) { // do not care for dirty jumps since we do not read data and only want to know types
            // # nocov start
           snprintf(internalErr, internalErrSize, _("Internal error: invalid head position. jump=%d, headPos=%p, thisJumpStart=%p, sof=%p"), jump, (void*)headPos, (void*)thisJumpStart, (void*)sof);
           stopTeam = true;
@@ -2503,7 +2503,7 @@ int freadMain(freadMainArgs _args) {
     }
     stopTeam = false;
 
-    if (extraAllocRows) {
+    if (extraAllocRows && nrowLimit>0) { // no allocating needed for nrows=0
       allocnrow += extraAllocRows;
       if (allocnrow > nrowLimit) allocnrow = nrowLimit;
       if (verbose) DTPRINT(_("  Too few rows allocated. Allocating additional %"PRIu64" rows (now nrows=%"PRIu64") and continue reading from jump %d\n"),
@@ -2512,7 +2512,7 @@ int freadMain(freadMainArgs _args) {
       extraAllocRows = 0;
       goto read;
     }
-    if (restartTeam) {
+    if (restartTeam && nrowLimit>0) { // no restarting needed for nrows=0 since we discard read data anyway
       if (verbose) DTPRINT(_("  Restarting team from jump %d. nSwept==%d quoteRule==%d\n"), jump0, nSwept, quoteRule);
       ASSERT(nSwept>0 || quoteRuleBumpedCh!=NULL, "Internal error: team restart but nSwept==%d and quoteRuleBumpedCh==%p", nSwept, quoteRuleBumpedCh); // # nocov
       goto read;