> Kazu Hirata writes: > >> Nathan writes: >> this is untrue. the elements hold the qualification. > > Then how do I know that an array is declared with const (or static > const)? When I looked at the CONSTRUCTOR stored in the DECL_INITIAL > of an array, I saw it with the TREE_STATIC flag set regardless of > whether the array is declared with const, so that's not useful to > determine whether the array is declared with const. > > All I need is a sufficient condition that works with C, C++, and any > other language that GCC supports to determine whether it's safe to > fold array[CST] into a constant.
My basic understanding based on GCC's documentation, is that only memory references to "static const" values (and presumably non-in-lined literals) are warranted to have MEM_READONLY_P (mem-ref) true (which I understand is designated by the unchanging attribute at the tree/rtx level); as these are the only references which may physically be read-only and traditionally warranted not to change (as as memory references to "const" only objects are more a semantic restriction, not necessarily a literal one, as non-"const" objects may be cast as being "const" and visa versa, therefore can't be reliably presumed to be READONLY/unchanging). However do note that something does appear to be broken with respect to literal string constants, as MEM_READONLY_P memory references to them are not reliably true as they should be, although char array literals do seem to be reliably designated (at least for the C FE).
