I am having the same problem for the integer workspace. Putting -1 into the size decriptor gives on a two processor parallel 9x9 matrix size test problem on proc 0 a 0 and on proc 1 a 72756096. This does not make sense and consequently this routine produces segfaults, strangely they come from PXERBLA,
+ /lib/x86_64-linux-gnu/libc.so.6(+0x364c0) [0x2adc193444c0]
+ function pxerbla_ (0x1714741)
+ function pdhseqr_ (0x1743FDF)
which does handle the error output for wrong parameters. Looks like a bug to me and I suspect it to be coming from the integer workspace array, whos size I cannot safely determine.
Is there any other routine that can determine the size of the integer array needed?
The workspace query to the real array size works fine with -1 tho, it returns reasonable numbers.
Have a nice day, Jan