When to emit diagnistics

[jacob navia]

Recently we had a discussion about the following code:

void f(long *lp) { *lp = 0; }
int main(void) { int i; f(&i); return i; }

lcc-win doesn't emit any diagnostic in normal mode, and doesn't
emit a diagnostic with the higher warning level.

It emits a diagnostic only in the highest warning level:

lcc -A -A foo.c

will diagnostic that with a warning.

Why?

Because under windows (32 and 64 bits) there is
absolutely no PRACTICAL difference between a
long and an int. They are completely equivalent types.

The diagnostic would just add CLUTTER and NOISE to
the output of the compiler.

The problem is that if the compiler emits too many diagnostics
the important ones will go unnoticed, swallowed by the noise
of the unimportant ones.

Note that Microsoft C, for instance will not diagnose this
even with the highest warning level.

This are the reasons behind the decision. Some people like
heathfield or becarisse use this as a "proof that lcc-win is
useless", etc. Their usual stuff. Note that lcc-win is not
just for pleasing pedants in comp.lang.c but it is useful
for doing REAL work. The objective is not to please
pedants here.


--
jacob navia
jacob at jacob point remcomp point fr
logiciels/informatique

http://www.cs.virginia.edu/~lcc-win32

[Joachim Schmitz]

Re: When to emit diagnistics

jacob navia wrote:

Recently we had a discussion about the following code:
>
void f(long *lp) { *lp = 0; }
int main(void) { int i; f(&i); return i; }
>
lcc-win doesn't emit any diagnostic in normal mode, and doesn't
emit a diagnostic with the higher warning level.
>
It emits a diagnostic only in the highest warning level:
>
lcc -A -A foo.c
>
will diagnostic that with a warning.

Good. One problem solved. Now document that switch in your manual.

Why?
>
Because under windows (32 and 64 bits) there is
absolutely no PRACTICAL difference between a
long and an int. They are completely equivalent types.
>
The diagnostic would just add CLUTTER and NOISE to
the output of the compiler.
>
The problem is that if the compiler emits too many diagnostics
the important ones will go unnoticed, swallowed by the noise
of the unimportant ones.

All valid points, indeed.

Note that Microsoft C, for instance will not diagnose this
even with the highest warning level.

Haven't tested this myself, but a far as I understood others (namely RH), it
it does diagnose this if called _in conforming mode_. So please don't
continue comparing apples with peaches.

This are the reasons behind the decision. Some people like
heathfield or becarisse use this as a "proof that lcc-win is
useless", etc. Their usual stuff. Note that lcc-win is not
just for pleasing pedants in comp.lang.c but it is useful
for doing REAL work. The objective is not to please
pedants here.

Nobody claimed win-lcc to be useless. The only claim here (and plenty proof
for it) was that win-lcc does not (unlike your claim) conform to either
C89/C90 or C99.
So it is a compiler for a C-like language, no more, no less. That doesn't
make it useless.

Bye, Jojo

[santosh]

Re: When to emit diagnistics

jacob navia wrote:

Recently we had a discussion about the following code:
>
void f(long *lp) { *lp = 0; }
int main(void) { int i; f(&i); return i; }
>
lcc-win doesn't emit any diagnostic in normal mode, and doesn't
emit a diagnostic with the higher warning level.
>
It emits a diagnostic only in the highest warning level:
>
lcc -A -A foo.c

The problem is you have not (as far as I can tell) *documented* this
switch combination. Undocumented features are IMO worse than
unimplemented features.

will diagnostic that with a warning.
>
Why?
>
Because under windows (32 and 64 bits) there is
absolutely no PRACTICAL difference between a
long and an int. They are completely equivalent types.

You are assuming here that any Standard C program compiled with lcc-win
under Windows has no chance of later being ported to a different
system.

This warning is absolutely essential for a user who considers
portability to be important.

The diagnostic would just add CLUTTER and NOISE to
the output of the compiler.

Then why not add a specific switch for enabling warnings of this class,
like say gcc? Or you could arrange for the '-A' switch to take a
numeric value which would specify the warning level. Necessitating
*two* instances of the '-A' switch is very counter-intuitive.

Don't be afraid to add more command line switches. A C compiler is a
serious tool, and a professional programmer is expected to study it
thoroughly as a part of his job.

