[HP aCC 6.15] lots of noise due to compiler remarks

• Key: STDCXX-695
• Type: Bug
• Status: Open
• Priority: Major
• Assignee: Martin Sebor
• Reporter: Martin Sebor
• Component/s: Build and Installation
• Affects Version/s: 4.2.0
• Fix Version/s: 4.2.2

Description

The latest HP compiler on IPF issues lots of remarks, making our builds exceedingly noisy.

Here's a breakdown of all the remarks we get in a recent 12S build along with the number of each:

F=hpux-11.31-ia64-acc-6.15-12S-610576-log; \
R=`sed -n "s/.*: remark \(#[1-9][0-9]*-.\): .*/\1/p" $F | sort | uniq`; \
for r in $R; do \
    n=`grep ": remark $r: " $F | wc -l`; echo $r: $n; \
done

#2177-D: 8   variable was declared but never referenced
#2181-D: 9   argument is incompatible with corresponding format string conversion
#2550-D: 28   variable was set but never used
#2826-D: 44   parameter was never referenced
#2961-D: 1   use of a type with no linkage to declare a variable with linkage
#3235-D: 1   nonstandard conversion between pointer to function and pointer to data
#3348-D: 799   declaration hides constant or variable
#4217-D: 3   comparison of enum (represented as an unsigned integer) with zero
#4219-D: 250   64 bit migration: type conversion may truncate value
#4234-D: 28   type conversion may lose significant bits
#4239-D: 2   case type mismatch with switch expression type
#4244-D: 4   extern storage class used with a function definition
#4248-D: 4   comparison of unsigned integer with a signed integer
#4271-D: 780   type conversion may lose sign
#4273-D: 18   floating-point equality and inequality comparisons may be inappropriate due to roundoff common in floating-point computation
#4278-D: 3   the subexpression in logical expression is a constant
#4282-D: 34   virtual function is called from constructor/destructor
#4283-D: 3   access global variable in constructor
#4296-D: 695   arithmetic operation on boolean type
#4297-D: 1220   boolean value is used as array index
#4298-D: 124   64 bit migration: addition result could be truncated before cast to bigger sized type
#4299-D: 20   64 bit migration: multiply result could be truncated before cast to bigger sized type
#4300-D: 60   Overflow while computing constant in left shift operation
#4315-D: 60   while loop without body, did you insert an extra ';'?

Comments

[Martin Sebor]

Dennis' comments:

-------- Original Message --------
Subject: Re: aCC 6.15 bogus remark #3348
Date: Fri, 11 Jan 2008 18:03:44 -0800 (PST)
From: Dennis Handly <[e-mail]>
To: [e-mail], [e-mail]
CC: [e-mail]

>From: Martin Sebor <[e-mail]>
>We get over 800 of these remarks in each of our builds. I haven't
>analyzed them all but I did reduce one to a test case showing that
>at least in some instances the warning is wrong.

I've seen cases where it was right but we know what we are doing.
And a few broken cases.

struct S { enum { X }; };
template <class T>
int foo() {
const int X = 1;
return X;
}
int main() { return foo<int>(); }
aCC: HP C/aC++ B3910B A.06.14 [Feb 22 2007]
line 4: remark #3348-D: declaration hides constant "S::X" (declared at line 1)
const int X = 1;
detected during instantiation of "int foo<T>() [with T=int]" at line 8

So this is an enum X that is in struct S, not even in scope here.

We released a new compiler A.06.15 in Sept, with a patch A.06.16 in Dec.

>From: "Xiaohua zhang" <[e-mail]>
>This bug has been fixed in ToT, you can try next week's weekly tarball to

Martin doesn't get them.

>From: Martin Sebor <[e-mail]>
>Btw., I just counted the number of remarks we get in stdcxx builds
>with aCC 6.15 and put together a breakdown. You might find the
>results interesting:
http://issues.apache.org/jira/browse/STDCXX-695

Well, interesting in that you haven't figured out that you should just
suppress some of them.

You do know that there is a command "cadvise report logs ..." that
can produce a nice summary??

#2177-D: 8 variable was declared but never referenced
#2550-D: 28 variable was set but never used
#2826-D: 44 parameter was never referenced

Suppress with +W2177,2550,2826.

#2181-D: 9 argument is incompatible with corresponding format string conversion

This is usually your bug.

#3235-D: 1 nonstandard conversion between pointer to function and pointer to data

This isn't legal. You must use a type pun, or bigger hammer.

#3348-D: 799 declaration hides constant or variable

As you said, most are bogus.

#4219-D: 250 64 bit migration: type conversion may truncate value
#4298-D: 124 64 bit migration: addition result could be truncated before cast to bigger sized type
#4299-D: 20 64 bit migration: multiply result could be truncated before cast to bigger sized type

Probably not helpful. We have removed it from +wlint.

#4244-D: 4 extern storage class used with a function definition

A useless coding standard that no right thinking developer would want.
We are thinking of hiding it.

#4296-D: 695 arithmetic operation on boolean type
#4297-D: 1220 boolean value is used as array index
Another two that might not be useful.

#4282-D: 34 virtual function is called from constructor/destructor

A comment in that these aren't virtual.

#4300-D: 60 Overflow while computing constant in left shift operation

Probably a bug.

#4315-D: 60 while loop without body, did you insert an extra ';'?

An attempt to find typos.

