On 07/04/2016 06:44 PM, Jakub Jelinek wrote:
On Mon, Jul 04, 2016 at 10:23:06AM -0600, Martin Sebor wrote:
No, it doesn't. What I meant is that the same code, when added
in builtins.c instead, could readily be extended to fold into
strings expressions like
sprintf (buf, "%i", 123);
I've commented in some PR a few years ago that I'm not convinced we want to
do it, or at least not without careful considerations, consider .rodata
size. Say if the user has in 1000x different places
sprintf (buf, "foobarbaz %i", NNN); for various values of NNN, then such
"optimization" would replace
a single string literal of length 13 bytes with 1000 string literals of 12-20
bytes.
Consider larger string literal, with %s and long additions and it might not
be a win even for 2 occurrences.
I think that's not a highly ligkely scenario, and it would still be a
massive speed optimization over calling sprintf. Each such call is
likely to be larger than the string literal anyway.
Bernd
