Data.table: Stack imbalance in fread

@HughParsonage, this looks similar to #2457. Perhaps try passing showProgress=FALSE and see if it completes.
@mattdowle could there have been a regression since 2017-11-09?

Running with showProgress=FALSE did indeed return the result (with only the expected warnings).

Thanks for all the detailed info. I doubt there has been a regression since 2017-11-09 but maybe the long verbose=TRUE output is having a similar impact to the ETA output. The file needs a reread which means more output is generated. I fear that @HughParsonage 's report that showProgress=TRUE works for him is spurious and that the problem will happen if it is run 5-10 times with verbose=TRUE.

There aren't any verbose messages printed from within the parallel section (other than progress ETA which is already fixed.) However, there are verbose messages after the first read and before the 2nd reread starts (which is happening for this file). I suppose it's possible that if those prints trigger the 100th CheckUserInterrupt (see #2457), it could cause the 2nd parallel region to fail (odd though). To rule that out anyway, I've just changed all verbose messages to use REprintf rather than Rprintf (same fix as #2457 for the ETA). That's failed because the tests aren't finding the output on stderr -- will fix. Once it's passing, the Windows .zip will be automatically created, and then you can try again please. I'll update here when ready.

Ok, the 2nd attempt is passing checks and Windows.zip is available. @HughParsonage would you mind trying again please? I've added a call to R_FlushConsole() after the messages in verbose mode just before rereading. That flush is only ever needed on Windows. I'm taking a guess that without the flush, the console sometimes updates a tiny bit later when the parallel reread is happening and that is known to cause problems. Please repeat 10 times, always with both verbose=TRUE and showProgress=TRUE. If you see 10 clear runs then we'll say that was it. Otherwise I'll have to think again.

Unfortunately, not fixed:

> library(data.table)
data.table 1.10.5 IN DEVELOPMENT built 2017-11-15 00:36:41 UTC; appveyor
  The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
  Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
  Release notes, videos and slides: http://r-datatable.com
> fread("SA2-by-DJZ-2011.csv", header = FALSE, na.strings = "", verbose = FALSE)
Read 26%. ETA 00:00 Warning: stack imbalance in '$', 20 then 22
Read 52%. ETA 00:00 Warning: stack imbalance in '$', 36 then 35
Warning: stack imbalance in '$', 21 then 22
Read 59%. ETA 00:00 Error in fread("SA2-by-DJZ-2011.csv", header = FALSE, na.strings = "",  : 
  unprotect_ptr: pointer not found
In addition: Warning: stack imbalance in '$', 26 then 28
Warning messages:
1: Warning: stack imbalance in '$', 26 then 27
In fread("SA2-by-DJZ-2011.csv", header = FALSE, na.strings = "",  :
  Starting data input on line 12 <<"Goulburn","110018063",3499,>> with 4 fields and discarding line 11 <<"Main Statistical Area Structu>> before it because it has a different number of fields (3).
2: In fread("SA2-by-DJZ-2011.csv", header = FALSE, na.strings = "",  :
  Found the last consistent line but text exists afterwards. Consider fill=TRUE and/or blank.lines.skip=TRUE. First 200 characters of discarded line: <<"Dataset: 2011 Census of Population and Housing">>
Warning: stack imbalance in 'lapply', 31 then 30
Warning: stack imbalance in '$', 14 then 15

Using verbose=TRUE, showProgress=TRUE even after 10 runs I get no error. Here's the result of the 10th output:

Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:02.094 wall clock time
Thread buffers were grown 0 times (if all 1 threads each grew once, this figure would be 1)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 99%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:01.752
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.004s (  0%) Memory map 0.341GB file
   0.008s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.173s (  4%) Allocation of 22885380 rows x 4 cols (0.469GB)
   3.660s ( 95%) Reading 360 chunks of 0.971MB (63547 rows) using 1 threads
   =    0.009s (  0%) Finding first non-embedded \n after each jump
   +    1.946s ( 51%) Parse to row-major thread buffers
   +    1.098s ( 29%) Transpose
   +    0.608s ( 16%) Waiting
   1.752s ( 46%) Rereading 1 columns due to out-of-sample type exceptions
   3.846s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:02.589 wall clock time
Thread buffers were grown 0 times (if all 1 threads each grew once, this figure would be 1)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 99%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:01.418
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.001s (  0%) Memory map 0.341GB file
   0.003s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.574s ( 14%) Allocation of 22885380 rows x 4 cols (0.469GB)
   3.428s ( 86%) Reading 360 chunks of 0.971MB (63547 rows) using 1 threads
   =    0.010s (  0%) Finding first non-embedded \n after each jump
   +    1.988s ( 50%) Parse to row-major thread buffers
   +    1.137s ( 28%) Transpose
   +    0.292s (  7%) Waiting
   1.418s ( 35%) Rereading 1 columns due to out-of-sample type exceptions
   4.007s        Total
There were 20 warnings (use warnings() to see them)

@HughParsonage Thanks! I'm confused though. You're saying it works fine with verbose=TRUE, showProgress=TRUE which is what we hoped for -- yay! That failed before didn't it? The default for showProgress is TRUE anyway but when you run with the default FALSE for verbose, _then_ it doesn't work and you see the stack imbalance? That's odd that _less_ output makes it fail. Please confirm. If that's the case then maybe I'm barking up the wrong tree. It works fine here for me on Linux so I'm reliant on you testing on Windows. Thanks.
(Also, at the bottom of the 10th run output, it says there were 20 warnings. I assume those are the 2 warnings shown higher up, repeated 10 times. If so, makes sense.)

Hi sorry for the confusion, Matt.

You're right that the original issue no longer results in a crash, namely the following works as expected:

fread("SA2-by-DJZ-2011.csv", header = FALSE, na.strings = "")

To clarify, in the original, when verbose =FALSE (the default) I got a crash. I ran it with verbose = TRUE before filing the issue, and noticed a 'stack imbalance' warning but didn't encounter a crash. With the latest version, I don't get a crash (or indeed any problems) with verbose = FALSE.

The reason I said 'not fixed' was that I noticed the warning messages:

Warning messages:
Warning: stack imbalance in '$', 26 then 27
Warning: stack imbalance in 'lapply', 31 then 30
Warning: stack imbalance in '$', 14 then 15

which looked strange and I thought might indicate a closely-related though not identical problem. Having said that, this morning in Australia I can no longer reproduce the warning messages.

Ok I see. Those warning messages about stack imbalance are essentially errors, yes. We can't skip them. I'm calling that warning about stack imbalance a crash, even though it hasn't actually crashed yet. (It's just a matter of time until it crashes after seeing that warning.)

When you do the 10 runs in a fresh R session with verbose=TRUE, showProgress=TRUE, are any of the 20 warnings about stack imbalance or are all those 20 just the following regular warnings.

1: In fread("SA2-by-DJZ-2011.csv", header = FALSE, na.strings = "",  :
  Starting data input on line 12 <<"Goulburn","110018063",3499,>> with 4 fields and discarding line 11 <<"Main Statistical Area Structu>> before it because it has a different number of fields (3).
2: In fread("SA2-by-DJZ-2011.csv", header = FALSE, na.strings = "",  :
  Found the last consistent line but text exists afterwards. Consider fill=TRUE and/or blank.lines.skip=TRUE. First 200 characters of discarded line: <<"Dataset: 2011 Census of Population and Housing">>

Once a stack imbalance warning has occurred, please start a new fresh R session. We can't trust anything from R after that has occurred even once.

I managed to get a crash when I ran with verbose=TRUE, showProgress=TRUE. Something about a const char with a SEXP. I'm trying to reproduce this from the command line (unfortunately it occurred in RStudio and RStudio closed down before I could read the entire message).

Can't reproduce the crash. Here is the result after rebooting. There was a stack imbalance warning:

