On 01/28/2012 02:59 PM, Bruno Haible wrote: > Up to now, we have always minimized the code size, > even if it costs some .m4 macros to run at configure time.
That's usually fine, and it has the nice properties that you list, but it's not the only objective -- there are also the goals of increasing reliability of 'configure', reducing the number of dependencies, and making 'configure' go faster. I listed these roughly in decreasing order of importance, but in this particular case I thought that the fix would reduce dependencies and increase 'configure' speed without causing harm in more-important areas. So far our luck has held, but even if it fails on some rarely-used host I hope the change is still an overall win and that we'll need to revisit this only if our luck fails on some relatively-important target (and obviously there are some judgment calls here).
