https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110284
Bug ID: 110284
Summary: [14 Regression] Bootstrap failures with m2
Product: gcc
Version: 14.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: modula2
Assignee: gaius at gcc dot gnu.org
Reporter: jakub at gcc dot gnu.org
Target Milestone: ---
As mentioned in
https://gcc.gnu.org/pipermail/gcc-patches/2023-June/620819.html
and later in
https://gcc.gnu.org/pipermail/gcc-patches/2023-June/621576.html
since the inclusion of insn-opinit.h (generated header) in internal-fn.h
parallel bootstraps with modula2 enabled randomly fail with errors like:
In file included from ../../gcc/m2/gm2-gcc/gcc-consolidation.h:74,
from ../../gcc/m2/gm2-lang.cc:24:
../../gcc/internal-fn.h:24:10: fatal error: insn-opinit.h: No such file or
directory
24 | #include "insn-opinit.h"
| ^~~~~~~~~~~~~~~
compilation terminated.
(not always, just sometimes with bad luck).
My r14-1603-g00bfc503cc3b3e8f3 commit for this was actually incorrect and
should be reverted - the problem is that gcc/m2/Make-lang.in is included before
generated_files
variable is set in gcc/Makefile.in and the order only rule
$(GM2_C_OBJS) : | $(generated_files)
is then expanded at that point as
$(GM2_C_OBJS) : |
and so doesn't actually do anything.
As I said, the right fix is mentioned in the second mail,
-$(GM2_C_OBJS) : | $(generated_files)
+m2_OBJS = $(GM2_C_OBJS)
should work correctly, because then the
$(ALL_HOST_OBJS) : | $(generated_files)
rule in gcc/Makefile.in will include the m2 FE files and will DTRT.
The problem is that this needs further work on the m2 FE, which I can't easily
do because I don't know much about the FE. Inclusion in m2_OBJS means they
will be
compiled with -DIN_GCC_FRONTEND and e.g. checking in system.h tries to make
sure
that rtl.h, except.h etc. headers aren't included in the FEs, but m2 FE does
include them (not sure why, most of them probably uselessly). There is a
further problem that
m2 compiles its own variant of stor-layout.cc which is clearly a backend file.
If this can't be made to work correctly, I guess as last resort perhaps
-$(GM2_C_OBJS) : | $(generated_files)
+ALL_HOST_OBJS += $(GM2_C_OBJS)
could be used, which would be an ugly hack, but I think putting it in m2_OBJS
is the right thing.
Gaius, could you please have a look?
Thanks.
