Kneaddata 0.10 produces unordered pairs of reads

Infrastructure and utilities KneadData
1

I run kneaddata this way:

kneaddata --remove-intermediate-output --threads 16 --input ERR1855535_1.fastq.gz --input ERR1855535_2.fastq.gz \
--output kneaddata_bowtie  --reference-db human_genome_index --reference-db Sscrofa11.1_Phix 
--trimmomatic  /env/share/trimmomatic  --max-memory  10G  --bowtie2-options "--very-sensitive --dovetail"

The process ends without errors, but when I run metawrap binning, bwa have problems with the name of the paired reads and fails:

[mem_sam_pe] paired reads have different names: "ERR1855535.63NS500633:37:H3VL5BGXY:1:11101:13061:1082", "ERR1855535.83NS500633:37:H3VL5BGXY:1:11101:10337:1103"

When I looked to these reads, I got :

ERR1855535_paired_2.tmp.fastq                                                                                                                                       
385:@ERR1855535.63NS500633:37:H3VL5BGXY:1:11101:13061:1082/2                            
                                                                                                       
ERR1855535_paired_1.tmp.fastq          
129:@ERR1855535.63NS500633:37:H3VL5BGXY:1:11101:13061:1082/1

(The names have tmp because I moved them before “repair” them)

When I run bbmap’s repair.sh on them I got:

ERR1855535_paired_1.fastq                                                                                                  
329:@ERR1855535.63NS500633:37:H3VL5BGXY:1:11101:13061:1082/1                                        
                                                                                                                                                                                                              
ERR1855535_paired_2.fastq                                                                                                                                           
329:@ERR1855535.63NS500633:37:H3VL5BGXY:1:11101:13061:1082/2

So repair.sh produces paired fastq files that have these reads in the same line and aligning them with bwa doesn’t fail.