> library(data.table)
data.table 1.10.5 IN DEVELOPMENT built 2017-11-15 00:36:41 UTC; appveyor
  The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
  Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
  Release notes, videos and slides: http://r-datatable.com
> for (i in 1:10) fread("SA2-by-DJZ-2011.csv", header = FALSE, na.strings = "", verbose = TRUE, showProgress = TRUE)
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.212 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 31%. ETA 00:00 Warning: stack imbalance in '$', 24 then 23
Read 91%. ETA 00:00 Warning: stack imbalance in '$', 27 then 26
Read 95%. ETA 00:00 Warning: stack imbalance in '$', 28 then 29
Read 98%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.895
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.029s (  1%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.314s ( 15%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.761s ( 84%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.015s (  1%) Finding first non-embedded \n after each jump
   +    0.599s ( 28%) Parse to row-major thread buffers
   +    0.400s ( 19%) Transpose
   +    0.746s ( 35%) Waiting
   0.895s ( 42%) Rereading 1 columns due to out-of-sample type exceptions
   2.107s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.335 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 98%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:01.049
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.006s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.402s ( 17%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.974s ( 83%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.209s (  9%) Parse to row-major thread buffers
   +    0.864s ( 36%) Transpose
   +    0.900s ( 38%) Waiting
   1.049s ( 44%) Rereading 1 columns due to out-of-sample type exceptions
   2.385s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.212 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 96%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.414
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.007s (  0%) sep=',' ncol=4 and header detection
   0.002s (  0%) Column type detection using 10027 sample rows
   0.293s ( 18%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.322s ( 81%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.199s ( 12%) Parse to row-major thread buffers
   +    0.822s ( 51%) Transpose
   +    0.301s ( 19%) Waiting
   0.414s ( 25%) Rereading 1 columns due to out-of-sample type exceptions
   1.626s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.451 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 97%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.409
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.006s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.403s ( 22%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.448s ( 78%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.194s ( 10%) Parse to row-major thread buffers
   +    0.974s ( 52%) Transpose
   +    0.279s ( 15%) Waiting
   0.409s ( 22%) Rereading 1 columns due to out-of-sample type exceptions
   1.860s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.480 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 99%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.412
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.007s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.459s ( 24%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.424s ( 75%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.197s ( 10%) Parse to row-major thread buffers
   +    0.938s ( 50%) Transpose
   +    0.288s ( 15%) Waiting
   0.412s ( 22%) Rereading 1 columns due to out-of-sample type exceptions
   1.892s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.381 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 97%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.401
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.005s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.384s ( 22%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.389s ( 78%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.196s ( 11%) Parse to row-major thread buffers
   +    0.911s ( 51%) Transpose
   +    0.281s ( 16%) Waiting
   0.401s ( 22%) Rereading 1 columns due to out-of-sample type exceptions
   1.781s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.384 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 98%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.480
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.006s (  0%) sep=',' ncol=4 and header detection
   0.002s (  0%) Column type detection using 10027 sample rows
   0.476s ( 26%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.378s ( 74%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.192s ( 10%) Parse to row-major thread buffers
   +    0.833s ( 45%) Transpose
   +    0.352s ( 19%) Waiting
   0.480s ( 26%) Rereading 1 columns due to out-of-sample type exceptions
   1.864s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.374 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 98%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.507
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.006s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.311s ( 17%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.562s ( 83%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.193s ( 10%) Parse to row-major thread buffers
   +    0.988s ( 52%) Transpose
   +    0.381s ( 20%) Waiting
   0.507s ( 27%) Rereading 1 columns due to out-of-sample type exceptions
   1.881s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.318 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 96%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.493
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.006s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.306s ( 17%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.496s ( 83%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.193s ( 11%) Parse to row-major thread buffers
   +    0.935s ( 52%) Transpose
   +    0.367s ( 20%) Waiting
   0.493s ( 27%) Rereading 1 columns due to out-of-sample type exceptions
   1.811s        Total
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapping ... ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \r-only line endings are not allowed because \n is found in the data
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 11051  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 11051
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[12] Finalizing the datatable
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:01.141 wall clock time
Thread buffers were grown 0 times (if all 12 threads each grew once, this figure would be 12)
Final type counts
         0 : drop     
         1 : bool8    
         0 : bool8    
         0 : bool8    
         0 : bool8    
         1 : int32    
         0 : int64    
         0 : float64  
         0 : float64  
         0 : float64  
         2 : string   
Rereading 1 columns due to out-of-sample type exceptions.
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 98%. ETA 00:00 
[12] Finalizing the datatable
Reread 22885380 rows x 1 columns in 00:00.506
Read 22885380 rows. Exactly what was estimated and allocated up front
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.007s (  0%) sep=',' ncol=4 and header detection
   0.001s (  0%) Column type detection using 10027 sample rows
   0.132s (  8%) Allocation of 22885380 rows x 4 cols (0.469GB)
   1.506s ( 91%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.001s (  0%) Finding first non-embedded \n after each jump
   +    0.195s ( 12%) Parse to row-major thread buffers
   +    0.938s ( 57%) Transpose
   +    0.371s ( 23%) Waiting
   0.506s ( 31%) Rereading 1 columns due to out-of-sample type exceptions
   1.647s        Total
Warning: stack imbalance in 'for', 2 then 8
There were 20 warnings (use warnings() to see them)

Oddly that certainty is great. Thanks. That means the flush didn't work and I'll have to find a way to avoid Rprintf after all. It works reliably with verbose=FALSE, showProgress=FALSE I assume (you wrote that near the top of this issue so I'm relying on that.) "Reliably" meaning 10 straight runs with only the two expected warnings and no sight of the stack imbalance warning.
Leave it with me then. Thanks again.

@HughParsonage Ok, please give it a try again with that recent 2nd attempt. It isn't merged into master yet so please be careful to fetch the Windows.zip from the branch here. As before, please provide the full output either way so that I can check it. Thanks!

First attempt of the following resulted in a crash (something about a pointer).

Second attempt (after rebooting) results in a stack imbalance in '$', 16 then 15 warning.

# Assert that `data.table` is not installed:
stopifnot(!requireNamespace("data.table", quietly = TRUE))

install.packages("https://ci.appveyor.com/api/buildjobs/bpsehtwybbbgbyy3/artifacts/data.table_1.10.5.zip",
                 repos = NULL)
# Installing package into ‘C:/Users/hughp/Documents/R/win-library/3.4’
# (as ‘lib’ is unspecified)
# trying URL 'https://ci.appveyor.com/api/buildjobs/bpsehtwybbbgbyy3/artifacts/data.table_1.10.5.zip'
# Content type 'application/octet-stream' length 1557502 bytes (1.5 MB)
# downloaded 1.5 MB

library(data.table)
# data.table 1.10.5 IN DEVELOPMENT built 2017-11-17 01:38:17 UTC; appveyor
# The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
# Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
# Release notes, videos and slides: http://r-datatable.com

setwd("~/ABS-data/inbox/SA2-by-DJZ-2011/")
fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "", header = FALSE)
# Input contains no \n. Taking this to be a filename to open
# [01] Check arguments
# Using 12 threads (omp_get_max_threads()=12, nth=12)
# NAstrings = [<<>>]
# None of the NAstrings look like numbers.
# show progress = 1
# 0/1 column will be read as boolean
# [02] Opening the file
# Opening file SA2-by-DJZ-2011.csv
# File opened, size = 349.4MB (366418725 bytes).
# Memory mapping ... ok
# [03] Detect and skip BOM
# [04] Arrange mmap to be \0 terminated
# \r-only line endings are not allowed because \n is found in the data
# [05] Skipping initial rows if needed
# Positioned on line 1 starting: <<Australian Bureau of Statistic>>
#   [06] Detect separator, quoting rule, and ncolumns
# Detecting sep ...
# sep=','  with 89 lines of 4 fields using quote rule 0
# Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
#   Quote rule picked = 0
# fill=false and the most number of columns found is 4
# [07] Detect column types, good nrow estimate and whether first row is column names
# 'header' changed by user from 'auto' to false
# Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
# Type codes (jump 000)    : 1551  Quote rule 0
# Type codes (jump 100)    : 1A51  Quote rule 0
# =====
#   Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
# Bytes from first data row on line 12 to the end of last row: 366418143
# Line length: mean=16.02 sd=0.21 min=16 max=29
# Estimated number of rows: 366418143 / 16.02 = 22877178
# Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
# =====
#   [08] Assign column names
# [09] Apply user overrides on column types
# After 0 type and 0 drop user overrides : 1A51
# [10] Allocate memory for the datatable
# Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
# [11] Read the data
# jumps=[0..360), chunk_size=1017828, total_size=366418143
# Read 78%. ETA 00:00 Warning: stack imbalance in '$', 16 then 15
# Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:02.677 wall clock time
# [12] Finalizing the datatable
# Type counts:
#   1 : bool8     '1'
# 1 : int32     '5'
# 2 : string    'A'
# =============================
#   0.002s (  0%) Memory map 0.341GB file
# 0.007s (  0%) sep=',' ncol=4 and header detection
# 0.001s (  0%) Column type detection using 10027 sample rows
# 0.297s ( 11%) Allocation of 25164895 rows x 4 cols (0.469GB) of which 22885380 ( 91%) rows used
# 2.369s ( 89%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
# =    0.003s (  0%) Finding first non-embedded \n after each jump
# +    0.273s ( 10%) Parse to row-major thread buffers (grown 0 times)
# +    1.313s ( 49%) Transpose
# +    0.780s ( 29%) Waiting
# 0.893s ( 33%) Rereading 1 columns due to out-of-sample type exceptions
# 2.677s        Total
# Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
# V1        V2      V3 V4
# 1: Goulburn 110018063    3499 NA
# 2:       NA 110018064     812 NA
# 3:       NA 110018065    2158 NA
# 4:       NA 110019999     402 NA
# 5:       NA 110028068      10 NA
# ---                              
#   22885376:       NA 997999799       0 NA
# 22885377:       NA 998999899      64 NA
# 22885378:       NA 994999499      34 NA
# 22885379:       NA 0&&&&&&&&  250796 NA
# 22885380:       NA 0@@@@@@@@ 7305367 NA
# Warning messages:
#   1: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
#                 Starting data input on line 12 <<"Goulburn","110018063",3499,>> with 4 fields and discarding line 11 <<"Main Statistical Area Structu>> before it because it has a different number of fields (3).
#               2: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
#               Found the last consistent line but text exists afterwards. Consider fill=TRUE and/or blank.lines.skip=TRUE. First 200 characters of discarded line: <<"Dataset: 2011 Census of Population and Housing">>

Hi, @mattdowle. There are versions of GCC still in use whose OpenMP is at best 3.1, not 4.0. I ran into that problem in one of my packages on CRAN (Delaporte) where I tried using a SIMD directive (OpenMP 4.0) which compiled with Rtools for Windows (based on 4.9.3) but threw and error on someone's Linux machine still using gcc 4.8.0. Even Windows can only use 4.0 and not 4.5 calls, if I remember correctly. Maybe that is contributing to the issue?

@HughParsonage Thanks for testing so quickly! Ok, I'll keep thinking then!
@aadler That's a good thought -- anything is possible.

