On Wed, Jul 16, 2008 at 10:17 PM, Nagy Gabor <ngaba@bibl.u-szeged.hu> wrote:
nicer, but harder ;-) See 84283672853350a84d2a71b72dc06e180cad1587, search for 'type mismatch'.
Hm. Wrong. With the new vercmp: 1.1 > 1.b (but in ASCII 'b' > '1'), strange. Our problem is at the end of string: '\0' vs. alpha, as I see, this case was handled after comment /* see if we ran out of segments on one string */
It indeed looks like we just need to handle the case where it runs out of segments on one string. But we have to handle two cases : run out of segments with the -release number or without it. So in both cases, I handle it differently if the last remaining segment starts with a letter or not. I am not 100% sure that it will work correctly in every single cases, but the logic seems alright, there is no regression on all existing vercmp test, and the 4 new tests you posted in an older thread now pass fine.