Tom Kistner reported that my constness updates from the "Compiler
masochism compliance" patch:
http://git.exim.org/exim.git/commit/1ba28e2b955b005ce4825fec792df17f75a8de1e
broke build on Linux; in his words:
> I was fiddling with the latest git state today. The "const"ification in
> src/osfunctions.h breaks build on my Linux box, because OS/os.h-Linux sets
> "#define os_strsignal strsignal". Then the prototype in
> /usr/include/string.h and aliased prototype in src/osfunctions.h have
> conflicting types
On the whole, I'm still happy with the change and don't want to revert
it, since it clears the way for more stringent compiler warnings which
should result in less buggy code. Heck, it led to my spotting the
ldap_require_cert bug.
I think that the fix is sufficiently simple to warrant patching rather
than reversion.
My proposed fix is:
http://git.exim.org/users/pdp/exim.git/commit/b52ce0604579f50b71e7527e5030dfb3f74c618b
----------------------------8< cut here >8------------------------------
Rely on system prototypes if we #define our os funcs.
The const-ness updates broke systems where `os_strsignal()` gets mapped
to `strsignal()`, which does *not* return `const char *` but `char *`.
If we #define away, then there should be a prototype from the system
headers.
----------------------------8< cut here >8------------------------------
Does this generally look sane to people? I temporarily modified
os.h-FreeBSD to use the same strsignal() redefine and without the commit
above, things broke but with the patch above, things compile.
I'm about to shut down my laptop to pack it for a plane flight, so I'm
not going to push changes to the master repo when I'm not going to be
around to deal with fall-out. If someone else is happy with this
change, feel free to push it to the master repo. :)
Sorry for the breakage, and thanks for the report, Tom.
-Phil