The problem is that if the compiler emits too many diagnostics
the important ones will go unnoticed, swallowed by the noise
of the unimportant ones.

That's what different diagnostic levels are for. If a programmer
explicitly asks for *all* possible diagnostics, then he is presumably
prepared to deal with some innocuous ones.

Note that Microsoft C, for instance will not diagnose this
even with the highest warning level.
>
This are the reasons behind the decision. Some people like
heathfield or becarisse use this as a "proof that lcc-win is
useless", etc. Their usual stuff. Note that lcc-win is not
just for pleasing pedants in comp.lang.c but it is useful
for doing REAL work. The objective is not to please
pedants here.

No one said lcc-win was "useless" or just for "pleasing pedants". This
thread developed because you claimed that lcc-win conforms to both C90
and C99 which is clearly not the case.

Regardless of conformance, one of the real advantages of HLLs is the
rigorous automated checking that is possible. People are prone to
errors and typos. A tool that catches them (even at the risk of some
false positives) is superior to one that has reduced ability in this
regard.

IME Intel's compiler has somewhat better diagnostics than even gcc. You
would do well to study it's behaviour and strive to emulate it. IMO of
course.

[Richard Heathfield]

Re: When to emit diagnistics

jacob navia said:

Recently we had a discussion about the following code:
>
void f(long *lp) { *lp = 0; }
int main(void) { int i; f(&i); return i; }
>
lcc-win doesn't emit any diagnostic in normal mode, and doesn't
emit a diagnostic with the higher warning level.
>
It emits a diagnostic only in the highest warning level:
>
lcc -A -A foo.c
>
will diagnostic that with a warning.

A diagnostic message is required for constraint violations and syntax
errors. If your implementation doesn't issue one, it doesn't conform to
the language spec.

Why?
>
Because under windows (32 and 64 bits) there is
absolutely no PRACTICAL difference between a
long and an int. They are completely equivalent types.
>
The diagnostic would just add CLUTTER and NOISE to
the output of the compiler.
>
The problem is that if the compiler emits too many diagnostics
the important ones will go unnoticed, swallowed by the noise
of the unimportant ones.

Syntax errors and constraint violations are pretty important.

>
Note that Microsoft C, for instance will not diagnose this
even with the highest warning level.

Liar. When you first claimed this, it was reasonable to assume you were
mistaken. But after you claimed it three times, I checked it and told you
"MSVC diagnoses this error properly when invoked in conforming mode", and
you replied "and lcc-win also", in an article timestamped 12 minutes or so
before your "When to emit diagnistics" article. So you knew and accepted
that Microsoft C does indeed diagnose this error, which makes you a liar
for pretending that it doesn't.

This are the reasons behind the decision. Some people like
heathfield or becarisse use this as a "proof that lcc-win is
useless", etc.

Liar. I have never claimed that lcc-win is useless. I *have* claimed that
it isn't a C compiler, and you seem to agree, since you don't appear to
claim either C90 or C99 conformance for it.

Their usual stuff. Note that lcc-win is not
just for pleasing pedants in comp.lang.c but it is useful
for doing REAL work. The objective is not to please
pedants here.

If you want to get real work done, use a compiler that implements the
language it claims to implement.

--
Richard Heathfield <http://www.cpax.org.uk >
Email: -http://www. +rjh@
Google users: <http://www.cpax.org.uk/prg/writings/googly.php>
"Usenet is a strange place" - dmr 29 July 1999

[Ian Collins]

Re: When to emit diagnistics

jacob navia wrote:

Recently we had a discussion about the following code:
>
void f(long *lp) { *lp = 0; }
int main(void) { int i; f(&i); return i; }
>
lcc-win doesn't emit any diagnostic in normal mode, and doesn't
emit a diagnostic with the higher warning level.
>
It emits a diagnostic only in the highest warning level:
>
lcc -A -A foo.c
>
will diagnostic that with a warning.
>
Why?
>
Because under windows (32 and 64 bits) there is
absolutely no PRACTICAL difference between a
long and an int. They are completely equivalent types.
>

But they are incompatible types. I thought you'd made a Linux port of
the compiler? There you would have an issue.

Both Sun cc and gcc warn on their default settings.

