On Mon, Mar 30, 2015 at 11:41:52AM +0200, Jean-Marc Lasgouttes wrote: > Le 26/03/2015 17:00, Enrico Forestieri a écrit : > >On Thu, Mar 26, 2015 at 12:09:41PM +0100, Jean-Marc Lasgouttes wrote: > >> > >>On a related idea, could you tell us what is the intent of the > >>following code? > >> > >> int iconsize = Length(1, Length::EM).inPixels(1); > >> percent_scale = (100 * iconsize + imgsize / 2)/imgsize; > > > >Sorry, I thought it was self-explanatory. The first line sets iconsize > >to be equal to the number of pixels corresponding to 1em (that should > >be the width in pixels of the letter 'M' as shown on screen). The second > >line computes how much the actual image should be scaled to match the > >wanted size (imgsize is the size in pixels of the image). It would > >simply be 100*(iconsize/imgsize), but in that way I round up or down > >the result to the integer nearest to the float value of the result. > >I mean, 100*10/15 = 66.666, int(100*10/15) = 66, int((100*10+7)/15) = 67. > > I understand now. I think this deserves a comment,
Ok, I will add one. > especially since this looks like you forgot parentheses like "(iconsize + > imgsize) /2". No, no. It's equivalent to rounding as int(a/b + 0.5), with a float-type division. Indeed, int(a/b + 0.5) = (a + b/2)/b, where we have an integer-type division on the RHS. -- Enrico
