I wholeheatedly agree that it would be a really good thing if we were all working around a common, recent, source tree... I'm finding this discussion useful because it's identifying people who are working in the area, and I'm wondering whether we can come together in the short term on building a better regression test suite... > There seems to have been some confusion in the community about which > versions to use, etc. But, for most purposes, i've found that the > latest releases have worked "pretty well" -- certainly well enough to > use for development with a small number of patches... Difference in perception here. Algorithmics have been maintaining GNU C for about ten years for our 'embedded' customers, some of whom have very large and demanding code bases. Up to and including the 2.96+ release we currently work with, no GCC version we've taken on has been fit for use by our MIPS customers without a large number of fixes, including significant changes to nominally non-machine-dependent code. If you experienced a big improvement in quality on moving to some more recent version, I'd love to know and that's worth telling everyone. But if you're saying "it's always been more or less all right" then we are bound to suspect you're not looking hard enough... > Presumably your customers desire both features of new GCC/binutils, > but also want support for your improvements. They want high reliability (which requires fairly lengthy development schedules) and the latest version... I'm sure yours do too, but the requirements compete somewhat. > FWIW, it's not that hard to get changes back in if you try. Thanks for the encouragement. Dominic Sweetman, Algorithmics Ltd
