1 Jan 2002 01:02
Re: [Patch] (Partial) fix for libstdc++/14078
Paolo Carlini <pcarlini <at> suse.de>
2002-01-01 00:02:00 GMT
2002-01-01 00:02:00 GMT
Pétur Runólfsson wrote: > Paolo Carlini wrote: > >> Better results, working around deficiencies of the current inliner >> (http://gcc.gnu.org/ml/libstdc++/2003-02/msg00302.html) could be >> obtained by adding const qualifiers to the signatures. > > But a top level const qualifier on a formal parameter is *not* part > of the signature, see 13.1. The only effect it has is to make the > parameter const within the function body. In other words, it's > equivalent to declaring a local variable const. Pétur, you don't need to explain that again. I know that technically we could add those 'const' here and there in order to help the defects of the current inliner. Still, all of this will become very soon redundant, since the tree-ssa (aka 3.5) already has those problems fixed and it would be the first time that we do that kind of fiddling. As you may have noticed we do our best not to add hacks and workarounds in the library, which is meant as a long term project, humbly a "reference implementation", not something which must perform better at any cost than a non-existant competitor (since in formatted output of integers, on x86 at least, this is the current situation ;) Paolo.(Continue reading)
however...
The problem is potentially serious (note, however, that nobody, to date, has
actually reported anything ;) but I have already a sketch of the fix and I
plan to have something ready by the end of the week.
Obviously, thank you very much for your kind offer of help: for sure, I'll
ask you to test thoroughly the patch on your machines (as often happens for
invasive changes)... ;)
Paolo.
RSS Feed