In PR#7172, Daniel Bünzli remarks that even if an user does only "make
world install" only, intentionally not installing the native code
compiler, the configuration variables exposed through
$(ocamlc -where)/Makefile.config will be misleading as they indicate
the configuration-time prediction that this installation has a native
compiler and support native dynlink.
The -no-native-compiler switch allows to avoid this issue by
intentionally indicating at configure-time that no native compiler
shall be built. However, its current implementation forgot to also
explicitly disable native dynlinking. Changing this fixes the
Makefile.config information and has no impact on the compiler build
process (a build with ARCH=none will never hit the {all,install}opt of
dynlink/Makefile that inspect NATDYNLINK).
On regression risk: it is easy to check in ./configure that this
change only affects configuration when the -no-native-compiler option
is passed, which did not exist before (late in) 4.03+dev.
MAX_TESTSUITE_DIR_RETRIES (which defaults to 1 for the native Windows
ports and 0 for all other platforms) causes the `all` and `list` targets
to re-run any given test directory an additional number of times in
order to eliminate random failures during tests.
This is needed because some required constants and prototypes are defined
inside `#ifdef _ALL_SOURCE`
Examples are:
- AI_NUMERICHOST constant in netdb.h (used from getaddrinfo.c)
- chroot prototype in unistd.h (used from chroot.c)
having them. The config script doesn't fail when there's no prototype to the
functions. It should maybe be compiled as "warnings as errors".
(obtained from PKGSRC patches.)
The locgr instruction is not available in z10, the baseline for this port.
Instead, generate pedestrian code with a conditional branch.
Pass -march=z10 to the assembler to enforce z10 compliance.
Following reports that -fwrapv has known issues in GCC prior to version 4.2,
use '-O' (and warn) for GCC [3.0, 4.2);
use '-O2 -fwrapv -fno-strict-aliasing' for GCC 4.2 and up.
git-svn-id: http://caml.inria.fr/svn/ocaml/trunk@16540 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
Auxiliary changes:
- Put GCC in gnu99 mode (= C99 + GNU extensions).
- Check C99 conformance, warn if not.
- Reject if gcc is too old ( < 3.0 )
- Stop C compilation on warnings if this is a development version of OCaml.
(I'm tired of C warnings being ignored.)
git-svn-id: http://caml.inria.fr/svn/ocaml/branches/cc-optim@16329 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
configure: deselect ocamlopt, which is not supported
signals_machdep.h: use i386 instruction sequence, not amd64.
git-svn-id: http://caml.inria.fr/svn/ocaml/trunk@16243 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
- Removed MacOS X support, long dead.
- Added support for PowerPC 64 bits, big-endian, ELF v1 ABI
(tested, mostly works, some issues remain with marshaling of code pointers)
- Added support for PowerPC 64 bits, little-endian, ELF v2 ABI
(completely untested)
git-svn-id: http://caml.inria.fr/svn/ocaml/branches/ppc64@16226 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
(Peter Zotov, review by Mark Shinwell)
NB: after applying this change you need to run ./configure again.
git-svn-id: http://caml.inria.fr/svn/ocaml/trunk@16068 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
Hannes Mehnert
Damien Doligez
Sébastien Hinderer
David Allsopp
Gabriel Scherer
François Bobot
Pierre Chambart
Mark Shinwell
tkob
Rich Neswold
Xavier Leroy
Damien Doligez
Jérémie Dimino
Mark Shinwell