From: Eric Friedman (ebf_at_[hidden])
Date: 2003-11-30 02:21:26
For some time now, variant_test5 has been failing under the Borland
compilers. It was not until recently that I determined why.
There problem is an alignment issue. I don't know whether the cause is
with the Borland compilers, the code for variant, the code for
alignment_of/type_with_alignment, or something else. Worse, I don't have
enough free time before the 1.31 release to really dive into this issue.
Fortunately I have devised a temporary solution: I have found that using
maximal alignment fixes the problem. (Currently, variant calculates the
minimal needed alignment for its storage so as to minimize variant data
size.) Accordingly, I plan to apply this as a temporary fix when
compiling under Borland.
My question: do I apply this temporary fix under *all* compilers?
My belief is that doing so would be unnecessary, and I know that it
would result in larger variants. On the other hand, if the problem is
*not* Borland-specific, then it would be critically important to have
this fix.
Please help me decide.
Thanks,
Eric