[Debian-med-packaging] Bug#800469: htslib FTBFS on armhf, bus error.

[peter green]

Package: htslib
Severify: important
Tags: patch

htslib is failing to build in Debian armhf and raspbian.

I reproduced the bug locally in a raspbian stretch environment and got 
the following backtrace.

root at odroidu2:/htslib-1.2.1# gdb --args test/sam test/ce.fa
GNU gdb (Raspbian 7.10-1) 7.10
Copyright (C) 2015 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later 
<http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "arm-linux-gnueabihf".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
<http://www.gnu.org/software/gdb/documentation/>.
For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from test/sam...done.
(gdb) run
Starting program: /htslib-1.2.1/test/sam test/ce.fa
[Thread debugging using libthread_db enabled]
Using host libthread_db library 
"/lib/arm-linux-gnueabihf/libthread_db.so.1".

Program received signal SIGBUS, Bus error.
0x00025640 in bam_aux2f (
     s=0xa967e "\333\017I at XddiW\024\213\n\277\005 at XZZHello, world!",
     s at entry=0xa967d "f\333\017I at XddiW\024\213\n\277\005 at XZZHello, world!")
     at sam.c:1181
1181        else if (type == 'f') return *(float*)s;
(gdb)

This is an alignment issue. VFP requires floating point loads and stores 
to be naturally aligned but s is only aligned to 2 bytes whiile a float 
is 4 bytes. Looking further at the code reveals the reason for the 
unalignment. It seems that the code is reading from a data format where 
values are prefixed with single bytes practically gauranteeing that many 
of the accesses will be unaligned.

In portable code unaligned accesses should be avoided. Sometimes they 
will give the right result, sometimes they will give bus errors, 
sometimes they will silently give wrong answers. To a large extent the 
behaviour is driven by architecture (and sometimes version of the 
architecture) but it can also be driven by what instructions the 
compiler choses to use. Even if it does the right thing it may do so 
very slowly. x86 tends to be the most forgiving architecture when it 
comes to unaligned accesses.

The attatched patch replaces a number of unaligned accesses in sam.c 
with memcpy calls. It resulted in a successful build on raspbian 
stretch. I did not include any conditional logic but it would be easy to 
add conditional logic to only use memcpy on non-x86 targets if that was 
considered desirable (I do not if memcpy is faster or slower than 
letting the x86 cpu fixup the unaligned accesses in hardware and if-so 
whether the difference is likely to be significant).


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: htslib.debdiff
URL: <http://lists.alioth.debian.org/pipermail/debian-med-packaging/attachments/20150929/2329dd77/attachment.ksh>