t_coffee segmentation fault in biopython test cases

[emollier]

Greetings,

In the context of the debian project, we noticed that t_coffee version 13.45.0.4846264 is causing a test failure in biopython 1.79. It turns out that the following command:

$ t_coffee \
        -output clustalw_aln \
        -infile Fasta/fa01 \
        -outfile Fasta/tc_out.aln \
        -type protein \
        -outorder input \
        -gapopen -2 \
        -gapext -5

is causing a segmentation fault when run with the following ./Fasta/fa01 input file:

>AK1H_ECOLI/1-378
CPDSINAALICRGEKMSIAIMAGVLEARGH-N--VTVIDPVEKLLAVG-HYLESTVDIAE
STRRIAASRIP------A-DHMVLMAGFTAGN-EKGELVVLGRNGSDYSAAVLAACLRAD
CCEIWTDVNGVYTCDP-------------RQVPDARLLKSMSYQEAMELSY--FGAKVLH
PRTITPIAQFQIPCLIKNTGNPQAPGTL-IG--ASRDEDELP----VKGISNLN------
NMAMFSVSGP-GMKGMVGMAARVFAAMS-------RARISVVLITQSSSEYSISFCVPQS
DCVRAERAMLEEFY-----LELKEGLLEPLAVAERLAIISV-VGDGLRTLRGISAKF---
---FAALARANINIVAIA
>AKH_HAEIN/1-382
-----------------VEDAVKATIDCRGEKLSIAMMKAWFEARGY-S--VHIVDPVKQ
LLAKG-GYLESSVEIEESTKRVDAANIA--K-DKVVLMAGF---TAGNEKGELVLLGRNG
SDYSAAC-----------------LAACLGASVCEIWTDVDGVYTCDP--RLVPDARLLP
TLSYREAMELSYFGAKVIHPRTIGPLLPQNIPCVIKNTGNPSAPGSI-ID--GNVKSESL
Q----VKGITNLDNLAMFNVSGPGMQGM---VGMASRVFSAMSGAGISVILITQSSSEYS
---ISFCVPVKSAEVAKTVLETEFA-----NELNEHQLEPIEVIKDLSIISV-VGDGMKQ
AKGIAARF------FSALAQAN

Here is a backtrace of the affected portion; the intercepted sequence pointer S is NULL, which is probably the root of the segmentation fault:

#0  0x0000555555675e73 in seq2unique_name_seq (S=0x0) at util_lib/aln_convertion_util.c:2091
#1  0x000055555574d656 in read_seq_in_n_list (fname=0x55555598cad0, n=1, type=0x55555593dfb0 "", SeqMode=0x555555d45fb0 "ANY") at util_lib/util_constraints_list.c:2533
#2  0x0000555555659491 in batch_main (argc=15, argv=0x555555944230) at t_coffee_lib/t_coffee.c:4720
#3  0x000055555564f9a0 in main (argc=15, argv=0x7fffffffe128) at t_coffee_lib/t_coffee.c:179

You can lookup Debian Bug#1022570 for some further details, but I tried to gather the most relevant information here.

Have a nice day, :)
Étienne.

[nileshpatra]

/cc: @JoseEspinosa @cnotred @athbaltzis

[emollier]

I tried to have a fresh look at this issue, and the bug seems to occur earlier in the code, before util_lib/util_constraints_list.c:2533, than what the trace reports. Sequence and alignments were already NULL, while I guess they were suppose to hold interesting information at this point:

(gdb) print S1
$6 = (Sequence *) 0x0
(gdb) print A
$7 = (Alignment *) 0x0
(gdb) print lname
$8 = 0x555555e5e2a1 "Fasta/fa01"

So I suppose something went wrong within main_read_aln perhaps.

Hope this helps; that issue is currently pulling a lot of packages out of the Debian 12 release in preparation. I welcome comments, as I'm not competent to decide whether this issue is grave enough to mandate a rollback of the t_coffee version, or if the test is about a fragile corner case normally not encountered in practice.

Thank you,
Étienne.