@HughParsonage Just to confirm please that the same command with just one change (verbose=FALSE) works fine? i.e. fread("SA2-by-DJZ-2011.csv", verbose = FALSE, na.strings = "", header = FALSE). The progress meter will still display.

Yes, running that command (ten times) returned the expected result (i.e. the data.table with only two warnings because it's badly formatted). No stack imbalance warnings.

Thanks. So it seems to be related to the console output then. Couple more things to try ...

In verbose mode, there are some branches inside the parallel region that call wallclock(). I've short-circuited that to always return 0.0 and avoid the system call, to rule that out. I thought it was thread safe but maybe not. Please try new Windows.zip from the rebuilt branch here.

First attempt:

install.packages("https://ci.appveyor.com/api/buildjobs/o0pn9ttkrbqgqw2k/artifacts/data.table_1.10.5.zip",
                 repos = NULL)
# Installing package into ‘C:/Users/hughp/Documents/R/win-library/3.4’
# (as ‘lib’ is unspecified)
# trying URL 'https://ci.appveyor.com/api/buildjobs/o0pn9ttkrbqgqw2k/artifacts/data.table_1.10.5.zip'
# Content type 'application/octet-stream' length 1556972 bytes (1.5 MB)
# downloaded 1.5 MB
# 
# package ‘data.table’ successfully unpacked and MD5 sums checked

library(data.table)
# data.table 1.10.5 IN DEVELOPMENT built 2017-11-17 03:49:20 UTC; appveyor
# The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
# Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
# Release notes, videos and slides: http://r-datatable.com
fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "", header = FALSE)

Second attempt, I get the following warnings:

Read 22%. ETA 00:00 Error in fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  : 
  unprotect_ptr: pointer not found
In addition: Warning messages:
1: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
  Starting data input on line 12 <<"Goulburn","110018063",3499,>> with 4 fields and discarding line 11 <<"Main Statistical Area Structu>> before it because it has a different number of fields (3).
2: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
 Warning: stack imbalance in '$', 29 then 28
 Found the last consistent line but text exists afterwards. Consider fill=TRUE and/or blank.lines.skip=TRUE. First 200 characters of discarded line: <<"Dataset: 2011 Census of Population and Housing">>
Warning: stack imbalance in 'lapply', 125 then 126
Warning: stack imbalance in 'lapply', 55 then 53
Warning: stack imbalance in 'lapply', 30 then 34
Warning: stack imbalance in '<-', 28 then 31
Warning: stack imbalance in '{', 24 then 27
Warning: stack imbalance in '{', 18 then 21

Just a thought: could this be a problem with RStudio? Running the script from the terminal doesn't seem to reproduce as easily. I'm running from RStudio because it makes copying the console output easier.

When you say it doesn't reproduce _as easily_ outside RStudio, does it reproduce _at all_ ? Even if it only happens inside RStudio, it's still something I'd aim to fix on the data.table side. I'm just asking as another route to confirm it definitely is "just" console output and not some other true stack imbalance in the fread logic.

I'm yet to reproduce outside of RStudio at all, and can reliable reproduce inside it (that is, I can reproduce some warning or crash). I've tried Windows command prompt and git shell (in Windows).

I'm using RStudio Version 1.1.383 on Windows. Would it be helpful to you if I raised this issue with them as well, or would you like me to wait?

Thanks. That's really useful to know that it's just inside RStudio. No need to raise it with them. It just means it is something to do with the output console buffering (or similar). I've gone ahead with a work around and about to push.

I don't see why Windows is not compiling that change :
fread.c:1054:3: warning: too many arguments for format [-Wformat-extra-args]
Works fine here on Linux and on Travis. That is preventing the Windows.zip being created for you to test this workaround. I'll have to sleep on it.
(It's complaining about line 1054 but not the very next line 1055 which is just the same afaics. Must be some difference. %llu a problem with __VA_ARGS__ on Windows -- surely not.)

Ok, finally windows.zip is ready for you to try again please here.

There are several workarounds in this branch now. If it works, I'll start removing the workarounds to establish which one it was. The llu compiler warnings look the most promising as that would give rise to a stack imbalance in verbose output, consistent with the explanation @st-pasha found here. Maybe the Rprintf layer was hiding that from the compiler which it can see now that it's using fprintf directly.

On second attempt (after rebooting)

stopifnot(!requireNamespace("data.table", quietly = TRUE))
install.packages("https://ci.appveyor.com/api/buildjobs/1pi0ae5iuyj9rhj8/artifacts/data.table_1.10.5.zip",
                 repos = NULL)
# Installing package into ‘C:/Users/hughp/Documents/R/win-library/3.4’
# (as ‘lib’ is unspecified)
# trying URL 'https://ci.appveyor.com/api/buildjobs/1pi0ae5iuyj9rhj8/artifacts/data.table_1.10.5.zip'
# Content type 'application/octet-stream' length 1559167 bytes (1.5 MB)
# downloaded 1.5 MB
# 
# package ‘data.table’ successfully unpacked and MD5 sums checked

library(data.table)
# data.table 1.10.5 IN DEVELOPMENT built 2017-11-18 04:58:23 UTC; appveyor
# The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
# Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
# Release notes, videos and slides: http://r-datatable.com

fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "", header = FALSE)
Log file:  C:\Users\hughp\AppData\Local\Temp\RtmpIT9H0D/fread.out 
Input contains no \n. Taking this to be a filename to open
Read 11%. ETA 00:00 Warning: stack imbalance in '$', 19 then 20
Read 28%. ETA 00:00 Warning: stack imbalance in '$', 19 then 20
Read 48%. ETA 00:00 Warning: stack imbalance in '$', 20 then 19
Read 98%. ETA 00:00 [01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapped ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \n has been found in the data so any mixture of line endings is allowed other than \r-only line endings. This is common and ideal.
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 1A51  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 1A51
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:02.822 wall clock time
[12] Finalizing the datatable
  Type counts:
         1 : bool8     '1'
         1 : int32     '5'
         2 : string    'A'
=============================
   0.000s (  0%) Memory map 0.341GB file
   0.001s (  0%) sep=',' ncol=4 and header detection
   0.000s (  0%) Column type detection using 10027 sample rows
   0.291s ( 10%) Allocation of 25164895 rows x 4 cols (0.469GB) of which 22885380 ( 91%) rows used
   2.531s ( 90%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.002s (  0%) Finding first non-embedded \n after each jump
   +    0.282s ( 10%) Parse to row-major thread buffers (grown 0 times)
   +    1.537s ( 54%) Transpose
   +    0.710s ( 25%) Waiting
   0.842s ( 30%) Rereading 1 columns due to out-of-sample type exceptions
   2.822s        Total
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
                V1        V2      V3 V4
       1: Goulburn 110018063    3499 NA
       2:       NA 110018064     812 NA
       3:       NA 110018065    2158 NA
       4:       NA 110019999     402 NA
       5:       NA 110028068      10 NA
      ---                              
22885376:       NA 997999799       0 NA
22885377:       NA 998999899      64 NA
22885378:       NA 994999499      34 NA
22885379:       NA 0&&&&&&&&  250796 NA
22885380:       NA 0@@@@@@@@ 7305367 NA
Warning messages:
1: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
  Starting data input on line 12 <<"Goulburn","110018063",3499,>> with 4 fields and discarding line 11 <<"Main Statistical Area Structu>> before it because it has a different number of fields (3).
2: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
  Found the last consistent line but text exists afterwards. Consider fill=TRUE and/or blank.lines.skip=TRUE. First 200 characters of discarded line: <<"Dataset: 2011 Census of Population and Housing">>

Again, not reproducible outside of RStudio.

Thanks for testing so quickly. Well, that certainly rules a lot out then! Two ideas left. First one pushed and passing. Please try new Windows.zip here. That alloca is on the stack and has to do with na.strings which you are setting as it happens. Definitely in the right area (stack imbalance) and worth trying.

No problem — I’ll be away for next 12 hours or so so can’t test till then.
On Sat, 18 Nov 2017 at 5:20 pm, Matt Dowle notifications@github.com wrote:

Thanks for testing so quickly. Well, that certainly rules a lot out then!
Two ideas left. First one pushed and passing. Please try new Windows.zip
here
https://ci.appveyor.com/project/Rdatatable/data-table/build/1.0.1363/job/fo02vnbu5ebhwy3w/artifacts.
That alloc is allocated on the stack and has to do with na.strings which
you are setting as it happens. Definitely in the right area (stack
imbalance) and worth trying.

—
You are receiving this because you were mentioned.

Reply to this email directly, view it on GitHub
https://github.com/Rdatatable/data.table/issues/2481#issuecomment-345421856,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AHvGDGa5Qnls5eSFBMaQO5s8DElfrpKSks5s3ncqgaJpZM4QcuPc
.

