On Tuesday 04 June 2002 11:00 am, Jean-Marc Lasgouttes wrote: > >>>>> "Angus" == Angus Leeming <[EMAIL PROTECTED]> writes: > > Angus> But there are NO signals in biblio.[Ch]; the difference here is > Angus> that biblio.C now makes use of boost::regex. > > I know. It is nevertheless an object file whose dec size got > multiplied by more than 5 by just using a template-happy boost lib. I > do not think that signals are the only ones to blame for the +50% bin > size increase. > > Angus> Could you perform the same test on a file containing a couple > Angus> of signals? > > Let's look at FormPreference.o: > > 1.2.x (reference): > lapinot: size build/src/frontends/xforms/FormPreferences.o > text data bss dec hex filename > 96298 592 128 97018 17afa > build/src/frontends/xforms/FormPreferences.o > > > 1.3.0cvs: > lapinot: size devbuild/src/frontends/xforms/FormPreferences.o > text data bss dec hex filename > 246079 20256 128 266463 410df > devbuild/src/frontends/xforms/FormPreferences.o > > same compiled with -finline-limit=100 > lapinot: size FormPreferences.o > text data bss dec hex filename > 114783 35536 128 150447 24baf FormPreferences.o > > And for fun, 1.2.x compiled with -finline-limit=100 > lapinot: size FormPreferences.o > text data bss dec hex filename > 83626 488 128 84242 14912 FormPreferences.o > > > JMarc
Ouch indeed. On a practical note, are there any downsides to using this -finline-limit flag with non-3-series gcc compilers? If not, shouldn't it be put into configure? Angus