[Martin Sebor]

Disabled formatting and bumped up the Original Estimate to 2 hours.

[Martin Sebor]

Most instances of remark #3348 are caused by the compiler bug described in STDCXX-694.

[Martin Sebor]

Remarks #4296-D arithmetic operation on boolean type (695 occurrences) and #4297-D boolean value is used as array index (1220 occurrences) are useless even according to HP. We should silence them.

Out of the total of 780 occurrences of remark #4271-D type conversion may lose sign, only about 120 are distinct. 11 of these are emitted from the library headers so we should silence those first:

$ grep "remark #4271" hpux-11.31-ia64-acc-6.15-15S-620132-log | grep "(TOPDIR)/include" | sort | uniq
"$(TOPDIR)/include/algorithm.cc", line 612: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/algorithm", line 82: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/list", line 407: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/loc/_messages.cc", line 108: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/loc/_money_get.cc", line 323: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/loc/_money_get.cc", line 328: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/sstream", line 201: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/string.cc", line 539: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/string.cc", line 633: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/string.cc", line 646: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/include/string", line 613: remark #4271-D: type conversion may lose sign

And another 11 come from our examples:

$ grep "remark #4271" hpux-11.31-ia64-acc-6.15-15S-620132-log | grep "(TOPDIR)/examples" | sort | uniq
"$(TOPDIR)/examples/manual/priority_queue.cpp", line 61: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/manual/priority_queue.cpp", line 78: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/manual/stack.cpp", line 67: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/tutorial/graph.cpp", line 86: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/tutorial/icecream.cpp", line 220: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/tutorial/radix.cpp", line 56: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/tutorial/sieve.cpp", line 45: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/tutorial/sieve.cpp", line 47: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/tutorial/sieve.cpp", line 51: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/tutorial/teller.cpp", line 60: remark #4271-D: type conversion may lose sign
"$(TOPDIR)/examples/tutorial/teller.cpp", line 64: remark #4271-D: type conversion may lose sign

[Farid Zaripov]

Fix merged in 4.2.x branch thus: http://svn.apache.org/viewcvs?view=rev&rev=637130

[Martin Sebor]

Scheduled for 4.2.1.

[Martin Sebor]

Here's nicer output from the duplicate STDCXX-788:

Latest HP aCC 6.16 builds produce many remarks making them exceedingly noisy:

$ gunzip -c hpux-11.31-ia64-acc-6.16-12D-637671-log.gz | grep "remark #" | wc -l
4187

The breakdown by remark produced by the following script is in the table below:

$ W=`gunzip -c hpux-11.31-ia64-acc-6.16-12D-637671-log.gz | sed -n "s/.* remark #\([1-9][0-9]*\)-D:.*/\1/p" | sort | uniq`; \
  for w in $W; do \
    printf "%s " $w && gunzip -c hpux-11.31-ia64-acc-6.16-12D-637671-log.gz | grep " remark #$w-D:" | wc -l; \
  done

remark	count	description	disabled
2177	10	variable was declared but never referenced
2550	28	variable was set but never used
2826	44	variable was never referenced
2961	1	use of a type with no linkage to declare a variable with linkage
3348	799	declaration hides variable	r639003
4217	3	comparison of enum (represented as an unsigned integer) with zero
4219	251	64 bit migration: type conversion may truncate value
4234	28	type conversion may lose significant bits
4239	2	case type mismatch with switch expression type
4244	4	extern storage class used with a function definition
4248	4	comparison of unsigned integer with a signed integer
4271	712	type conversion may lose sign
4273	18	floating-point equality and inequality comparisons may be inappropriate due to roundoff common in floating-point computation
4278	3	the subexpression in logical expression is a constant
4282	1	virtual function is called from constructor/destructor
4283	3	access global variable in constructor
4296	779	operation on boolean type	r639003
4297	1220	boolean value is used as array index	r639003
4298	124	64 bit migration: addition result could be truncated before cast to bigger sized type
4299	20	64 bit migration: multiply result could be truncated before cast to bigger sized type
4300	72	Overflow while computing constant in left shift operation
4315	61	loop without body, did you insert an extra ';'?

[Martin Sebor]

This is remotely related to the cadvise warnings issue.

[Martin Sebor]

Even though all the subtasks have been resolved we're still not done. We forgot about 32-bit builds... As the dump below shows, we have two noisy remarks to worry about there:

$ for r in `sed -n "s/.*: remark #\([1-9][0-9]*\)-.*/\1/p" hpux-11.31-ia64-acc-6.16-12d-650697.log | sort | uniq`; do \
      n=`grep ": remark #$r-D" hpux-11.31-ia64-acc-6.16-12d-650697.log | wc -l`; \
      printf "remark #$r: %4u\n" $n; \
  done
remark #2177:    8
remark #2181:   22
remark #2550:   28
remark #2826:    6
remark #2961:    1
remark #4217:    3
remark #4219:   32
remark #4230: 2892   # 64 bit migration: conversion may cause target of pointers to have a different size
remark #4234:   29
remark #4239:    2
remark #4244:    4
remark #4248:    4
remark #4271:  692   # type conversion may lose sign
remark #4273:   18
remark #4278:    3
remark #4280:    1
remark #4282:    1
remark #4283:    3
remark #4298:   18
remark #4300:   27
remark #4315:   61

[Martin Sebor]

Deferred until 4.2.2.