directories. This is compatible with the SunOS behaviour. Also,
it fixes a problem where it was not possible for some 3rd party
package to have a library with the same name as a library already
in the system, and a lower version number. The highest version
would always be picked. This particular problem shows up with Mozilla,
which has a libutil.so.1.0
and referencing externs in non-common code to the appropriate modules
- add -DLANG_SPECIFIC_DRIVER for g++ and g77 as required by a "GNU build"
- Cleanup: $() -> ${}
by any cc1 module and breaks a shared libcc1.
- Add first shot at a shared library of libcc1, only turned on if
CC1_SHLIB is set in make's environment. Doesn't yet know how to install
the library without installing the .a files. This is much faster than
you may think, and shares nearly 2MB worth of code between the cc1
modules.
* ns32k_gen_operand is no longer needed.
* new function symbolic_reference_mentioned_p (stolen from i386).
* make print_operand dtrt when printing symbolic "constants" in picmode
and abort for the cases that it can't handle (e.g. symbol+offset).
ns32k.h:
* Use NS32K_DISPLACEMENT_P wherever suitable.
* make LEGITIMATE_PIC_OPERAND_P dtrt to make the ns32k_gen_operand hack
unnecessary.
ns32k.md:
* Use general_operand instead of ns32k_gen_operand everywhere.
* replace "rmn" by "g" again.
* make addsi3 dtrt for pic.
The comments in ns32k.md still need some updates.
Thank's again to Ian Dall for his help.
This commits several enhancements done over the last years:
- adddi3 subdi3 insn definitions: Jon Buller (ns32k.md, ns32k.c)
- ffssi2 adddi3 and subdi3 bug fixes: Matthias Pfaller (ns32k.c, ns32k.md)
- movstrsi reimplimentation: Ian Dall (ns32k.c, ns32k.md, ns32k.h)
- 32381 fpu support (extra registers, multiply - add instructions):
Ian Dall (ns32k.c, ns32k.h, ns32k.md)
- divmoddisi: (re) implimentation Ian Dall (ns32k.md)
- use predicate rather than register constraints to allow better code for
non pic case: Ian Dall (ns32k.md, ns32k.c)
- clean up of many macros in ns32k.h: Ian Dall
that this is unnecessary (only permission on the file's directory is
necessary), and it has security implications (see PR #5094). There may be
other race-condition-related implications here that this change fixes, as
well.
command line. This behaviour is how it's documented; it should work
that way. Also fixes building a release tree on a -current system (which
bit me with the sparc secr set I built for pk).