Re: -flto to become more of a routine - any change in opinion since 2011?
Steffen Möller writes ("-flto to become more of a routine - any change in opinion since 2011?"):
> I admit to be a fan of link time optimisation and would like to see this
> challenge promoted towards more of a routine challenge to establish for
> our packages. I found this informative thread
>
> https://lists.debian.org/debian-devel/2011/06/msg00181.html
I have a concern not yet addressed in this thread.
Recently we have seen spectacular advances in compiler optimisation.
Spectacular in that large swathes of existing previously-working code
have been discovered, by diligent compilers, to be contrary to the
published C standard, and `optimised' into non-working machine code.
In fact, it turns out that there is practically no existing C code
which is correct according to said standards (including C compilers
themselves). (A full discussion of how this situation came to be is
probably out of scope for debian-devel, and also might involve me
becoming quite rude. So I will avoid that.)
I worry that LTO will exacerbate this problem, by extending the
categories of technical non-compliance (with rules which are very
difficult to fully comply with) which are detected by compilers and
transformed into actual non-working code.
IMO Debian should not arrange for users to be using LTO-affected
executables (in general[1]) until there have been major advances in
the manageability of the C dialect we are using.
To give an idea of what I think would be necessary, here is an
excellent posting from Pascal Cuoq, Matthew Flatt, and John Regehr:
http://blog.regehr.org/archives/1180
(I don't necessarily agree with this in every detail, but it gives a
very good idea of the breadth and depth of the changes I think are
needed.)
In general I highly reccommend Regehr's blog for this kind of topic.
Thanks,
Ian.
[1] Of course if there are specific programs that are somehow known to
be in compliance with the rules being newly enforced in LTO, then it
might be reasonable for those specific packages Debian build systems
to enable LTO.
However it seems like it will be very rarely in practice possible to
establish that a program is correct enough to safely enable LTO.
Reply to: