Re: Java failures in packs on hurd-i386... [Bug#753791: simgrid: FTBFS on hurd-i386]

[Samuel Thibault <sthibault@debian.org>]

Hello,

I've dug a bit more the java failures we've been encountering for some
time.  The particular case of simply running gjdoc was helpful: an
object gets allocated, then other happens, then the object gets used,
but its method table is completely nuts.  It happens that the class
field of the object was modified (?!). It happens that this is a side
effect of glibc making a symbol resolution (??!!). That seems to be due
to the fact that boehm-gc uses sbrk to make its allocation, while glibc
also uses it for its own allocator. No wonder they step on each other.
Some time ago, boehm-gc was made to use mmap on kfreebsd too, not only
Linux. I've just started a gcc build with using mmap on hurd too, I
guess that will just fix all our issues...  I'll then commit that and
push upstream.

Samuel