problem building a GCJ-friendly GNU Crypto

[Tom Tromey]

>>>>> "Roger" == Roger Sayle <roger@eyesopen.com> writes:

Roger> I'm currently working on a patch to add an -fevaluation-order command
Roger> line option to GCC, that sets a flag_evaluation_order variable that
Roger> can be checked during constant folding to preserve left-to-right
Roger> evaluation order of binary expressions.  This can then be enabled by
Roger> default  by the java front-end, which should help make libbackend.a
Roger> more java friendly.

What about just encoding the required sequencing information in the
trees somehow?  Actually, I thought we did that already.

Last time this came up we also discussed having new operators
reflecting whether or not a given operation was required to be
strictfp.

Anyway, if you're interested in this area, there is a lot of
documentation on what is required from a java compiler here, and there
are also a fair number of tests in jacks.  ISTR that we fail a pretty
large fraction of the tests in this area, though it has been a while
since I looked into this.

Tom