Opened 6 years ago

Closed 6 years ago

Last modified 6 years ago

#3690 closed defect (fixed)

r.stats.quantile crashing with data linked through r.external

Reported by: neteler Owned by: grass-dev@…
Priority: normal Milestone: 7.4.3
Component: Raster Version: svn-releasebranch74
Keywords: r.external, r.stats.quantile Cc: AnikaBettge
CPU: Unspecified Platform: Unspecified

Description

We discovered that r.stats.quantile crashes with data linked through r.external (when importing the same data sets with r.in.gdal it works fine).

It seems that a corruption occurs with GDAL?

GRASS 7.4.3svn (utm30N):  > r.external input="covermap.tif" band=1 output=tmp15409185587112 --overwrite -o
Reading band 1 of 1...
r.external komplett. Link to raster map <tmp15409185587112> created.

GRASS 7.4.3svn (utm30N):  > r.external input="basemap.tif" band=1 output=tmp15409185587113 --overwrite -oÜ
Reading band 1 of 1...
r.external komplett. Link to raster map <tmp15409185587113> created.

GRASS 7.4.3svn (utm30N):  > r.stats.quantile -p base="tmp15409185587112" cover="tmp15409185587113" bins="2000"
Computing histograms
 100%
Speicherzugriffsfehler (Speicherabzug geschrieben)

... i.e. Core dump.

Debugging:

GRASS 7.4.3svn (utm30N):  > gdb r.stats.quantile 
GNU gdb (Ubuntu 7.11.1-0ubuntu1~16.5) 7.11.1
...
(gdb) r -p base="tmp15409185587112" cover="tmp15409185587113" bins="2000"
Starting program: /home/user/src/grass-7.4.svn/dist.x86_64-pc-linux-gnu/bin/r.stats.quantile -p base="tmp15409185587112" cover="tmp15409185587113" bins="2000"
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Computing histograms
 100%
  27%
Program received signal SIGSEGV, Segmentation fault.
0x0000000000401fa0 in get_slot_counts (basefile=0, coverfile=1) at main.c:161
161		    bc->slots[i]++;

 
(gdb) bt
#0  0x0000000000401fa0 in get_slot_counts (basefile=0, coverfile=1) at main.c:161
#1  0x0000000000403c89 in main (argc=5, argv=0x7fffffffd878) at main.c:642


(gdb) bt full
#0  0x0000000000401fa0 in get_slot_counts (basefile=0, coverfile=1) at main.c:161
        basebuf = 0x6c5760
        coverbuf = 0x6c6eb0
        bc = 0x6f3ab0
        row = 467
        col = 1305
        i = 0
        allnull = 0
#1  0x0000000000403c89 in main (argc=5, argv=0x7fffffffd878) at main.c:642
        module = 0x7ffff7ba0ea8 <state+40>
        opt = {quant = 0x640ec0, perc = 0x641000, slots = 0x641140, basemap = 0x7ffff7ba0f08 <state+136>, covermap = 0x640d80, output = 0x641280, file = 0x641440, fs = 0x641640}
        flag = {r = 0x7ffff7ba0ed8 <state+88>, p = 0x6426a0, t = 0x6426d0}
        basemap = 0x61edb0 "tmp15409185587112"
        covermap = 0x61ee80 "tmp15409185587113"
        outputs = 0x0
        fs = 0x7fffffffd72f "@"
        reclass = 0
        print = 1
        cover_fd = 1
        base_fd = 0
        range = {min = 1, max = 1061, first_time = 0}
        fprange = {min = 7.0767250061034996, max = 9.9984693527221999, first_time = 0}
        i = 1
(gdb) 

Looking at memory allocation and G_free() usage, I see that not all is freed:

cd r.stats.quantile/
grep 'G_calloc\|G_free' main.c  | grep bc
	bc->slots = G_calloc(bc->num_slots, sizeof(unsigned int));
	bc->bins = G_calloc(num_quants, sizeof(struct bin));
	bc->slot_bins = G_calloc(bc->num_slots, sizeof(unsigned char));
	G_free(bc->slots);
	bc->values = G_calloc(num_values, sizeof(DCELL));
	G_free(bc->slot_bins);

but I am not sure if that is related.

