Date: Sun, 15 Aug 2004 16:27:03 +0100 From: Doug Rabson <dfr@nlsystems.com> To: "Justin T. Gibbs" <gibbs@scsiguy.com> Cc: freebsd-current@freebsd.org Subject: Re: New nvidia drivers available Message-ID: <200408151627.03820.dfr@nlsystems.com> In-Reply-To: <D5005C10BFCCF98A80A9F3D2@aslan.scsiguy.com> References: <Pine.GSO.4.43.0408151001050.15254-100000@sea.ntplx.net> <200408151611.04764.dfr@nlsystems.com> <D5005C10BFCCF98A80A9F3D2@aslan.scsiguy.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sunday 15 August 2004 16:16, Justin T. Gibbs wrote: > >> I thought that static constructor invocation was deterministic > >> based on link order. Does the C++ spec really indicate that the > >> order of construction can be random? > > > > I don't think the spec places any restrictions on constructor > > ordering. The problem here is that you get different behaviour > > depending on whether you link with libGL first followed by > > libpthread (in that case libpthread initialises first) or if you > > link in the other order (in which case libGL initialises first). As > > far as I can see, rtld calls the _init sections of each shared > > library in reverse order with the last library linked against being > > initialised first. > > But such ordering restrictions also apply to things like weak > symbols, so I don't think that imposing a link order restriction to > solve this issue is really a problem. The algorithm for weak symbols is pretty simple - you always get either the first strong symbol that rtld finds in its search or the last weak symbol. I think we arrange for stuff like open(2) to be weak in libc and strong in libpthread. > > >From my Microsoft days, I know that at least PowerPoint took > > advantage > > of the known order of static constructor invocation. The splash > screen was executed from a static constructor in the first .o linked > into the executable. I'm just curious if this is something the C++ > spec says anything about. I'd be surprised if it didn't. I'm sure that constructor ordering differs between binutils toolchains on unix systems and Microsoft toolchains on win32 systems. The software I write in my day job uses C++ constructors heavily and I'm certain that things get constructed in a different order on the two platforms. We just accept it and write code that can initialise in any order.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200408151627.03820.dfr>