[tree-ssa] -fno-merge-constants vs. string literals

[Frank Ch. Eigler]

rth wrote:

> > The side-effect I'm interested in is related to -fmudflap, where
> > I'd like to force -fno-merge-constants for similar reasons.
> 
> Why?  Seems like you'd want the constants to still be merged
> regardless of bounds checking.

I believe keeping source-level string literals apart will improve the
error-checking potential of the mudflap runtime.  Consider for example
erroneous negative indexes on a merged string.  Actually, this
suggests that -fmudflap would benefit from some extra padding in the
varasm.c rendering of data/bss/constant objects.


- FChE