Incorrect TLEN sign for reverse complement paired reads?

I'm working with results generated by bowtie2-2.2.3 on PE Illumina data, with SAM output (converted to BAM using samtools), and extracting the template length (TLEN). The SAM spec states that TLEN should have a plus sign for the leftmost read, and a minus sign for the rightmost read in a mapped pair. When the paired reads are reverse complements of each other (i.e. TLEN is shorter than the read length so both sequence reads read through the template and both reads map to the same location) bowtie assigns a minus sign to the TLEN for both reads. The absolute value of TLEN in the output is correct.

I realize this may have been a conscious design decision, or an unanticipated edge case - to me it would make more sense if in these cases R1 was given a plus sign, and R2 given a minus sign. It would also simplify parsing of TLEN from bowtie SAM output.