Ok no worries. Thanks! I've pushed the 2nd idea now too. I seem to remember \r causing a problem on Windows in the past but I don't recall a stack imbalance. Anyway, to rule that out, I've removed the \r from the progress meter. The stack imbalance message does seem to be printed where the ETA lines occur. It's feasible the console catches \r and treats it differently so that the last line is replaced. You should now see a new line every time the ETA is updated. Just temporarily to rule that out. New Windows.zip built and passing here.

fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "", header = FALSE)
Log file:  C:\Users\hughp\AppData\Local\Temp\RtmpcVjZ1f/fread.out 
Input contains no \n. Taking this to be a filename to open
Read 5%. ETA 00:00
Read 8%. ETA 00:00
Read 11%. ETA 00:00
Read 15%. ETA 00:00
Read 18%. ETA 00:00
Read 21%. ETA 00:00
Read 25%. ETA 00:00
Read 28%. ETA 00:00
Read 31%. ETA 00:00
Read 35%. ETA 00:00
Read 38%. ETA 00:00
Read 41%. ETA 00:00
Read 45%. ETA 00:00
Read 48%. ETA 00:00
Read 51%. ETA 00:00
Read 55%. ETA 00:00
Warning: stack imbalance in '$', 30 then 31
Warning: stack imbalance in '$', 17 then 16
Read 58%. ETA 00:00
Read 61%. ETA 00:00
Read 65%. ETA 00:00
Read 68%. ETA 00:00
Read 71%. ETA 00:00
Read 75%. ETA 00:00
Read 78%. ETA 00:00
Read 81%. ETA 00:00
Read 85%. ETA 00:00
Read 88%. ETA 00:00
Read 91%. ETA 00:00
Read 95%. ETA 00:00
Read 98%. ETA 00:00
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapped ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \n has been found in the data so any mixture of line endings is allowed other than \r-only line endings. This is common and ideal.
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 1A51  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 1A51
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:02.894 wall clock time
[12] Finalizing the datatable
  Type counts:
         1 : bool8     '1'
         1 : int32     '5'
         2 : string    'A'
=============================
   0.001s (  0%) Memory map 0.341GB file
   0.003s (  0%) sep=',' ncol=4 and header detection
   0.000s (  0%) Column type detection using 10027 sample rows
   0.316s ( 11%) Allocation of 25164895 rows x 4 cols (0.469GB) of which 22885380 ( 91%) rows used
   2.574s ( 89%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.004s (  0%) Finding first non-embedded \n after each jump
   +    0.284s ( 10%) Parse to row-major thread buffers (grown 0 times)
   +    1.450s ( 50%) Transpose
   +    0.837s ( 29%) Waiting
   0.953s ( 33%) Rereading 1 columns due to out-of-sample type exceptions
   2.894s        Total
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
                V1        V2      V3 V4
       1: Goulburn 110018063    3499 NA
       2:       NA 110018064     812 NA
       3:       NA 110018065    2158 NA
       4:       NA 110019999     402 NA
       5:       NA 110028068      10 NA
      ---                              
22885376:       NA 997999799       0 NA
22885377:       NA 998999899      64 NA
22885378:       NA 994999499      34 NA
22885379:       NA 0&&&&&&&&  250796 NA
22885380:       NA 0@@@@@@@@ 7305367 NA
Warning messages:
1: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
  Starting data input on line 12 <<"Goulburn","110018063",3499,>> with 4 fields and discarding line 11 <<"Main Statistical Area Structu>> before it because it has a different number of fields (3).
2: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
  Found the last consistent line but text exists afterwards. Consider fill=TRUE and/or blank.lines.skip=TRUE. First 200 characters of discarded line: <<"Dataset: 2011 Census of Population and Housing">>

FYI: I could not reproduce this stack imbalance error on a different Windows machine with a slightly older version of RStudio.

In that case, looks like it's time to ask RStudio support as you suggested then please. I've had a look through the fread code again and I'm out of ideas on my side. Please tell them the two version numbers of RStudio. It doesn't necessarily mean it is RStudio, it could be a fault on data.table's side that just happens to show up on one version of RStudio. But it is strange that it seems related to the console output and that is something that is different and RStudio-specific. I've searched for "RStudio stack imabalance" but a lot of issues come up about package faults, not RStudio per se. Difficult problem to search for. Let's keep the issue open here and see what they say.

I doubt that last attempt will help, but for completeness, please give it a try here. Perhaps the MinGW compiler which is used on Windows does something strange with those two ints. One of them is constant 0 which perhaps is optimized away and then causes a stack imbalance somehow.

However, that particular stack imbalance message is coming from eval.c:491 in R itself. Some thread must be running that line but I don't think it is fread or data.table. That check_stack_balance() is only called from 5 places in R internals :
in names.c at the end of do_internal()
in objects.c, twice in applyMethod()
in eval.c, twice in eval()
I don't see how any of those can be reached while fread.c is in its parallel section. The only entry point being called is REprintf and I don't see how that could reach check_stack_balance(). All I can think currently is that RStudio has a thread which is doing something in the background which perhaps interacts with console output, perhaps differently on Windows.
Finally, for completeness, it seems using REprintf is the right way to go as base R is using that (rather than Rprintf) in its progress meter in libcurl.c:354 and internet.c:409. It's a shame R's progress bar at C level is not available in R's API (it seems to be implemented twice in R at C level, too).

@mattdowle , would this be helpful? https://github.com/r-lib/progress

@aader Yes -- thanks! Its source contains this comment :
// In R Studio we should print to stdout, because printing a \r
// to stderr is buggy (reported)
But I already removed the \r and the stack imbalance still happens. I wonder where it was reported.

Maybe this?

https://support.rstudio.com/hc/en-us/community/posts/203136268-Carriage-return-does-not-work-on-stderr-

Last build also did not work:

Reported at https://community.rstudio.com/t/stack-imbalance-possibly-in-stderr/3009

Timely question on R-devel: [Rd] Are Rprintf and REprintf thread-safe?

Upshot "Rprintf and REprintf are not thread-safe."

Yoiks!

Thanks all for the links and Hugh for raising the issue with RStudio.

data.table::fwrite() and data.table::fread() are aware Rprintf and REprintf are not thread-safe so for their progress meters, they only call them from the master thread. Not only don't two data.table threads ever call that R entry point at the same time, but only the master thread ever calls it, too, and that's the only R entry point called by any of the threads at any time during the parallel section. However, Rprintf calls R_CheckUserInterrupt every 100 prints. I think that's the part that is likely not safe even from just the master thread. That's the reason for now using REprintf as that doesn't call R_CheckUserInterrupt. R internals use REprintf for progress meters, so switching to REprintf for consistency with core R makes sense; i.e. that choice is nothing to do with stderr vs stdout, per se.

@kevinushey would you mind taking a look at this thread and letting me know anything else I can try? Could it be RStudio related, somehow, perhaps related to a background thread? If RStudio has a background thread then it could be that Rprintf/REprintf could be called from two threads at the same time. But, if that were the case, then we would have seen many more problems before now. So that seems very unlikely. Perhaps RStudio replaces the ptr_* callbacks mentioned in section 8.1.2 of R-exts -- those are related to console output and interaction. However that section starts with "For unix-alikes", so I don't know how Windows comes in. Maybe section 8.1.5 threading issues is relevant too. Both are subsections of section 8: "Linking GUIs and other front-ends to R".

I'm going to be out until early December, so unfortunately I won't have a chance to take a look until then. However, RStudio does run almost everything on the main thread using the R event loop; the only exceptions are for e.g. project-level file indexing and those background threads typically don't touch any R APIs.

RStudio does take over the various ptr_* callbacks for handling console input and output; I can't immediately think of how they might be a cause here but I'll try to take a deeper look when I'm back in.

Ok, please try this one here. Before, it was updating the progress status every 2%. In your case your file is only taking just under 3 seconds so that was a new progress update to the RStudio console every 0.06 seconds. Maybe that was too much for RStudio. So this attempt prints a bar. It doesn't use \r at all. This should be better for reports and log files anyway where the \r could fill up the output.

Since your timing of 3 seconds is quite quick, I've reduced the progress bar to start at 1 second if there's a 1 second ETA from there. Otherwise it wouldn't display at all and work for your file just because it wasn't being displayed. After you've tested, I'll increase to what fwrite has; i.e. start at 2 secs if ETA is 2 secs from there.

Hello, @mattdowle. My last comment in #2503 may be related to this issue too.

Looks good! No warnings (after 5 runs). First run below (note that leading spaces look different in the actual output):

stopifnot(!requireNamespace("data.table", quietly = TRUE))
install.packages("https://ci.appveyor.com/api/buildjobs/1o9s06o31v8i3ljr/artifacts/data.table_1.10.5.zip",
                 repos = NULL)
# Installing package into ‘C:/Users/hughp/Documents/R/win-library/3.4’
# (as ‘lib’ is unspecified)
# trying URL 'https://ci.appveyor.com/api/buildjobs/1o9s06o31v8i3ljr/artifacts/data.table_1.10.5.zip'
# Content type 'application/octet-stream' length 1557423 bytes (1.5 MB)
# downloaded 1.5 MB
# 
# package ‘data.table’ successfully unpacked and MD5 sums checked

