[Debian-med-packaging] Bug#969608: makeblastdb 2.10.x on 32-bit architectures

[Étienne Mollier]

Hi Aaron,

Aaron M. Ucko, on 2020-09-13 22:44:00 -0400:
> Étienne Mollier <etienne.mollier at mailoo.org> writes:
> 
> > regression looks still there in version 2.10.0-3, at least on
> > arm64 architecture.
> 
> How can I reproduce it?

I worked within an arm64 schroot built with sbuild-createchroot,
on a native amd64 machine.  The reproducer is the test from
Biopython stored in Tests/test_NCBI_BLAST_tools.py, which boils
down to running the command:

	makeblastdb \
	        -in NC_005816.faa \
	        -dbtype prot \
	        -hash_index \
	        -max_file_sz 20MB \
	        -parse_seqids \
	        -taxid 10

where NC_005816.faa can be obtained from the Biopython test
suite on Salsa:

	https://salsa.debian.org/med-team/python-biopython/-/raw/master/Tests/GenBank/NC_005816.faa

When I run it within that arm64 environment, I get the following
output, and the error code 255 sent back to the shell:

	Building a new DB, current time: 09/14/2020 18:35:05
	New DB name:   /tmp/NC_005816.faa
	New DB title:  NC_005816.faa
	Sequence type: Protein
	Keep MBits: T
	Maximum file size: 20000000B
	
	No volumes were created.
	
	Error: mdb_env_open: Cannot allocate memory

while specifying "-blastdb_version 4" in additions restores the
former behavior:

	Building a new DB, current time: 09/14/2020 18:44:51
	New DB name:   /tmp/NC_005816.faa
	New DB title:  NC_005816.faa
	Sequence type: Protein
	Deleted existing Protein BLAST database named /tmp/NC_005816.faa
	Keep MBits: T
	Maximum file size: 20000000B
	Adding sequences from FASTA; added 10 sequences in 0.0821829 seconds.

I haven't straced the execution of the program to check the
amount of memory in map, so maybe the issue is different but
with similar symptoms as previously?

>                         AFAICT python-biopython's autopkgtest boils
> down to the upstream test suite, which runs fine as part of a build of
> revision 15084ca (immediately preceding your patch to drop back to
> version 4 on selected architectures) on the porterbox amdahl.  Perhaps
> system limits come into play in some other environments, but then I
> expect they would have done the same on 2.10.0-1.  Also, the telltale
> warning
> 
>   /<<PKGBUILDDIR>>/c++/include/objtools/blast/seqdb_writer/writedb_lmdb.hpp:52:59: warning: integer overflow in expression of type ‘int’ results in ‘-647710720’ [-Woverflow]
> 
> occurs only in the logs for 2.10.0-2.

Ouch, looks good to know, thank you for the warning.

Kind Regards,
-- 
Étienne Mollier <etienne.mollier at mailoo.org>
Old rsa/3072: 5ab1 4edf 63bb ccff 8b54  2fa9 59da 56fe fff3 882d
New rsa/4096: 8f91 b227 c7d6 f2b1 948c  8236 793c f67e 8f0d 11da
Sent from /dev/pts/2, please excuse my verbosity.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://alioth-lists.debian.net/pipermail/debian-med-packaging/attachments/20200914/31892484/attachment.sig>