Compaq C V6.5 on VMS 7.3-2 compiler crash

[John Carrette]

We've recently installed Compaq C V6.5 on VMS V7.3-2. We have an intermittent problem compiling C files, some modules always crash the compiler, some will compile about 50% of the time so we don't think it's the programs, also if we introduce a syntax error then the compiler reports the error every time, i.e. doesn't crash, it's only "clean" code where the error occurs. We have made all the checks we can think of, any help much appreciated. The error is as follows:

$ cc testprog
Assertion failure: Compiler internal error - please submit problem report
%GEM-F-ASSERTION, Compiler internal error - please submit problem report
%TRACE-F-TRACEBACK, symbolic stack dump follows
image module routine line rel PC abs PC
DECC$COMPILER GEM_DB GEM_DB_ABORT_FAST
736 00000000000000CC 00000000006ED6CC
DECC$COMPILER GEM_CD_CALL_ALPHA_VMS GEM_CD_BIND_ADDRESS_MODE
1539 0000000000000F44 00000000008DD694
DECC$COMPILER GEM_CD_ALPHA GEM_CD_LOAD_ADDRESS
2043 000000000000155C 00000000008E2F2C
DECC$COMPILER GEM_CD CD_INTERP 7750 0000000000007524 00000000008EBE04
DECC$COMPILER GEM_CD GEM_CD 1575 000000000000013C 00000000008E4A1C
DECC$COMPILER GEM_CG GEM_CG 820 0000000000000414 00000000008EFB84
DECC$COMPILER GEM_CO GEM_CO_COMPILE_ROUTINE
5898 0000000000008548 00000000006DAB18
DECC$COMPILER GEM_CO GEM_CO_COMPILE_MODULE
3600 0000000000000D30 00000000006D3300
DECC$COMPILER COMPILE gemc_be_master
84816 0000000000000E88 00000000001BE898
DECC$COMPILER COMPILE gem_xx_compile
84039 0000000000000000 0000000000000000
DECC$COMPILER GEM_CP_VMS GEM_CP_MAIN
2603 00000000000018CC 00000000006BA1AC
0 FFFFFFFF8025FE94 FFFFFFFF8025FE94

[Kris Clippeleyr]

John,
Have you tried compiling /NOOPTIMIZE ?
Sometimes the C compiler gets confused. I also have a few sources that make it crash; /NOOPTIMIZE did the trick.
Regards,
Kris (aka Qkcl)

[Ian Miller.]

V6.5 is ot the most recent version so when you submit a problem report hp may ask you to upgrade.

Check process quotas - increasing them may help to workaround the bug.

[John Carrette]

Thanks for the reply but forgit to mention that quite a few command line switches, including /noopt, have been tried. Some of our modules will actually compile sometimes, at other times they won't!

[John Carrette]

Thanks for the reply. We have increased many quotas but to no avail. The intermittent nature of this does point to a problem like that but I'm afraid I can't find it, I wondered if anyone has a similar problem ?

[Antoniov.]

John,
do compile with vax mode? Can you try something like standard (/ACCEPT & /STANDARD qualifiers)?
May it be something like /ARCHITECT?
If you use some fortran library look at /PD qualifier!
Good Luck!

Antonio Vigliotti

[John Carrette]

Antonio

Thanks. We do need to use /standard=vax but it makes no difference to the compiler crashing. Have also tried many different combinations of /accept and /arch. The problem is intermittent, we have modules it will compile about 50% of the attempts we try with exactly the same switches or no switches, which makes us believe it may be a compiler or possibly a VMS set-up problem.

John Carrette

[Craig A Berry]

Let's just double check you don't have a field test version of the compiler by doing:

$ cc/version

If there are any ECOs available for 6.5-001 I haven't heard of them and couldn't find any after a few minutes searching. 6.5A is a documentation only release and 7.1 is still only on Itanium.

It's worth making sure you are up-to-date on system ECOs; I have heard of compiler problems caused by problems in the infrastructure the compiler depends on.

I would also recommend getting a free account on eisner.encompasserve.org or testdrive.hp.com, upload some code, and see if you can get the same error.

And finally, try compiling with /WARNINGS=LEVEL6 to see if there is something about your code it doesn't like but dies before it can tell you. This will be very noisy if you are accustomed to using /STANDARD=VAXC.

[John Carrette]

Craig, thanks for the ideas.

cc/version shows
Compaq C V6.5-001 on OpenVMS Alpha V7.3-2

cc /warnings=(informationals=level6, warnings=level6, errors=level6, fatals=level6)
does not give any other clues on the module that always crashes,
but it did coax the following from the module that sometimes compiles before it went
into the now familiar crash, but this does not seem a likely cause to crash the compiler, especially as it will compile this module every now and again.

LOG_OUTPUT( SFLOG_ORA, CONCAT(SFLOG_MESS25, CSHUTDOWN.arr) );
..^
%CC-W-PTRMISMATCH1, In this statement, the referenced type of the pointer value "(CSHUTDOWN.arr)" is "unsigned char", which is not
compatible with "const char" because they differ by signed/unsigned attribute.
at line number 1143 in file $DISK2:[JD_CARRETTE]SFAPCORA.C;1
Assertion failure: Compiler internal error - please submit problem report
%GEM-F-ASSERTION, Compiler internal error - please submit problem report
%TRACE-F-TRACEBACK, symbolic stack dump follows
image module routine line rel PC abs PC
DECC$COMPILER GEM_DB GEM_DB_ABORT_FAST
736 00000000000000CC 00000000006ED6CC
DECC$COMPILER GEM_CD_CALL_ALPHA_VMS GEM_CD_BIND_ADDRESS_MODE
1539 0000000000000F44 00000000008DD694

I will check ECOs for VMS, that's a good idea, OpenVMS V7.3-2 is fairly new to us.

Thanks

John Carrette

[John Gillings]

John,

A compiler assertion failure is always a bug.

If you're on the latest compiler the best thing to do is what the message says: "please submit problem report".

Log a case with your local customer support centre, with a copy of one of your source files (plus any required #includes).

[Antoniov.]

John,
I remember an old vax c bug about unsigned constant char in parameter list; may be this bug is again appeared .
On old vax, I remember, I removed the constant declaration every parameter list, leaving them only in alpha platform.
Do you have some declaration like this?

Antonio Vigliotti

[John Carrette]

Antonio, thanks, no "unsigned const char", the problem is intermittent anyway, some modules will compile sometimes.
John, thanks, I have been trying to log a call with HP, think I'm getting soemwhere now. Will update this when I get a resolution.