library(data.table)
# data.table 1.10.5 IN DEVELOPMENT built 2017-11-30 00:21:00 UTC; appveyor
# The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
# Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
# Release notes, videos and slides: http://r-datatable.com
setwd("~/ABS-data/inbox/SA2-by-DJZ-2011/")
fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "", header = FALSE)
# Input contains no \n. Taking this to be a filename to open
# [01] Check arguments
# Using 12 threads (omp_get_max_threads()=12, nth=12)
# NAstrings = [<<>>]
# None of the NAstrings look like numbers.
# show progress = 1
# 0/1 column will be read as boolean
# [02] Opening the file
# Opening file SA2-by-DJZ-2011.csv
# File opened, size = 349.4MB (366418725 bytes).
# Memory mapped ok
# [03] Detect and skip BOM
# [04] Arrange mmap to be \0 terminated
# \n has been found in the data so any mixture of line endings is allowed other than \r-only line endings. This is common and ideal.
# [05] Skipping initial rows if needed
# Positioned on line 1 starting: <<Australian Bureau of Statistic>>
#   [06] Detect separator, quoting rule, and ncolumns
# Detecting sep ...
# sep=','  with 89 lines of 4 fields using quote rule 0
# Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
#   Quote rule picked = 0
# fill=false and the most number of columns found is 4
# [07] Detect column types, good nrow estimate and whether first row is column names
# 'header' changed by user from 'auto' to false
# Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
# Type codes (jump 000)    : 1551  Quote rule 0
# Type codes (jump 100)    : 1A51  Quote rule 0
# =====
#   Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
# Bytes from first data row on line 12 to the end of last row: 366418143
# Line length: mean=16.02 sd=0.21 min=16 max=29
# Estimated number of rows: 366418143 / 16.02 = 22877178
# Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
# =====
#   [08] Assign column names
# [09] Apply user overrides on column types
# After 0 type and 0 drop user overrides : 1A51
# [10] Allocate memory for the datatable
# Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
# [11] Read the data
# jumps=[0..360), chunk_size=1017828, total_size=366418143
# jumps=[0..360), chunk_size=1017828, total_size=366418143
# |--------------------------------------------------|
#   |==================================================|
#   Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:02.280 wall clock time
# [12] Finalizing the datatable
# Type counts:
#   1 : bool8     '1'
# 1 : int32     '5'
# 2 : string    'A'
# =============================
#   0.005s (  0%) Memory map 0.341GB file
# 0.037s (  2%) sep=',' ncol=4 and header detection
# 0.000s (  0%) Column type detection using 10027 sample rows
# 0.321s ( 14%) Allocation of 25164895 rows x 4 cols (0.469GB) of which 22885380 ( 91%) rows used
# 1.917s ( 84%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
# =    0.011s (  0%) Finding first non-embedded \n after each jump
# +    0.560s ( 25%) Parse to row-major thread buffers (grown 0 times)
# +    0.488s ( 21%) Transpose
# +    0.858s ( 38%) Waiting
# 0.999s ( 44%) Rereading 1 columns due to out-of-sample type exceptions
# 2.280s        Total
# Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
# V1        V2      V3 V4
# 1: Goulburn 110018063    3499 NA
# 2:       NA 110018064     812 NA
# 3:       NA 110018065    2158 NA
# 4:       NA 110019999     402 NA
# 5:       NA 110028068      10 NA
# ---                              
#   22885376:       NA 997999799       0 NA
# 22885377:       NA 998999899      64 NA
# 22885378:       NA 994999499      34 NA
# 22885379:       NA 0&&&&&&&&  250796 NA
# 22885380:       NA 0@@@@@@@@ 7305367 NA
# Warning messages:
#   1: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
#                 Starting data input on line 12 <<"Goulburn","110018063",3499,>> with 4 fields and discarding line 11 <<"Main Statistical Area Structu>> before it because it has a different number of fields (3).
#               2: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
#               Found the last consistent line but text exists afterwards. Consider fill=TRUE and/or blank.lines.skip=TRUE. First 200 characters of discarded line: <<"Dataset: 2011 Census of Population and Housing">>

@HughParsonage Relief! I think that's a win then. I'll tidy up, merge and move on. Huge thanks to you for testing.

@aadler Yes agreed your comment in issue #2503 looks just the same. Could you test latest from dev too please and confirm it's now fixed? Here's hoping the problem with as.IDate you found was indeed caused by the earlier stack imbalance.

No good :(

> library(data.table)
data.table 1.10.5 IN DEVELOPMENT built 2017-11-30 00:21:00 UTC; appveyor
  The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
  Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
  Release notes, videos and slides: http://r-datatable.com
> sessionInfo()
R version 3.4.2 beta (2017-09-17 r73296)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows Server >= 2012 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252  LC_CTYPE=English_United States.1252    LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C                           LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] data.table_1.10.5

loaded via a namespace (and not attached):
[1] compiler_3.4.2 tools_3.4.2   
> DT <- fread('2017-11-22_1999_Performance.csv', header = TRUE, colClasses = CLS, select = SEL, key = c('LoanID', 'Month'), verbose = TRUE)
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 40 threads (omp_get_max_threads()=40, nth=40)
  NAstrings = [<<NA>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file 2017-11-22_1999_Performance.csv
  File opened, size = 6.355GB (6823372783 bytes).
  Memory mapped ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \n has been found in the data so any mixture of line endings is allowed other than \r-only line endings. This is common and ideal.
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Orig_Year,Orig_Qtr,LoanID,Mont>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 100 lines of 37 fields using quote rule 0
  Detected 37 columns on line 1. This line is either column names or first data row. Line starts as: <<Orig_Year,Orig_Qtr,LoanID,Mont>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 37
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to true
  Number of sampling jump points = 101 because (6823372781 bytes from row 1 to eof) / (2 * 12905 jump0size) == 264369
  Type codes (jump 000)    : 51AA7155A15A7111111111111771111177715  Quote rule 0
  Type codes (jump 001)    : 51AA7155A15A711111111111177111117771A  Quote rule 0
  Type codes (jump 008)    : 51AA7555A15A711111111111177111117771A  Quote rule 0
  Type codes (jump 009)    : 51AA7555AA5A71A155555557177111117775A  Quote rule 0
  Type codes (jump 042)    : 55AA7555AA5A71A155555557177111117775A  Quote rule 0
  Type codes (jump 064)    : 55AA7555AA5A71A1A5555557177111117775A  Quote rule 0
  Type codes (jump 100)    : 55AA7555AA5A71A1A5555557177111117775A  Quote rule 0
  =====
  Sampled 10049 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 1 to the end of last row: 6823372781
  Line length: mean=126.15 sd=8.30 min=100 max=359
  Estimated number of rows: 6823372781 / 126.15 = 54088821
  Initial alloc = 62279495 rows (54088821 + 15%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 4 type and 23 drop user overrides : 00AA700000000000000000070775555077750
[10] Allocate memory for the datatable
  Allocating 14 column slots (37 - 23 dropped) with 62279495 rows
[11] Read the data
  jumps=[0..6520), chunk_size=1046529, total_size=6823372422
|--------------------------------------------------|
|=======Warning: stack imbalance in '$', 27 then 26
===Warning: stack imbalance in '$', 26 then 27
================Error in fread("2017-11-22_1999_Performance.csv", header = TRUE, colClasses = CLS,  : 
  unprotect_ptr: pointer not found

@aadler Thanks for that report. I went through freadR and localized the protection. There's a 30% chance that might work since in your case you are overriding types and there were quite a few protects in that part of the code. Please try again using this build.

@aadler If you haven't tried the last build yet, please go straight to trying this one. Also, if it's possible to get a copy of your file to me, I might be able to try myself on Windows RStudio.

:(

> library(data.table)
data.table 1.10.5 IN DEVELOPMENT built 2017-12-01 01:54:04 UTC; appveyor
  The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
  Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
  Release notes, videos and slides: http://r-datatable.com
> sessionInfo()
R version 3.4.2 beta (2017-09-17 r73296)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows Server >= 2012 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252  LC_CTYPE=English_United States.1252    LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C                           LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] data.table_1.10.5

loaded via a namespace (and not attached):
[1] compiler_3.4.2 tools_3.4.2   
> ColCLASS <- c(rep('integer', 2L), 'character', 'Date', 'numeric',
+               rep('integer', 3L), rep('character', 2L),
+               'integer', 'Date', rep('numeric', 2L), 'Date',
+               rep('numeric', 12L), rep('integer', 5),
+               rep('numeric', 3L), 'integer', 'character')
> SELCOL <- c(WHATEVER)
> DT <- fread('LargeFile.csv', header = TRUE, colClasses = ColCLASS, select = SELCOL, key = c('LoanID', 'Month'), verbose = TRUE)
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 40 threads (omp_get_max_threads()=40, nth=40)
  NAstrings = [<<NA>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file LargeFile.csv
  File opened, size = 6.355GB (6823372783 bytes).
  Memory mapped ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \n has been found in the data so any mixture of line endings is allowed other than \r-only line endings. This is common and ideal.
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Orig_Year,Orig_Qtr,LoanID,Mont>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 100 lines of 37 fields using quote rule 0
  Detected 37 columns on line 1. This line is either column names or first data row. Line starts as: <<Orig_Year,Orig_Qtr,LoanID,Mont>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 37
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to true
  Number of sampling jump points = 101 because (6823372781 bytes from row 1 to eof) / (2 * 12905 jump0size) == 264369
  Type codes (jump 000)    : 51AA7155A15A7111111111111771111177715  Quote rule 0
  Type codes (jump 001)    : 51AA7155A15A711111111111177111117771A  Quote rule 0
  Type codes (jump 008)    : 51AA7555A15A711111111111177111117771A  Quote rule 0
  Type codes (jump 009)    : 51AA7555AA5A71A155555557177111117775A  Quote rule 0
  Type codes (jump 042)    : 55AA7555AA5A71A155555557177111117775A  Quote rule 0
  Type codes (jump 064)    : 55AA7555AA5A71A1A5555557177111117775A  Quote rule 0
  Type codes (jump 100)    : 55AA7555AA5A71A1A5555557177111117775A  Quote rule 0
  =====
  Sampled 10049 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 1 to the end of last row: 6823372781
  Line length: mean=126.15 sd=8.30 min=100 max=359
  Estimated number of rows: 6823372781 / 126.15 = 54088821
  Initial alloc = 62279495 rows (54088821 + 15%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 4 type and 23 drop user overrides : 00AA700000000000000000070775555077750
[10] Allocate memory for the datatable
  Allocating 14 column slots (37 - 23 dropped) with 62279495 rows
[11] Read the data
  jumps=[0..6520), chunk_size=1046529, total_size=6823372422
|--------------------------------------------------|
|Error in fread("LargeFile.csv", header = TRUE, colClasses = ColCLASS,  : 
  unprotect_ptr: pointer not found

Thanks to @aadler on email, I can now reproduce. R 3.4.2, latest RStudio 1.1.383 and Windows 10 Pro 10.0.16299 Build 16299.

