Jim Nance wrote: > In article <aelj3f$[EMAIL PROTECTED]>, Dan Mosedale wrote: > >>Even if all components survive an ABI change, both Mozilla and some of >>the plugins link directly against libstdc++. So right now, I've got a >>gcc 3.1 build on my disk which can't use the flash plugin, because >>Mozilla is linked against the gcc 3.1 version of libstdc++, and the >>plugin is expecting the name mangling of the old version. >> >>That said, I think we should take a list of plugin developers and try >>to get rough consensus that we should bite the bullet and upgrade >>compilers, in large part because of the optimization wins we get can. > > > Hi Dan, > I really think we should hold off on this. It is going to create a > lot of confusion to have incompatable sets of plugins floating around the > net and thats going to generate a lot of support problems for the plugin > authors. And the last thing we want is for mozilla to get a "difficult to > support" reputation.
Well, yeah, it sucks. But we have to do it sooner or later, and it'll suck whenever we do it. And since there's a significant performance win involved, might as well be sooner. > The last I heard (and admittedly its been several months) gcc3 generated > slower code than gcc 2.9X. Thats expected to change, but it will take > time and I think we should wait until there is some clear advantage to > using gcc 3 before we push people in that direction. Even if this were true, which I'm not sure I believe, gcc 3.1 -O2 is significantly faster than egcs 1.1.2 -O, which is what we're using now. And it's got an ABI that is believed to be stable. > On a slightly related topic, will netscape 4.X plugins continue to work > on mozilla if we switch to gcc 3? I think several of the plugins I commonly > use are really 4.X plugins (flash & realplayer). I suspect most people are > in the same boat. Breaking these would be really bad. That's why I suggested getting consensus among the plugin developers first; we would indeed need them to rebuild their plugins in order to pull this off. Dan