We can make the two GeoTIFF files offlist available.

Change History (5)

in reply to:  description ; comment:1 by mmetz, 6 years ago

Replying to neteler:

We discovered that r.stats.quantile crashes with data linked through r.external (when importing the same data sets with r.in.gdal it works fine).

It seems that a corruption occurs with GDAL?

GRASS 7.4.3svn (utm30N):  > r.external input="covermap.tif" band=1 output=tmp15409185587112 --overwrite -o
Reading band 1 of 1...
r.external komplett. Link to raster map <tmp15409185587112> created.

GRASS 7.4.3svn (utm30N):  > r.external input="basemap.tif" band=1 output=tmp15409185587113 --overwrite -oÜ
Reading band 1 of 1...
r.external komplett. Link to raster map <tmp15409185587113> created.

GRASS 7.4.3svn (utm30N):  > r.stats.quantile -p base="tmp15409185587112" cover="tmp15409185587113" bins="2000"
Computing histograms
 100%
Speicherzugriffsfehler (Speicherabzug geschrieben)

... i.e. Core dump.

The above commands do not work for me because basemap and covermap are swapped and the base map must be CELL type.

These commands

r.external input=covermap.tif band=1 output=covermap_ext
r.external input=basemap.tif band=1 output=basemap_ext
r.stats.quantile -p base=basemap_ext cover=covermap_ext bins=2000

work for me without segfault and valgrind does not find any memory violations.

I am using GDAL 2.3.2.

in reply to:  1 ; comment:2 by neteler, 6 years ago

Replying to mmetz:

The above commands do not work for me because basemap and covermap are swapped

ops, maybe I flipped that while anonymising the data using the output from the QGIS Processing log file.

and the base map must be CELL type.

Would it be possible to add a test for that?

These commands

r.external input=covermap.tif band=1 output=covermap_ext
r.external input=basemap.tif band=1 output=basemap_ext
r.stats.quantile -p base=basemap_ext cover=covermap_ext bins=2000

work for me without segfault and valgrind does not find any memory violations.

I am using GDAL 2.3.2.

The machine causing problems is Linux Mint (Ubuntu). The versions I don't have right now.

in reply to:  2 comment:3 by mmetz, 6 years ago

Replying to neteler:

Replying to mmetz:

The above commands do not work for me because basemap and covermap are swapped

ops, maybe I flipped that while anonymising the data using the output from the QGIS Processing log file.

and the base map must be CELL type.

Would it be possible to add a test for that?

This test exists already:

ERROR: The base map must be an integer (CELL) map

I am using GDAL 2.3.2.

The machine causing problems is Linux Mint (Ubuntu). The versions I don't have right now.

Can you please test again on this machine, directly in GRASS, and ideally with both GRASS 7.4 and 7.6?

comment:4 by mmetz, 6 years ago

Resolution: fixed
Status: newclosed

In 73639:

r.external: do not rely on statistics stored in metadata, they can be approximations (fixes #3690)

in reply to:  4 comment:5 by mmetz, 6 years ago

Replying to mmetz:

In 73639:

r.external: do not rely on statistics stored in metadata, they can be approximations (fixes #3690)

These approximate statistics stored in metadata are a nightmare! The stored maximum value in basemap is too low:

gdalinfo -mm basemap.tif
  Min=1.000 Max=1061.000   Computed Min/Max=1.000,1063.000

However, metadata say

TIFFTAG_SOFTWARE=GRASS GIS 7.4.3svn with GDAL 2.2.2

in which case min and max should not be set at all. That means some other software has edited the metadata in basemap.tif, inserting approximate and thus wrong values.

Note: See TracTickets for help on using tickets.