I'm seeing odd behaviour in RStudio, recorded here :
https://www.youtube.com/watch?v=tl2x2vmZxMU
It seems that RStudio is generating GCs just by typing. Why is that and is there anyway to turn it off? It might be that when fread() is printing its progress bar, RStudio's separate event loop is thinking that the output to the console is the user typing and calls R which gives rise to the GC and trips everything up? Perhaps RStudio users here know, could point me in the right direction, or perhaps @kevinushey is back (you did say early December Kevin, and it's the 1st today : - ) )

I can reproduce the stack imbalance reliably in the RStudio Console. Using the RStudio Terminal tab, I can't reproduce it at all, even with gcinfo(TRUE). Interestingly, GCs do occur as the progress bar is printing and that seems fine, as it is fine on Linux too. Given the behaviour in that video of the RStudio Console, I'm reaching the conclusion that this is an RStudio Console bug. I was unable to copy the text from the RStudio Terminal window (Edit->Copy doesn't work and neither does Ctrl-C) so I took a screenshot of the Terminal tab to show that GC during the progress bar is ok. I'd expect it to be ok because only the master thread is calling REprintf and the other threads are not calling any R API at all.

Works fine in RStudio Terminal :

Note there are GC's while the progress bar is printing the first time and that works fine in RStudio Terminal. The progress bar prints a second time because there's an out-of-sample type exception in this test file which triggers an auto reread for just those columns.

But in RStudio Console there is either stack imbalance or unprotect_ptr: pointer not found :

R version 3.4.2 (2017-09-28) -- "Short Summer"
> gcinfo(TRUE)
[1] FALSE
Garbage collection 22 = 16+3+3 (level 0) ... 
25.5 Mbytes of cons cells used (80%)
6.6 Mbytes of vectors used (66%)
Garbage collection 23 = 16+4+3 (level 1) ... 
24.9 Mbytes of cons cells used (79%)
6.5 Mbytes of vectors used (64%)
Garbage collection 24 = 17+4+3 (level 0) ... 
25.1 Mbytes of cons cells used (79%)
6.5 Mbytes of vectors used (64%)
Garbage collection 25 = 18+4+3 (level 0) ... 
25.1 Mbytes of cons cells used (79%)
6.5 Mbytes of vectors used (64%)
Garbage collection 26 = 19+4+3 (level 0) ... 
25.1 Mbytes of cons cells used (79%)
6.5 Mbytes of vectors used (65%)
Garbage collection 27 = 20+4+3 (level 0) ... 
25.2 Mbytes of cons cells used (80%)
6.5 Mbytes of vectors used (65%)
Garbage collection 28 = 20+5+3 (level 1) ... 
25.1 Mbytes of cons cells used (79%)
6.5 Mbytes of vectors used (65%)
Garbage collection 29 = 21+5+3 (level 0) ... 
25.1 Mbytes of cons cells used (79%)
6.5 Mbytes of vectors used (65%)
Garbage collection 30 = 22+5+3 (level 0) ... 
25.2 Mbytes of cons cells used (80%)
6.5 Mbytes of vectors used (65%)
Garbage collection 31 = 23+5+3 (level 0) ... 
25.2 Mbytes of cons cells used (80%)
6.6 Mbytes of vectors used (66%)
Garbage collection 32 = 24+5+3 (level 0) ... 
25.3 Mbytes of cons cells used (80%)
6.6 Mbytes of vectors used (66%)
Garbage collection 33 = 25+5+3 (level 0) ... 
25.4 Mbytes of cons cells used (80%)
6.7 Mbytes of vectors used (66%)
Garbage collection 34 = 25+5+4 (level 2) ... 
24.6 Mbytes of cons cells used (61%)
6.4 Mbytes of vectors used (50%)
Garbage collection 35 = 26+5+4 (level 0) ... 
25.0 Mbytes of cons cells used (62%)
6.5 Mbytes of vectors used (52%)
> require(data.table)
Loading required package: data.table
Garbage collection 36 = 27+5+4 (level 0) ... 
27.2 Mbytes of cons cells used (68%)
7.1 Mbytes of vectors used (56%)
data.table 1.10.5 IN DEVELOPMENT built 2017-12-02 01:04:34 UTC; appveyor
  The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
  Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
  Release notes, videos and slides: http://r-datatable.com
Garbage collection 37 = 28+5+4 (level 0) ... 
27.7 Mbytes of cons cells used (69%)
7.3 Mbytes of vectors used (58%)
Garbage collection 38 = 29+5+4 (level 0) ... 
28.0 Mbytes of cons cells used (70%)
7.4 Mbytes of vectors used (58%)
Garbage collection 39 = 30+5+4 (level 0) ... 
28.1 Mbytes of cons cells used (70%)
7.4 Mbytes of vectors used (59%)
Garbage collection 40 = 31+5+4 (level 0) ... 
28.2 Mbytes of cons cells used (70%)
7.5 Mbytes of vectors used (59%)
Garbage collection 41 = 32+5+4 (level 0) ... 
28.4 Mbytes of cons cells used (71%)
7.5 Mbytes of vectors used (59%)
> DT = fread("/Users/pasha/Downloads/LargeFile.csv")
Garbage collection 42 = 32+5+5 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
7.1 Mbytes of vectors used (2%)
Garbage collection 43 = 32+5+6 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
244.7 Mbytes of vectors used (42%)
Garbage collection 44 = 32+5+7 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
482.3 Mbytes of vectors used (42%)
Garbage collection 45 = 32+5+8 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
957.4 Mbytes of vectors used (56%)
Garbage collection 46 = 32+5+9 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
1432.6 Mbytes of vectors used (63%)
Garbage collection 47 = 32+5+10 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
2145.3 Mbytes of vectors used (75%)
Garbage collection 48 = 32+5+11 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
2620.4 Mbytes of vectors used (71%)
Garbage collection 49 = 32+5+12 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
3570.8 Mbytes of vectors used (78%)
Garbage collection 50 = 32+5+13 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
4283.5 Mbytes of vectors used (75%)
Garbage collection 51 = 32+5+14 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
5709.0 Mbytes of vectors used (77%)
Garbage collection 52 = 32+5+15 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
7372.0 Mbytes of vectors used (81%)
Garbage collection 53 = 32+5+16 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
8797.5 Mbytes of vectors used (79%)
Garbage collection 54 = 32+5+17 (level 2) ... 
27.4 Mbytes of cons cells used (54%)
10935.7 Mbytes of vectors used (80%)
|--------------------------------------------------|
|=====Error in fread("LargeFile.csv") : 
  unprotect_ptr: pointer not found
> 

showProgress=FALSE resolves it reliably in RStudio Console. To reproduce, it needs to be the very first run in a fresh RStudio Console with showProgress=TRUE (i.e. default). Seems related to whether there's a GC during the progress meter; there is in the first run in a fresh session. It just needs to be a large file so that the progress meter displays. Nothing to do with reread or arguments passed to fread. If the first run in a fresh RStudio Console is with showProgress=FALSE to make it work, that run expands R's heap then a subsequent run in the same session with showProgress=TRUE also then works. But just because there's then no GC during the progress meter due to the first run already expanding the heap.
Why a GC on master thread during the progress meter is ok on Linux and in Windows RStudio Terminal but not in RStudio Console is the outstanding question.

Ok, this fixes it. Problem was on data.table side and not RStudio. Works now for me reliably in RStudio Console on Windows. It was a problem that could occur on Linux and Max too, it's just that the memory patterns weren't triggering it. The other threads did have an entry point to R (when pushing their buffers with string columns) which could happen at the same time as the master thread printing progress using REprintf. That's why it only happened in the first run in a fresh session. In the 2nd run onwards, all the strings in the file had been seen before so the cache lookups were hitting (thread-safe) and not alloc'ing (not thread-safe).

So, @aadler and @HughParsonage, please try this one. 95% chance this works now!

