pcc blows-up with major internal compiler error when confronted with AST mamake...

[Roland Mainz]

Hi1

----

Trying to compile the attached source file (attached as "mamake.c.gz")
blows-up pcc (build from today's (=2012-08-14, 22:40h CEST) cvs) like
this:
-- snip --
major internal compiler error:
/home/test001/work/ast_ksh_20120813/build_pcc/src/cmd/INIT/mamake.c,
line 372
/home/test001/pccinstall/lib/ccom terminated with status 1
-- snip --
... it seems to be related to the |register| keyword... the error goes
away if I remove them or add a |#define register| at the top.

Any help/workaround/magic would be greatly appreciated since this is
currently a blocker issue before we can start testing "pcc" ...

----

Bye,
Roland

[Roland Mainz]

Post by Roland Mainz
Trying to compile the attached source file (attached as "mamake.c.gz")
blows-up pcc (build from today's (=2012-08-14, 22:40h CEST) cvs) like
-- snip --
/home/test001/work/ast_ksh_20120813/build_pcc/src/cmd/INIT/mamake.c,
line 372
/home/test001/pccinstall/lib/ccom terminated with status 1
-- snip --
... it seems to be related to the |register| keyword... the error goes
away if I remove them or add a |#define register| at the top.

PCC-400 is probably the same as this

Erm... if you mean the bug "PCC-399" ('ast-ksh/mamake build fails with
"major internal compiler error" in src/cmd/INIT/mamake.c, line 372')
... then "yes". We're trying to evaluate pcc and crawl from issue to
issue... slowly...;-/

It is always difficult to reproduce an issue which happens in an
environment or with a lot of source code.. I always try to reduce it to a
standalone segment, since a person who might fix it may not have the same
environment.

It happens consistently for on SuSE 12.1+Redhat with 32bit and 64bit
targets... LC_ALL=C... and trying to cut-down the sources didn't help
much. I tried to find a "lowest common denominator" (because this
issue occurs for many of the AST sources) ... but without success yet.

Post by Roland Mainz
Any help/workaround/magic would be greatly appreciated since this is
currently a blocker issue before we can start testing "pcc" ...

as you suggest, CPPFLAGS=-Dregister might do it..

CPPFLAGS=-Dregister= ... if I just use -Dregister the CPP value will
AFAIK default to "1" and that causes build breakage...
... then we hit YAI(=yet another issue) where "pcc" puts output which
is (AFAIK) intended for stderr into stdout... but I'll file a seperate
bug for that mess once the Jira bugtracker is online again...

----

Bye,
Roland

[Iain Hibbert]

Post by Roland Mainz
I tried to find a "lowest common denominator" (because this
issue occurs for many of the AST sources) ... but without success yet.

I think this lowest common denominator might be PCC-400 which is

void foo(void)
{
register int a = 0;
}

I did not have any idea about your PCC-399 since I have neither that
environment, nor that source, nor the wget tool required to follow your
instructions to fetch it..

iain

[Iain Hibbert]

Post by Roland Mainz
Hi1
----
Trying to compile the attached source file (attached as "mamake.c.gz")
blows-up pcc (build from today's (=2012-08-14, 22:40h CEST) cvs) like
-- snip --
/home/test001/work/ast_ksh_20120813/build_pcc/src/cmd/INIT/mamake.c,
line 372
/home/test001/pccinstall/lib/ccom terminated with status 1
-- snip --
... it seems to be related to the |register| keyword... the error goes
away if I remove them or add a |#define register| at the top.

PCC-400 is probably the same as this

It is always difficult to reproduce an issue which happens in an
environment or with a lot of source code.. I always try to reduce it to a
standalone segment, since a person who might fix it may not have the same
environment.

Post by Roland Mainz
Any help/workaround/magic would be greatly appreciated since this is
currently a blocker issue before we can start testing "pcc" ...

as you suggest, CPPFLAGS=-Dregister might do it..

iain