The diagnostic would just add CLUTTER and NOISE to
the output of the compiler.
>
The problem is that if the compiler emits too many diagnostics
the important ones will go unnoticed, swallowed by the noise
of the unimportant ones.
>

Most shops have a "no warnings in the build" rule for that very reason.
It isn't hard to do.

--
Ian Collins.

[Richard Heathfield]

Re: When to emit diagnistics

Ian Collins said:

<snip>

Most shops have a "no warnings in the build" rule for that very reason.
It isn't hard to do.

Not if you only have to cater for one compiler, no. There's usually a way
of shutting most compilers up, for any given spurious warning.
Unfortunately, the way to shut one compiler up might well trigger a
warning under a different compiler. Compiling the same source base cleanly
under multiple implementations is not a trivial exercise.

--
Richard Heathfield <http://www.cpax.org.uk >
Email: -http://www. +rjh@
Google users: <http://www.cpax.org.uk/prg/writings/googly.php>
"Usenet is a strange place" - dmr 29 July 1999

[Ian Collins]

Re: When to emit diagnistics

Richard Heathfield wrote:

Ian Collins said:
>
<snip>
>

>Most shops have a "no warnings in the build" rule for that very reason.
> It isn't hard to do.

>
Not if you only have to cater for one compiler, no. There's usually a way
of shutting most compilers up, for any given spurious warning.
Unfortunately, the way to shut one compiler up might well trigger a
warning under a different compiler. Compiling the same source base cleanly
under multiple implementations is not a trivial exercise.
>

In that case use minimum warning levels and lint.

--
Ian Collins.

[santosh]

Re: When to emit diagnistics

Keith Thompson wrote:

jacob navia <jacob@nospam.c omwrites:

>Recently we had a discussion about the following code:
>>
>void f(long *lp) { *lp = 0; }
> int main(void) { int i; f(&i); return i; }

>
Which clearly violates a constraint.
>

>lcc-win doesn't emit any diagnostic in normal mode, and doesn't
>emit a diagnostic with the higher warning level.

>
Out of curiosity, does lcc-win emit a diagnostic for this?
>
int main(void)
{
long obj = 42;
int *ptr = &obj;
return 0;
}
>
What about the equivalent using an assignment rather than an
initialization?

It does (at least my copy of it). In both cases the diagnostic is:

Warning test.c: 5 assignment of pointer to long int to pointer to int

In addition:

Warning test.c: 4 ptr is assigned a value that is never used

is also emitted.

<snip>

[Spiro Trikaliotis]

Re: When to emit diagnistics

Hello,

jacob navia wrote:

Recently we had a discussion about the following code:
>
void f(long *lp) { *lp = 0; }
int main(void) { int i; f(&i); return i; }

[...]

Note that Microsoft C, for instance will not diagnose this
even with the highest warning level.

This is not true. I just tried it with Visual C++ 6.0, all service packs
applied:

default settings: (warning level 3): No warning
incremented warning level to 4: No warning

warning level back to 3, but "disable language extensions" active:

C:\test\a.c(9) : warning C4133: 'function' : incompatible types - from 'int *' to 'long *'

This is exactly what I have expected.

I do not expect a compiler must emit the warning if it does not claim to
be conforming ("language extensions") - although, I would love it.
However, I expect it to emit these warnings if in compliant mode - and
MSVC6 does.

Note that this is MSVC6, which is ten years old.

Regards,
Spiro.

"Anybody who would spend considerable effort emulating a C64 is not
particularly sane to begin with; it is unreasonable to expect them to produce
sane software."
("asuffield" in http://forums.thedailywtf.com/forums/p/7199/134136.aspx)

--
Spiro R. Trikaliotis http://opencbm.sf.net/
http://www.trikaliotis.net/ http://www.viceteam.org/

[Joachim Schmitz]

Re: When to emit diagnistics

Spiro Trikaliotis wrote:

Hello,
>
jacob navia wrote:
>

>Recently we had a discussion about the following code:
>>
>void f(long *lp) { *lp = 0; }
> int main(void) { int i; f(&i); return i; }

>
[...]
>

>Note that Microsoft C, for instance will not diagnose this
>even with the highest warning level.

>
This is not true. I just tried it with Visual C++ 6.0, all service
packs
applied:
>
default settings: (warning level 3): No warning
incremented warning level to 4: No warning
>
warning level back to 3, but "disable language extensions" active:
>
C:\test\a.c(9) : warning C4133: 'function' : incompatible types -
from 'int *' to 'long *'
>
This is exactly what I have expected.

