Jan> I am reading this remote so I cannot check all the details. Ok, I fill you in. Jan> Andreas Jaeger <aj(a)suse.de&gt; writes: Jan> Is there a documented guarantee this will never be needed? I'll happy Jan> to take your word for glibc related things, but isn't it is possible Jan> that for instance the gcc folk could decide to set it in libgcc.a? Or Jan> should I take your comment to mean that glic's malloc doesn't use the Jan> hook at all? If we compile with DOUG_LEA_MALLOC, XEmacs nowhere sets or uses __malloc_hook - besides in this one place. __malloc_hook is essential for GNU malloc - but with Doug Lea's malloc, we don't need it. glibc needs to initialize its malloc implementation and uses __malloc_hook for this initilisation. If you set __malloc_hook to NULL, the initialisation code is not executed at all - and the malloc implementation will fail. Therefore, we've got two solutions for XEmacs for now: - call malloc before setting __malloc_hook so that it gets initialised (my first patch) - don't set __malloc_hook at all. Since __malloc_hook is part of glibc and not part of gcc, we only have to deal with glibc here. Jan> Is the version in Debian potato an official released version. If not Jan> would it may be better just to make this patch when there is actually Jan> a version with this problem released? The version 2.1.2-6 from Debian is based on the current CVS version of glibc 2.1.3 - nothing official. BUT: All distributions[1] use at least some patches from the CVS archive. I don't think there's a distribution that uses plain 2.1.2 without patches. I guess that users of the current Debian potato, the latest RedHat release (6.1?) and of the coming SuSE release (6.3) who try to compile their own XEmacs, get these problems. Jan> Jan Jan> P.S. I am interpreting your patch correctly as saying glibc doesn't Jan> actually define malloc_hook until after the first malloc. Yes - and this is a bug IMO. glibc uses __malloc_hook internally - and also externally. After malloc is initialized, __malloc_hook can be used by user programs - but before malloc is initialized you've got to be careful. This is broken. But there's no trivial patch for glibc. Andreas Footnotes: [1] Ok, I know only SuSE, RedHat and Debian in this regard. -- Andreas Jaeger SuSE Labs aj(a)suse.de private aj(a)arthur.rhein-neckar.de