No warning, not sure if you're looking for anything else:

> gcinfo(TRUE)
[1] FALSE
> fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "", header = FALSE)
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 12 threads (omp_get_max_threads()=12, nth=12)
  NAstrings = [<<>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file SA2-by-DJZ-2011.csv
  File opened, size = 349.4MB (366418725 bytes).
  Memory mapped ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \n has been found in the data so any mixture of line endings is allowed other than \r-only line endings. This is common and ideal.
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Australian Bureau of Statistic>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 89 lines of 4 fields using quote rule 0
  Detected 4 columns on line 12. This line is either column names or first data row. Line starts as: <<"Goulburn","110018063",3499,>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 4
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to false
  Number of sampling jump points = 101 because (366418375 bytes from row 1 to eof) / (2 * 1457 jump0size) == 125744
  Type codes (jump 000)    : 1551  Quote rule 0
  Type codes (jump 100)    : 1A51  Quote rule 0
  =====
  Sampled 10027 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 12 to the end of last row: 366418143
  Line length: mean=16.02 sd=0.21 min=16 max=29
  Estimated number of rows: 366418143 / 16.02 = 22877178
  Initial alloc = 25164895 rows (22877178 + 9%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 0 type and 0 drop user overrides : 1A51
[10] Allocate memory for the datatable
  Allocating 4 column slots (4 - 0 dropped) with 25164895 rows
Garbage collection 53 = 36+5+12 (level 2) ... 
30.3 Mbytes of cons cells used (60%)
7.9 Mbytes of vectors used (1%)
[11] Read the data
  jumps=[0..360), chunk_size=1017828, total_size=366418143
|--------------------------------------------------|
|==================================================|
Garbage collection 54 = 37+5+12 (level 0) ... 
30.8 Mbytes of cons cells used (61%)
566.6 Mbytes of vectors used (74%)
Garbage collection 55 = 37+6+12 (level 1) ... 
30.8 Mbytes of cons cells used (61%)
549.2 Mbytes of vectors used (72%)
  jumps=[0..360), chunk_size=1017828, total_size=366418143
|--------------------------------------------------|
|==================================================|
Read 22885380 rows x 4 columns from 349.4MB (366418725 bytes) file in 00:02.626 wall clock time
[12] Finalizing the datatable
  Type counts:
         1 : bool8     '1'
         1 : int32     '5'
         2 : string    'A'
=============================
   0.002s (  0%) Memory map 0.341GB file
   0.005s (  0%) sep=',' ncol=4 and header detection
   0.000s (  0%) Column type detection using 10027 sample rows
   0.469s ( 18%) Allocation of 25164895 rows x 4 cols (0.469GB) of which 22885380 ( 91%) rows used
   2.150s ( 82%) Reading 360 chunks of 0.971MB (63547 rows) using 12 threads
   =    0.103s (  4%) Finding first non-embedded \n after each jump
   +    0.230s (  9%) Parse to row-major thread buffers (grown 0 times)
   +    0.718s ( 27%) Transpose
   +    1.099s ( 42%) Waiting
   0.745s ( 28%) Rereading 1 columns due to out-of-sample type exceptions
   2.626s        Total
Column 1 ("") bumped from 'bool8' to 'string' due to <<"Goulburn">> on row 0
Garbage collection 56 = 37+6+13 (level 2) ... 
31.1 Mbytes of cons cells used (62%)
531.9 Mbytes of vectors used (70%)
Garbage collection 57 = 38+6+13 (level 0) ... 
31.1 Mbytes of cons cells used (62%)
532.0 Mbytes of vectors used (70%)
                V1        V2      V3 V4
       1: Goulburn 110018063    3499 NA
       2:       NA 110018064     812 NA
       3:       NA 110018065    2158 NA
       4:       NA 110019999     402 NA
       5:       NA 110028068      10 NA
      ---                              
22885376:       NA 997999799       0 NA
22885377:       NA 998999899      64 NA
22885378:       NA 994999499      34 NA
22885379:       NA 0&&&&&&&&  250796 NA
22885380:       NA 0@@@@@@@@ 7305367 NA
Warning messages:
1: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
  Starting data input on line 12 <<"Goulburn","110018063",3499,>> with 4 fields and discarding line 11 <<"Main Statistical Area Structu>> before it because it has a different number of fields (3).
2: In fread("SA2-by-DJZ-2011.csv", verbose = TRUE, na.strings = "",  :
  Found the last consistent line but text exists afterwards. Consider fill=TRUE and/or blank.lines.skip=TRUE. First 200 characters of discarded line: <<"Dataset: 2011 Census of Population and Housing">>

Thanks Hugh. Yep that's a clean run, assuming that was in a fresh RStudio console session. No sign of either the stack imbalance or "unprotect_ptr: pointer not found" messages, and the progress meter is running correctly (twice in this case as there's a reread). Now just @aadler to confirm.

SUCCESS.

First run, fresh instance of RStudio.

> library(data.table)
data.table 1.10.5 IN DEVELOPMENT built 2017-12-02 12:05:42 UTC; appveyor
  The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
  Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
  Release notes, videos and slides: http://r-datatable.com
> sessionInfo()
R version 3.4.2 beta (2017-09-17 r73296)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows Server >= 2012 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252  LC_CTYPE=English_United States.1252    LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C                           LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] data.table_1.10.5

loaded via a namespace (and not attached):
[1] compiler_3.4.2 tools_3.4.2   
> DT <- fread('LargeFile.csv', colClasses = colCLASS, select = colSEL, header = TRUE, key = c('LoanID', 'Month'), verbose = TRUE)
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 40 threads (omp_get_max_threads()=40, nth=40)
  NAstrings = [<<NA>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file LargeFile.csv
  File opened, size = 6.355GB (6823372783 bytes).
  Memory mapped ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \n has been found in the data so any mixture of line endings is allowed other than \r-only line endings. This is common and ideal.
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Orig_Year,Orig_Qtr,LoanID,Mont>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 100 lines of 37 fields using quote rule 0
  Detected 37 columns on line 1. This line is either column names or first data row. Line starts as: <<Orig_Year,Orig_Qtr,LoanID,Mont>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 37
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to true
  Number of sampling jump points = 101 because (6823372781 bytes from row 1 to eof) / (2 * 12905 jump0size) == 264369
  Type codes (jump 000)    : 51AA7155A15A7111111111111771111177715  Quote rule 0
  Type codes (jump 001)    : 51AA7155A15A711111111111177111117771A  Quote rule 0
  Type codes (jump 008)    : 51AA7555A15A711111111111177111117771A  Quote rule 0
  Type codes (jump 009)    : 51AA7555AA5A71A155555557177111117775A  Quote rule 0
  Type codes (jump 042)    : 55AA7555AA5A71A155555557177111117775A  Quote rule 0
  Type codes (jump 064)    : 55AA7555AA5A71A1A5555557177111117775A  Quote rule 0
  Type codes (jump 100)    : 55AA7555AA5A71A1A5555557177111117775A  Quote rule 0
  =====
  Sampled 10049 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 1 to the end of last row: 6823372781
  Line length: mean=126.15 sd=8.30 min=100 max=359
  Estimated number of rows: 6823372781 / 126.15 = 54088821
  Initial alloc = 62279495 rows (54088821 + 15%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 4 type and 23 drop user overrides : 00AA700000000000000000070775555077750
[10] Allocate memory for the datatable
  Allocating 14 column slots (37 - 23 dropped) with 62279495 rows
[11] Read the data
  jumps=[0..6520), chunk_size=1046529, total_size=6823372422
|--------------------------------------------------|
|==================================================|
Read 53945186 rows x 14 columns from 6.355GB (6823372783 bytes) file in 00:25.938 wall clock time
[12] Finalizing the datatable
  Type counts:
        23 : drop      '0'
         5 : int32     '5'
         7 : float64   '7'
         2 : string    'A'
=============================
   0.005s (  0%) Memory map 6.355GB file
   0.025s (  0%) sep=',' ncol=37 and header detection
   0.001s (  0%) Column type detection using 10049 sample rows
   4.681s ( 18%) Allocation of 62279495 rows x 37 cols (5.336GB) of which 53945186 ( 87%) rows used
  21.226s ( 82%) Reading 6520 chunks of 0.998MB (8295 rows) using 40 threads
   =    0.485s (  2%) Finding first non-embedded \n after each jump
   +    1.465s (  6%) Parse to row-major thread buffers (grown 0 times)
   +    9.095s ( 35%) Transpose
   +   10.181s ( 39%) Waiting
   0.000s (  0%) Rereading 0 columns due to out-of-sample type exceptions
  25.938s        Total

Closed down RStudio and re-opened it to prevent the string caching from activating and ran it again with gcinfo(TRUE). Added bonus, the conversion to IDate completed (took over 40 seconds, though :) ).

> colCLASS <- c(rep('integer', 2L), 'character', 'Date', 'numeric',
+               rep('integer', 3L), rep('character', 2L),
+               'integer', 'Date', rep('numeric', 2L), 'Date',
+               rep('numeric', 12L), rep('integer', 5),
+               rep('numeric', 3L), 'integer', 'character')
> library(data.table)
data.table 1.10.5 IN DEVELOPMENT built 2017-12-02 12:05:42 UTC; appveyor
  The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
  Documentation: ?data.table, example(data.table) and browseVignettes("data.table")
  Release notes, videos and slides: http://r-datatable.com
> gcinfo(TRUE)
[1] FALSE
> sessionInfo()
R version 3.4.2 beta (2017-09-17 r73296)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows Server >= 2012 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252  LC_CTYPE=English_United States.1252    LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C                           LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] data.table_1.10.5

