Re: PR java/16927 [Was Re: PATCH for better assertion control.]

[Andrew Haley <aph at redhat dot com>]

Tom Tromey writes:
 > Andrew> Indeed.  However, I'm pretty sure that patch_string does do the right
 > Andrew> thing, and that calling it at this point solves the problem.  But the
 > Andrew> whole approach of patching tree nodes is fraught with problems.
 > 
 > In the particular case of assert, I see the problem as being that the
 > front end is generating trees for both analysis and code generation,
 > so we wind up making "tricky" decisions like this in order to get the
 > same structure past both of those passes.  It would probably be
 > cleaner to just introduce a new ASSERT_STMT tree for analysis, then
 > lower that to something else (perhaps a nop if we aren't generating
 > asserts) after analysis and before code generation.

Perhaps, but the analysis proceeds after the trees have been patched.
I suppose we could leave the lowering of ASSERT_STMT until
gimplification...

Andrew.