Fixme, this code is simply broken,
nobody seems to use it.
One would need to change the socket protocol (message length)
in agreement with the users of the protocol.
For the time beeing,
just suffocate the warnings in a way which does not change
the broken behaviour of this code.
identified and reported by Marcel Hendrix,
expressions of this kind could trigger a segmentation violation.
PTdifferentiate() roughly evaluates to
ternary_fcn(ge0(0-expr), 0, PTdifferentate(expr))
and mkb() optimizes
0 - expr --> unary_minus(expr)
IFeval() invokes PTeval() for the derivative too,
PTeval() looks at the incorrect tree->funcnum
and tries to PTeval for a second argument which is not there,
(unary_minus does not have a second argument)
causing a segmentation fault.
in autogen.sh:
libtoolize generates m4/*.m4 files
autoheader generates config.h.in
but only when needed
Thus a second autogen.sh invocation generated new .m4 files
but left config.h.in untouched with old timestamp
The makefiles did recognice this situation and tried to regenerate
config.h.in and config.h
But the subdir src/include was not the first one in src/Makefile.am
and thus this update did occure too late,
causing further rebuilds in the next make invocation.
Avoid this with --force arguments to the autotools,
and put src/include into the first position.
and drop
'-fno-omit-frame-pointer'. This entry compensates for a compiler bug of actual mingw gcc 4.6.1.
which was introduced in commit
commit 5b1be4d32d
Author: h_vogt <h_vogt>
Date: Sat Jan 28 09:59:27 2012 +0000
prepare for release 24
rlar
h_vogt
mhx