And the considerably younger VC 8.0, AKA Visual Studio 2005, is giving
exatly the same warning, down to warning level 1, as long as language
extension are disabled (i.e. working in conforming mode).

Bye, Jojo

[jacob navia]

Re: When to emit diagnistics

Joachim Schmitz wrote:

Spiro Trikaliotis wrote:

>Hello,
>>
>jacob navia wrote:
>>

>>Recently we had a discussion about the following code:
>>>
>>void f(long *lp) { *lp = 0; }
>> int main(void) { int i; f(&i); return i; }

>[...]
>>

>>Note that Microsoft C, for instance will not diagnose this
>>even with the highest warning level.

>This is not true. I just tried it with Visual C++ 6.0, all service
>packs
>applied:
>>
>default settings: (warning level 3): No warning
>incremented warning level to 4: No warning
>>
>warning level back to 3, but "disable language extensions" active:
>>
>C:\test\a.c( 9) : warning C4133: 'function' : incompatible types -
>from 'int *' to 'long *'
>>
>This is exactly what I have expected.

And the considerably younger VC 8.0, AKA Visual Studio 2005, is giving
exatly the same warning, down to warning level 1, as long as language
extension are disabled (i.e. working in conforming mode).
>
Bye, Jojo
>
>

I did not know that you have to disable language extensions
to get a warning.

Anyway, that obscure option matches lcc -A -A obscure option.


--
jacob navia
jacob at jacob point remcomp point fr
logiciels/informatique

http://www.cs.virginia.edu/~lcc-win32

[Eric Sosman]

Re: When to emit diagnistics

jacob navia wrote:

Recently we had a discussion about the following code:
>
void f(long *lp) { *lp = 0; }
int main(void) { int i; f(&i); return i; }
>
lcc-win doesn't emit any diagnostic in normal mode, and doesn't
emit a diagnostic with the higher warning level.
>
It emits a diagnostic only in the highest warning level:
>
lcc -A -A foo.c
>
will diagnostic that with a warning.

Since the diagnostic is required, it follows that using the
highest warning level is a prerequisite for operating the compiler
in a conforming mode.

Why?
>
Because under windows (32 and 64 bits) there is
absolutely no PRACTICAL difference between a
long and an int. They are completely equivalent types.

They are distinct types that happen (on this implementation)
to share the same underlying representation.

This is a difficult notion for assembly programmers to grasp:
They tend to think about "type" in the way the hardware does, so
the idea of distinct types with identical behavior seems to them
obfuscatory. Angels on pinheads, academic piffle, and so on.

But the distinction is real, and also useful. Let me pose
a rhetorical question: If int and long are really the same, why
do programmers write `int' at some places and `long' at others?
Clearly the programmers who make the choices think there are --
or at least can be -- differences between int and long, and by
writing one or the other they express an intention about how they
want their variables to behave. To put the rhetorical question
even more crudely: If int and long are identical, why not just
eliminate `long' and flag it as an error?

Here's another, somewhat more abstract reason to pay attention
to type distinctions. To begin, we all know that `char' has the
same representation and behavior as `signed char' on some systems,
and as `unsigned char' on others. Yet even though `char' is "just
like" one or the other of these types, it is a type distinct from
both of them. By using plain `char', the programmer expresses a
preference that the machine use whichever type is "easier" for it
to deal with; the programmer does not need to commit to signed or
to unsigned and find that his code is unnecessarily slow on half
the machines that run it. That's why it is valuable to be able to
write `char' as distinct from `signed char' and `unsigned char',
even though it always behaves like one or the other. A distinction
that makes no difference on any one machine turns out to make a
difference when considering the wider universe of target machines.

Similar concerns motivate other types, too, which is why the
sizeof operator produces a size_t, why time() returns a time_t,
why sig_atomic_t exists, and so on. On any implementation you
can point at, these abstracted types (and others) are very likely
to be synonyms for various flavors of integers -- but for different
flavors on different systems. That's why portability is improved
by paying attention to type distinctions: They permit the programmer
to express his intentions to many platforms at once rather than to
just one at a time.

