[devkitARM] C++ bigger in r26 than r25 - sprintf usage?

[Quirky]

It seems like cp-demangle.c is using sprintf rather than siprintf and this adds a fair bit to the size of the final .nds, presumably due to all the floating point stuff. I have no idea why cp-demangle is linked in to my code (some missing C++ flag?), so that could be the real cause of my particular problem, but in r25 it was linked in too but didn't pull in sprintf.

[Quirky]

Logically enough, it seems adding -g is doing this. Removing -g reduces the final .nds file by 30-40k. I guess I had gotten used to the fact that -g on the NDS didn't have much overhead.

[WinterMute]

Interesting, I just did some quick tests with C code and the -g option doesn't seem to make any difference to final binary size. I'll have to do a bit of digging and see what's going on with this in C++.