loaded via a namespace (and not attached):
[1] compiler_3.4.2 tools_3.4.2   
Garbage collection 46 = 36+5+5 (level 0) ... 
38.6 Mbytes of cons cells used (77%)
11.2 Mbytes of vectors used (71%)
Garbage collection 47 = 37+5+5 (level 0) ... 
38.7 Mbytes of cons cells used (77%)
11.2 Mbytes of vectors used (71%)
Garbage collection 48 = 38+5+5 (level 0) ... 
38.8 Mbytes of cons cells used (77%)
11.2 Mbytes of vectors used (71%)
Garbage collection 49 = 39+5+5 (level 0) ... 
39.0 Mbytes of cons cells used (78%)
11.2 Mbytes of vectors used (71%)
Garbage collection 50 = 40+5+5 (level 0) ... 
39.1 Mbytes of cons cells used (78%)
11.3 Mbytes of vectors used (71%)
Garbage collection 51 = 40+6+5 (level 1) ... 
38.8 Mbytes of cons cells used (77%)
11.3 Mbytes of vectors used (71%)
Garbage collection 52 = 41+6+5 (level 0) ... 
38.9 Mbytes of cons cells used (77%)
11.3 Mbytes of vectors used (71%)
Garbage collection 53 = 42+6+5 (level 0) ... 
41.5 Mbytes of cons cells used (83%)
12.2 Mbytes of vectors used (77%)
Garbage collection 54 = 42+7+5 (level 1) ... 
43.4 Mbytes of cons cells used (86%)
12.8 Mbytes of vectors used (81%)
Garbage collection 55 = 42+7+6 (level 2) ... 
44.7 Mbytes of cons cells used (72%)
13.0 Mbytes of vectors used (67%)
Garbage collection 56 = 43+7+6 (level 0) ... 
46.5 Mbytes of cons cells used (74%)
13.6 Mbytes of vectors used (70%)
Garbage collection 57 = 44+7+6 (level 0) ... 
47.0 Mbytes of cons cells used (75%)
13.8 Mbytes of vectors used (71%)
Garbage collection 58 = 45+7+6 (level 0) ... 
47.4 Mbytes of cons cells used (76%)
13.9 Mbytes of vectors used (71%)
Garbage collection 59 = 46+7+6 (level 0) ... 
47.7 Mbytes of cons cells used (76%)
14.2 Mbytes of vectors used (73%)
Garbage collection 60 = 47+7+6 (level 0) ... 
48.0 Mbytes of cons cells used (77%)
14.2 Mbytes of vectors used (73%)
Garbage collection 61 = 48+7+6 (level 0) ... 
48.1 Mbytes of cons cells used (77%)
14.3 Mbytes of vectors used (73%)
> DT <- fread('LargeFile.csv', header = TRUE, colClasses = colCLASS, select = colSEL, key = c('LoanID', 'Month'), verbose = TRUE)
Input contains no \n. Taking this to be a filename to open
[01] Check arguments
  Using 40 threads (omp_get_max_threads()=40, nth=40)
  NAstrings = [<<NA>>]
  None of the NAstrings look like numbers.
  show progress = 1
  0/1 column will be read as boolean
[02] Opening the file
  Opening file LargeFile.csv
  File opened, size = 6.355GB (6823372783 bytes).
  Memory mapped ok
[03] Detect and skip BOM
[04] Arrange mmap to be \0 terminated
  \n has been found in the data so any mixture of line endings is allowed other than \r-only line endings. This is common and ideal.
[05] Skipping initial rows if needed
  Positioned on line 1 starting: <<Orig_Year,Orig_Qtr,LoanID,Mont>>
[06] Detect separator, quoting rule, and ncolumns
  Detecting sep ...
  sep=','  with 100 lines of 37 fields using quote rule 0
  Detected 37 columns on line 1. This line is either column names or first data row. Line starts as: <<Orig_Year,Orig_Qtr,LoanID,Mont>>
  Quote rule picked = 0
  fill=false and the most number of columns found is 37
[07] Detect column types, good nrow estimate and whether first row is column names
  'header' changed by user from 'auto' to true
  Number of sampling jump points = 101 because (6823372781 bytes from row 1 to eof) / (2 * 12905 jump0size) == 264369
  Type codes (jump 000)    : 51AA7155A15A7111111111111771111177715  Quote rule 0
  Type codes (jump 001)    : 51AA7155A15A711111111111177111117771A  Quote rule 0
  Type codes (jump 008)    : 51AA7555A15A711111111111177111117771A  Quote rule 0
  Type codes (jump 009)    : 51AA7555AA5A71A155555557177111117775A  Quote rule 0
  Type codes (jump 042)    : 55AA7555AA5A71A155555557177111117775A  Quote rule 0
  Type codes (jump 064)    : 55AA7555AA5A71A1A5555557177111117775A  Quote rule 0
  Type codes (jump 100)    : 55AA7555AA5A71A1A5555557177111117775A  Quote rule 0
  =====
  Sampled 10049 rows (handled \n inside quoted fields) at 101 jump points
  Bytes from first data row on line 1 to the end of last row: 6823372781
  Line length: mean=126.15 sd=8.30 min=100 max=359
  Estimated number of rows: 6823372781 / 126.15 = 54088821
  Initial alloc = 62279495 rows (54088821 + 15%) using bytes/max(mean-2*sd,min) clamped between [1.1*estn, 2.0*estn]
  =====
[08] Assign column names
[09] Apply user overrides on column types
  After 4 type and 23 drop user overrides : 00AA700000000000000000070775555077750
[10] Allocate memory for the datatable
  Allocating 14 column slots (37 - 23 dropped) with 62279495 rows
Garbage collection 62 = 48+7+7 (level 2) ... 
46.5 Mbytes of cons cells used (60%)
13.6 Mbytes of vectors used (2%)
Garbage collection 63 = 48+7+8 (level 2) ... 
46.5 Mbytes of cons cells used (60%)
488.7 Mbytes of vectors used (42%)
Garbage collection 64 = 48+7+9 (level 2) ... 
46.5 Mbytes of cons cells used (60%)
963.9 Mbytes of vectors used (56%)
Garbage collection 65 = 48+7+10 (level 2) ... 
46.5 Mbytes of cons cells used (60%)
1439.1 Mbytes of vectors used (63%)
Garbage collection 66 = 48+7+11 (level 2) ... 
46.5 Mbytes of cons cells used (60%)
1914.2 Mbytes of vectors used (67%)
Garbage collection 67 = 48+7+12 (level 2) ... 
46.5 Mbytes of cons cells used (60%)
2864.5 Mbytes of vectors used (77%)
Garbage collection 68 = 48+7+13 (level 2) ... 
46.5 Mbytes of cons cells used (60%)
3577.3 Mbytes of vectors used (78%)
Garbage collection 69 = 48+7+14 (level 2) ... 
46.5 Mbytes of cons cells used (60%)
4290.0 Mbytes of vectors used (75%)
[11] Read the data
  jumps=[0..6520), chunk_size=1046529, total_size=6823372422
|--------------------------------------------------|
|============================Garbage collection 70 = 49+7+14 (level 0) ... 
76.5 Mbytes of cons cells used (99%)
5487.5 Mbytes of vectors used (96%)
=Garbage collection 71 = 49+8+14 (level 1) ... 
77.0 Mbytes of cons cells used (100%)
5487.6 Mbytes of vectors used (96%)
Garbage collection 72 = 49+8+15 (level 2) ... 
77.0 Mbytes of cons cells used (81%)
5487.1 Mbytes of vectors used (80%)
==============Garbage collection 73 = 50+8+15 (level 0) ... 
94.3 Mbytes of cons cells used (100%)
5494.0 Mbytes of vectors used (80%)
Garbage collection 74 = 50+9+15 (level 1) ... 
94.5 Mbytes of cons cells used (100%)
5494.1 Mbytes of vectors used (80%)
Garbage collection 75 = 50+9+16 (level 2) ... 
94.5 Mbytes of cons cells used (82%)
5493.1 Mbytes of vectors used (67%)
=======|
Read 53945186 rows x 14 columns from 6.355GB (6823372783 bytes) file in 00:24.772 wall clock time
[12] Finalizing the datatable
  Type counts:
        23 : drop      '0'
         5 : int32     '5'
         7 : float64   '7'
         2 : string    'A'
=============================
   0.005s (  0%) Memory map 6.355GB file
   0.018s (  0%) sep=',' ncol=37 and header detection
   0.000s (  0%) Column type detection using 10049 sample rows
   5.496s ( 22%) Allocation of 62279495 rows x 37 cols (5.336GB) of which 53945186 ( 87%) rows used
  19.253s ( 78%) Reading 6520 chunks of 0.998MB (8295 rows) using 40 threads
   =    0.433s (  2%) Finding first non-embedded \n after each jump
   +    1.482s (  6%) Parse to row-major thread buffers (grown 0 times)
   +    9.515s ( 38%) Transpose
   +    7.822s ( 32%) Waiting
   0.000s (  0%) Rereading 0 columns due to out-of-sample type exceptions
  24.772s        Total
Garbage collection 76 = 51+9+16 (level 0) ... 
105.3 Mbytes of cons cells used (91%)
5500.3 Mbytes of vectors used (67%)
Garbage collection 77 = 51+10+16 (level 1) ... 
105.4 Mbytes of cons cells used (91%)
5500.2 Mbytes of vectors used (67%)
> DT[, Month := as.IDate(Month, format = "%Y-%m-%d")]
Garbage collection 78 = 51+10+17 (level 2) ... 
107.5 Mbytes of cons cells used (76%)
8174.1 Mbytes of vectors used (81%)
Garbage collection 79 = 51+11+17 (level 1) ... 
107.5 Mbytes of cons cells used (76%)
5910.4 Mbytes of vectors used (59%)
> gcinfo(FALSE)
[1] TRUE

awesome! :tada: great work to everyone involved, especially @mattdowle who must be short a few hairs by now with this :)

It seems like my strategy of 'stay on vacation until the problem is fixed' seems to have worked out here :-)

Is there anything else that I should try to take a look at or is this issue considered resolved?

Thanks @aadler and @HughParsonage! Relief.
@kevinushey Ha ha. Yes it was data.table side and now resolved (PR #2488). Thanks.