This are the reasons behind the decision. Some people like
heathfield or becarisse use this as a "proof that lcc-win is
useless", etc.

I don't recall seeing anyone describe lcc-win as "useless."
"Non-conforming," yes, but that's a different matter.

Note that lcc-win is not
just for pleasing pedants in comp.lang.c but it is useful
for doing REAL work.

It seems to me that "REAL" workers pay due attention to
portability, in light of the objectives of the project at hand.
A compiler that points out accidental strayings is helpful to
the programmers doing the REAL work, hence useful in that work.

--
Eric Sosman
esosman@ieee-dot-org.invalid

[Spiro Trikaliotis]

Re: When to emit diagnistics

jacob navia schrieb:

Joachim Schmitz wrote:

>Spiro Trikaliotis wrote:

[...]

>>warning level back to 3, but "disable language extensions" active:

[...]

>>This is exactly what I have expected.

>And the considerably younger VC 8.0, AKA Visual Studio 2005, is giving
>exatly the same warning, down to warning level 1, as long as language
>extension are disabled (i.e. working in conforming mode).

[...]

I did not know that you have to disable language extensions
to get a warning.
>
Anyway, that obscure option matches lcc -A -A obscure option.

For me, "disable language extensions" matches something like "--ansi"
(or similar).

Regards,
Spiro.

"Anybody who would spend considerable effort emulating a C64 is not
particularly sane to begin with; it is unreasonable to expect them to produce
sane software."
("asuffield" in http://forums.thedailywtf.com/forums/p/7199/134136.aspx)

--
Spiro R. Trikaliotis http://opencbm.sf.net/
http://www.trikaliotis.net/ http://www.viceteam.org/

[jacob navia]

Re: When to emit diagnistics

Eric Sosman wrote:

This is a difficult notion for assembly programmers to grasp:
They tend to think about "type" in the way the hardware does, so
the idea of distinct types with identical behavior seems to them
obfuscatory. Angels on pinheads, academic piffle, and so on.

Pedants love that.

"char is different from signed char and unsigned char. It is another
type".

"int and long are different types even if they are the same in some
implementations "

And so on.

Yes, how CLEVER, how WELL you have mastered the C language, incredible.

I am just an "assembly language programmer". Obviously for you they
are kind of thick heads that have "difficulti es" grasping elementary
concepts like how many bugs can you fit in a pinhead.

Contrary to the pedants here that despise assembly and assembly
language programmers I write assembly every day. Even worst

o I generate assembly language from C input
o I assemble it
o I link it

Can you imagine that?

I *must* be a thick head then, obviously.


--
jacob navia
jacob at jacob point remcomp point fr
logiciels/informatique

http://www.cs.virginia.edu/~lcc-win32

[Eric Sosman]

Re: When to emit diagnistics

jacob navia wrote:

Eric Sosman wrote:

> This is a difficult notion for assembly programmers to grasp:
>They tend to think about "type" in the way the hardware does, so
>the idea of distinct types with identical behavior seems to them
>obfuscatory. Angels on pinheads, academic piffle, and so on.

>
>
Pedants love that.
>
"char is different from signed char and unsigned char. It is another
type".
>
"int and long are different types even if they are the same in some
implementations "
>
And so on.
>
Yes, how CLEVER, how WELL you have mastered the C language, incredible.
>
I am just an "assembly language programmer". Obviously for you they
are kind of thick heads that have "difficulti es" grasping elementary
concepts like how many bugs can you fit in a pinhead.

I've never said you have a thick head, not in this nor in
any other thread. Nor did I use the word "just" as in "mere"
or "only" to describe assembly language programmers; in fact,
I wrote "just" only three times altogether, none of them having
to do with programmers of any stripe.

What I wrote is that assembly language programmers have a
hard time understanding the importance of type, and I speculated
about the reason. That's my opinion, and by dismissing type as
pedantry you offer supporting evidence.

Contrary to the pedants here that despise assembly and assembly
language programmers

If you intend to include me among the despisers, you are
flat-out wrong. R-O-N-G, wrong.

I write assembly every day. Even worst
>
o I generate assembly language from C input
o I assemble it
o I link it
>
Can you imagine that?
>
I *must* be a thick head then, obviously.

Not a thick head, but an extraordinarily thin skin.

--
Eric.Sosman@sun .com