On Wed, 5 Jun 2002, Lars Gullik Bjønnes wrote: > Allan Rae <[EMAIL PROTECTED]> writes: > > | On Wed, 5 Jun 2002, Andre Poenitz wrote: > > > >> On Tue, Jun 04, 2002 at 06:47:35PM +0200, Lars Gullik Bjønnes wrote: > >> > Not even an feeble attempt at trying to help fix anything, or even look > >> > at alternative solutions. > >> > >> Lars: You lost quite a few of us. There aren't too many people left who > >> are able to help you. > > > | I can't even compile 1.3.0cvs at present because automake-1.4 just > | isn't good enough anymore. > > Hmm... isn't that the case with 1.2.0 as well? No changes has been > made in that respect.
No, the 1.2.x branch is happy with automake-1.4. There have been plenty of changes in configure.in and such in 1.3.0cvs. > >> And a working alternative solution had been there, at least in a few cases. > >> But that's the "revert mantra" in your ears. > > > | I really would have thought that after all the talk about branches at > | various times in the last year that experimental changes like > | switching to boost::signals would have been done in a branch so if it > | didn't work it could be abandoned but kept for reference without much > | hassle or merged into the "stable" trunk if deemed successful. > > Not exactly experimental... There was discussion. A couple of signal/slot mechanisms were suggested as potential replacements. Boost got put straight into the trunk without evaluation or a chance for comparison to another implementation (apart from the previous libsigc++ -- although you seem to be refusing to accept criticism of boost when it compares unfavourably with libsigc++ -- code readability, template bloat etc.) > | Despite that, it seems to be more and more necessary for LyX > | developers _and users_ to be using gcc-cvs rather than > | gcc-some-recent-release if they want to compile LyX! > > Bullshit! > > Please try to backup that claim. You said yourself in an earlier email how 3.1 is the compiler we should be using. And how 3.2 will have some wonderful new template code that reduces overhead by a factor of 5. Likewise, you are also claiming that the only compilers worth using with exceptions is gcc > 3.0. You are similarly calling 2.96 a non-compiler or at least a compiler everyone should avoid. That pretty much leaves compilers that only been available this year. Upgrading to these compilers involves upgrading every other C++ library and app also present on the system. So a complete system upgrade is probably the simplest option. This requires either a purchase of a boxed Linux set or a lot of cheap bandwidth (at about $0.02 per megabyte in Australia an iso image is very expensive). For people with OSes other than Linux and particularly for users in institutions this is getting beyond a joke. > | This might be > | reasonable if LyX were as politically big a project as the Linux > | kernel and hence could drive gcc development or have some other > | significant contribution/influence on other projects but we aren't. > > > | boost is starting to be a PITA much as XTL was before its time and > | beyond readily available compilers. > > But now the compilers are available... Maybe you have cheap unlimited bandwidth but many users and developers don't. Readily available compilers is 2.95.x or at best 2.96 not 3.x. Mandrake 8.2 has been out a while and while it carries 3.0.4 everything was actually compiled with 2.96. And when the two are installed together 2.96 is the default compiler. I'm sure there are other distros doing similar (dumb?) things. On another note: Most developers have agreed there is a need to revise the various class hierarchies (particularly the GUII hierarchy). We can do that while still being compilable with 2.95.3. But we can't if we keep riding the bleeding edge of C++. Allan. (ARRae)
