Bus Error

[hasindu2008]

When I run the tool f5ls on a fast5 file I get the following error. ... eventdetection/events/size=12393 (mean=92.6346, stdv=2.68832, start=28195744, length=7) basecall(0)/group_list=1D_000 basecall(0)/seq_size=6997 Bus error

Similar behaviour is observed for f5-full or any other tool (such as nanopolish) that uses fast5 library when accessing events in fast5 files. The information about the system which I get the error is as follows. Processor: ARMv7 Processor rev 3 Operating system: Ubuntu 16.04.3 LTS

The output from gdb and backtrace is attached herewith. gdb_out.txt

The bus error seems to have originated from inside the HDF functions. However, I do not think that this is a bug in the HDF library as the h5dump tool provided by HDF output the following without any issues. h5dump.txt

Can you shed some light on this to fix the issue?

[hasindu2008]

After 2 days of hectic debugging, I think I found the issue. The data type for the Events group as given by h5dump is as follows,

         GROUP "BaseCalled_template" {
            DATASET "Events" {
               DATATYPE  H5T_COMPOUND {
                  H5T_IEEE_F64LE "mean";
                  H5T_IEEE_F64LE "start";
                  H5T_IEEE_F64LE "stdv";
                  H5T_IEEE_F64LE "length";
                  H5T_STRING {
                     STRSIZE 5;
                     STRPAD H5T_STR_NULLPAD;
                     CSET H5T_CSET_ASCII;
                     CTYPE H5T_C_S1;
                  } "model_state";
                  H5T_STD_I64LE "move";
                  H5T_IEEE_F32LE "weights";
                  **H5T_IEEE_F32LE "p_model_state**";

For p_model_state the data type is H5T_IEEE_F32LE. However, in fast5.hpp:393 it is declared as double instead of float as below.

struct Basecall_Event
{
    double mean;
    double stdv;
    double start;
    double length;
    **double p_model_state**;
    long long move;

The Bus error dissappeared after I did this change.

[jts]

Hi @hasindu2008,

Thanks for looking into this. I don't have access to an ARM development machine so I have limited ability to help.

This issue may have been caused by changes to ONT's fast5 file format (the internal data types have changed a few times in the past). Can you let me know which dataset caused the problem?

Thanks, Jared

[hasindu2008]

Hi @jts

The dataset that I used was downloaded from (https://github.com/nanopore-wgs-consortium/NA12878/blob/master/Genome.md) under the FAST5 (Signal Level files) section. I am attaching one of those fast5 files here for your convenience. After I changed the data type for p_model_state from double to float, now things are working perfectly.

test.zip